Midway through the first set of Monday’s Region 3D Tournament opener at Bassett High School, Bassett senior Zoie Pace made a pass over her head to fellow senior Annie Laine. Laine then made the kill to give the Bengals a slight lead over rival Magna Vista.

High School Volleyball: Bassett's Zoie Pace records 1,500th career assist

The 11 Best Whiskeys We Tasted In October 2023, Ranked 35 Creamy Soup Recipes False claim 'climate crisis hoax' statement signed by thousands of scientists | Fact check US announces deadly new nuclear weapon days after China announced warhead expansion ...

Red and Rover by Brian Basset

Generative AI products like ChatGPT and Google's ( GOOG) Bard took the world by storm late last year. Yahoo Finance UK visited this year's Chatbot Summit in London on 12 October to discover what the next success story to come out of the AI sector could look like.

The next generative AI success story could come from the call centre market

Several of the women on The Real Housewives of Potomac have had to deal with cheating allegations that involved their husbands. Despite being in the middle of her own fallout from her marital strife,

‘RHOP’: Candiace Dillard Bassett Slams Robyn Dixon and Gizelle Bryant For Believing Cheating Rumors About Her Husband

Tumi Mhlongo Fred Jagueneau/Bravo Tumi Mhlongo is at the center of Below Deck Mediterranean season 8 — but that doesn’t mean she doesn’t have questions as she watches the show back. “It’s so interesting watching the show back,

Below Deck Med's Tumi Mhlongo Has Questions While Watching Season 8 Back

Is something keeping you from pursuing your dreams? This basset hound is here to remind you that you're capable of more than you think!

Adorable: Brave Basset Hound Overcomes Daunting Obstacle!

A basset hound that was swept away by a rip current at Eerste Rivier Beach in the Tsitsikamma area of the Western Cape was rescued by a Good Samaritan on Friday. National Sea Rescue Institute (NSRI) Oyster Bay operation controller Jaen Smit said volunteer duty crew responded to reports of a man and his dog being washed off the rocks.

Basset hound caught in rip current rescued by former SA Navy diver in Tsitsikamma

Asda fans could see a brand new store open in their local area in the coming weeks as part of plans to expand the retailer's portfolio.

Asda to open more than 350 stores across the UK - full list of confirmed locations

Allies of Colombian President Gustavo Petro have been defeated by wide margins in municipal and provincial elections, in what analysts say is a sign of growing discontent with the country's first left

Colombia veers to the right as President Petro’s allies lose by wide margins in regional elections

Basset

basset

Managing digital assets in large scale cloud ecosystems is a pressing challenge. Organizations accumulate millions of assets, databases, virtual machines, logs, models each needing proper ownership tracking for security, compliance, and cost optimization. Traditional methods like manual tagging, role-based access control (RBAC), and static access logs are inefficient, outdated, and prone to security breaches. To solve this, I developed an AI based Actor-Asset Relationship Modeling System, #US Patent 10511606 and 10388040, that dynamically discovers asset ownership using interaction based probability models. In this article, we’ll dive into: The flaws of traditional asset management
Graph based modeling of ownership relations
Machine learning algorithms for dynamic ownership scoring
Real world implementation using Python, NetworkX, and FAISS
Deployment strategies at scale with distributed databases and cloud APIs Why Static Asset Ownership Tracking Fails Cloud assets can be highly dynamic and volatile often changing hands between users, services and automated processes. Consider these real world challenges Security Risks: Orphaned assets (e.g., unused S3 buckets, VM instances) become security vulnerabilities.


Compliance Failures: Auditing data lineage is complex when ownership isn’t properly tracked.


Cost Inefficiencies: Resources continue accruing costs, even after their owners leave the company. A static RBAC approach cannot adapt to this fluid ownership model requiring a proactive approach. The Solution? Graph-Based Actor-Asset Ownership Modeling Instead of manually assigning asset ownership we use an actor-asset graph, where: Nodes represent users, services or service accounts.
Edges represent interactions (e.g., API calls, file access, logins, modifications).
Edge weights are time based decayed scores indicating interaction strength. Let’s Implement it? Graph Construction with NetworkX import networkx as nx
import numpy as np

#Initialize directed graph
G = nx.DiGraph()

#Sample interactions (User, Asset, Timestamp)
interactions = [
    ("User_A", "Asset_1", 100),
    ("User_B", "Asset_1", 200),
    ("User_A", "Asset_1", 300),
    ("User_C", "Asset_1", 400),
    ("User_B", "Asset_1", 500),
]

#Add edges (user -> asset) with timestamps
for user, asset, timestamp in interactions:
    G.add_edge(user, asset, weight=timestamp)

print("Graph Nodes:", G.nodes)
print("Graph Edges:", G.edges(data=True)) Decaying Historical Interactions with Exponential Time Decay Older interactions should contribute less weight to ownership scoring. We apply exponential decay, this ensures recent activity dominates the ownership score. #Apply decay function
def apply_decay(timestamp, current_time, decay_rate=0.01):
    return np.exp(-decay_rate * (current_time - timestamp))

#Compute ownership scores
current_time = 600  #Assume current time
ownership_scores = {}

for user, asset, data in G.edges(data=True):
    score = apply_decay(data['weight'], current_time)
    ownership_scores[user] = ownership_scores.get(user, 0) + score

#Rank users by highest ownership score
sorted_owners = sorted(ownership_scores.items(), key=lambda x: x[1], reverse=True)

print("Ownership Scores:", sorted_owners) Ownership Prediction Once a graph is built, we apply Graph Neural Networks (GNNs) to learn asset ownership patterns. import torch
import torch.nn.functional as F
from torch_geometric.nn import GCNConv
from torch_geometric.data import Data

#Convert graph to PyTorch Geometric format
edge_index = torch.tensor(list(G.edges)).t().contiguous()
x = torch.tensor(np.random.rand(len(G.nodes), 128), dtype=torch.float)  # 128D embeddings

#Define Graph Convolutional Network (GCN)
class OwnershipGCN(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = GCNConv(128, 64)
        self.conv2 = GCNConv(64, 1)  # Output ownership score
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

#Initialize and train
model = OwnershipGCN()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(50):
    optimizer.zero_grad()
    out = model(x, edge_index)
    loss = F.mse_loss(out, x.mean(dim=0))  # Self-supervised learning
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}") Optimizing for Scale: FAISS and Vector Search for Fast Ownership Retrieval For real time ownership lookup, embeddings can be stored in FAISS (Facebook AI Similarity Search) for efficient retrieval. import faiss

#Convert learned GNN embeddings to NumPy
learned_embeddings = model(x, edge_index).detach().numpy()

#Create FAISS index
d = 128  #Embedding dimensionality
index = faiss.IndexFlatL2(d)
index.add(learned_embeddings)

#Query for top-k most similar asset owners
query_vector = learned_embeddings[0].reshape(1, -1)  #Query with first user
distances, indices = index.search(query_vector, k=5)

print("Top-5 most likely asset owners:", indices) Cloud-Scale Deployment: Integrating AI-Driven Ownership Tracking into Cloud Services
A production-ready system integrates with cloud IAM systems:

AWS IAM - Enforce dynamic access policies
Azure Active Directory - Automate resource reassignment
Google Cloud IAM - Secure data lineage tracking For Example, if an asset is abandoned, an ownership score below threshold triggers automatic reassignment or deletion alerts. threshold = 0.1  #Define ownership confidence threshold
for user, score in sorted_owners:
    if score < threshold:
        print(f"🚨 Alert: {user} has low ownership confidence!") Real World Applications of AI Powered Asset Ownership Cybersecurity & Insider Threat Detection -  Detect orphaned accounts with access to critical data.
AI Model Governance - Ensure proper lineage tracking for AI datasets.
Regulatory Compliance (GDPR, HIPAA, SOC2) - Provide automated audit trails.
Cloud Cost Optimization - Identify and delete unused resources. Why Dynamic Ownership Tracking is the Future Traditional static access policies cannot keep up with the scale and dynamism of modern cloud ecosystems. This model solves this by: Improving security - Prevents abandoned asset vulnerabilities.
Enhancing compliance - Ensures traceable asset ownership.
Optimizing costs - Reduces waste from unused cloud resources. As cloud infrastructure grows, smart governance will be essential for scalable, secure and cost effective cloud management. The future of cloud security isn’t just about access control - it’s about understanding ownership at scale! Managing digital assets in large scale cloud ecosystems is a pressing challenge. Organizations accumulate millions of assets, databases, virtual machines, logs, models each needing proper ownership tracking for security, compliance, and cost optimization. Traditional methods like manual tagging, role-based access control (RBAC), and static access logs are inefficient, outdated, and prone to security breaches. To solve this, I developed an AI based Actor-Asset Relationship Modeling System, #US Patent 10511606 and 10388040 , that dynamically discovers asset ownership using interaction based probability models. 10511606 10388040 In this article, we’ll dive into: The flaws of traditional asset management Graph based modeling of ownership relations Machine learning algorithms for dynamic ownership scoring Real world implementation using Python, NetworkX, and FAISS Deployment strategies at scale with distributed databases and cloud APIs The flaws of traditional asset management Graph based modeling of ownership relations Machine learning algorithms for dynamic ownership scoring Real world implementation using Python, NetworkX, and FAISS Deployment strategies at scale with distributed databases and cloud APIs Why Static Asset Ownership Tracking Fails Cloud assets can be highly dynamic and volatile often changing hands between users, services and automated processes. Consider these real world challenges Security Risks: Orphaned assets (e.g., unused S3 buckets, VM instances) become security vulnerabilities. Compliance Failures: Auditing data lineage is complex when ownership isn’t properly tracked. Cost Inefficiencies: Resources continue accruing costs, even after their owners leave the company. Security Risks: Orphaned assets (e.g., unused S3 buckets, VM instances) become security vulnerabilities. Security Risks: Orphaned assets (e.g., unused S3 buckets, VM instances) become security vulnerabilities. Compliance Failures: Auditing data lineage is complex when ownership isn’t properly tracked. Compliance Failures: Auditing data lineage is complex when ownership isn’t properly tracked. Cost Inefficiencies: Resources continue accruing costs, even after their owners leave the company. Cost Inefficiencies: Resources continue accruing costs, even after their owners leave the company. A static RBAC approach cannot adapt to this fluid ownership model requiring a proactive approach. The Solution? Graph-Based Actor-Asset Ownership Modeling Instead of manually assigning asset ownership we use an actor-asset graph, where: Nodes represent users, services or service accounts. Edges represent interactions (e.g., API calls, file access, logins, modifications). Edge weights are time based decayed scores indicating interaction strength. Nodes represent users, services or service accounts. Edges represent interactions (e.g., API calls, file access, logins, modifications). Edge weights are time based decayed scores indicating interaction strength. Let’s Implement it? Graph Construction with NetworkX Graph Construction with NetworkX Graph Construction with NetworkX import networkx as nx
import numpy as np

#Initialize directed graph
G = nx.DiGraph()

#Sample interactions (User, Asset, Timestamp)
interactions = [
    ("User_A", "Asset_1", 100),
    ("User_B", "Asset_1", 200),
    ("User_A", "Asset_1", 300),
    ("User_C", "Asset_1", 400),
    ("User_B", "Asset_1", 500),
]

#Add edges (user -> asset) with timestamps
for user, asset, timestamp in interactions:
    G.add_edge(user, asset, weight=timestamp)

print("Graph Nodes:", G.nodes)
print("Graph Edges:", G.edges(data=True)) import networkx as nx
import numpy as np

#Initialize directed graph
G = nx.DiGraph()

#Sample interactions (User, Asset, Timestamp)
interactions = [
    ("User_A", "Asset_1", 100),
    ("User_B", "Asset_1", 200),
    ("User_A", "Asset_1", 300),
    ("User_C", "Asset_1", 400),
    ("User_B", "Asset_1", 500),
]

#Add edges (user -> asset) with timestamps
for user, asset, timestamp in interactions:
    G.add_edge(user, asset, weight=timestamp)

print("Graph Nodes:", G.nodes)
print("Graph Edges:", G.edges(data=True)) Decaying Historical Interactions with Exponential Time Decay Decaying Historical Interactions with Exponential Time Decay Decaying Historical Interactions with Exponential Time Decay Older interactions should contribute less weight to ownership scoring. We apply exponential decay, this ensures recent activity dominates the ownership score. #Apply decay function
def apply_decay(timestamp, current_time, decay_rate=0.01):
    return np.exp(-decay_rate * (current_time - timestamp))

#Compute ownership scores
current_time = 600  #Assume current time
ownership_scores = {}

for user, asset, data in G.edges(data=True):
    score = apply_decay(data['weight'], current_time)
    ownership_scores[user] = ownership_scores.get(user, 0) + score

#Rank users by highest ownership score
sorted_owners = sorted(ownership_scores.items(), key=lambda x: x[1], reverse=True)

print("Ownership Scores:", sorted_owners) #Apply decay function
def apply_decay(timestamp, current_time, decay_rate=0.01):
    return np.exp(-decay_rate * (current_time - timestamp))

#Compute ownership scores
current_time = 600  #Assume current time
ownership_scores = {}

for user, asset, data in G.edges(data=True):
    score = apply_decay(data['weight'], current_time)
    ownership_scores[user] = ownership_scores.get(user, 0) + score

#Rank users by highest ownership score
sorted_owners = sorted(ownership_scores.items(), key=lambda x: x[1], reverse=True)

print("Ownership Scores:", sorted_owners) Ownership Prediction Ownership Prediction Once a graph is built, we apply Graph Neural Networks (GNNs) to learn asset ownership patterns. import torch
import torch.nn.functional as F
from torch_geometric.nn import GCNConv
from torch_geometric.data import Data

#Convert graph to PyTorch Geometric format
edge_index = torch.tensor(list(G.edges)).t().contiguous()
x = torch.tensor(np.random.rand(len(G.nodes), 128), dtype=torch.float)  # 128D embeddings

#Define Graph Convolutional Network (GCN)
class OwnershipGCN(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = GCNConv(128, 64)
        self.conv2 = GCNConv(64, 1)  # Output ownership score
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

#Initialize and train
model = OwnershipGCN()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(50):
    optimizer.zero_grad()
    out = model(x, edge_index)
    loss = F.mse_loss(out, x.mean(dim=0))  # Self-supervised learning
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}") import torch
import torch.nn.functional as F
from torch_geometric.nn import GCNConv
from torch_geometric.data import Data

#Convert graph to PyTorch Geometric format
edge_index = torch.tensor(list(G.edges)).t().contiguous()
x = torch.tensor(np.random.rand(len(G.nodes), 128), dtype=torch.float)  # 128D embeddings

#Define Graph Convolutional Network (GCN)
class OwnershipGCN(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = GCNConv(128, 64)
        self.conv2 = GCNConv(64, 1)  # Output ownership score
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

#Initialize and train
model = OwnershipGCN()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

for epoch in range(50):
    optimizer.zero_grad()
    out = model(x, edge_index)
    loss = F.mse_loss(out, x.mean(dim=0))  # Self-supervised learning
    loss.backward()
    optimizer.step()
    print(f"Epoch {epoch+1}, Loss: {loss.item()}") Optimizing for Scale: FAISS and Vector Search for Fast Ownership Retrieval Optimizing for Scale: FAISS and Vector Search for Fast Ownership Retrieval For real time ownership lookup, embeddings can be stored in FAISS (Facebook AI Similarity Search) for efficient retrieval. import faiss

#Convert learned GNN embeddings to NumPy
learned_embeddings = model(x, edge_index).detach().numpy()

#Create FAISS index
d = 128  #Embedding dimensionality
index = faiss.IndexFlatL2(d)
index.add(learned_embeddings)

#Query for top-k most similar asset owners
query_vector = learned_embeddings[0].reshape(1, -1)  #Query with first user
distances, indices = index.search(query_vector, k=5)

print("Top-5 most likely asset owners:", indices) import faiss

#Convert learned GNN embeddings to NumPy
learned_embeddings = model(x, edge_index).detach().numpy()

#Create FAISS index
d = 128  #Embedding dimensionality
index = faiss.IndexFlatL2(d)
index.add(learned_embeddings)

#Query for top-k most similar asset owners
query_vector = learned_embeddings[0].reshape(1, -1)  #Query with first user
distances, indices = index.search(query_vector, k=5)

print("Top-5 most likely asset owners:", indices) Cloud-Scale Deployment: Integrating AI-Driven Ownership Tracking into Cloud Services
A production-ready system integrates with cloud IAM systems:

AWS IAM - Enforce dynamic access policies
Azure Active Directory - Automate resource reassignment
Google Cloud IAM - Secure data lineage tracking Cloud-Scale Deployment: Integrating AI-Driven Ownership Tracking into Cloud Services A production-ready system integrates with cloud IAM systems: AWS IAM - Enforce dynamic access policies
Azure Active Directory - Automate resource reassignment
Google Cloud IAM - Secure data lineage tracking Cloud-Scale Deployment: Integrating AI-Driven Ownership Tracking into Cloud Services A production-ready system integrates with cloud IAM systems: AWS IAM - Enforce dynamic access policies Azure Active Directory - Automate resource reassignment Google Cloud IAM - Secure data lineage tracking AWS IAM - Enforce dynamic access policies Azure Active Directory - Automate resource reassignment Google Cloud IAM - Secure data lineage tracking For Example, if an asset is abandoned, an ownership score below threshold triggers automatic reassignment or deletion alerts. threshold = 0.1  #Define ownership confidence threshold
for user, score in sorted_owners:
    if score < threshold:
        print(f"🚨 Alert: {user} has low ownership confidence!") threshold = 0.1  #Define ownership confidence threshold
for user, score in sorted_owners:
    if score < threshold:
        print(f"🚨 Alert: {user} has low ownership confidence!") Real World Applications of AI Powered Asset Ownership Cybersecurity & Insider Threat Detection -  Detect orphaned accounts with access to critical data. AI Model Governance - Ensure proper lineage tracking for AI datasets. Regulatory Compliance (GDPR, HIPAA, SOC2) - Provide automated audit trails. Cloud Cost Optimization - Identify and delete unused resources. Cybersecurity & Insider Threat Detection -  Detect orphaned accounts with access to critical data. AI Model Governance - Ensure proper lineage tracking for AI datasets. Regulatory Compliance (GDPR, HIPAA, SOC2) - Provide automated audit trails. Cloud Cost Optimization - Identify and delete unused resources. Why Dynamic Ownership Tracking is the Future Traditional static access policies cannot keep up with the scale and dynamism of modern cloud ecosystems. This model solves this by: Improving security - Prevents abandoned asset vulnerabilities. Enhancing compliance - Ensures traceable asset ownership. Optimizing costs - Reduces waste from unused cloud resources. Improving security - Prevents abandoned asset vulnerabilities. Enhancing compliance - Ensures traceable asset ownership. Optimizing costs - Reduces waste from unused cloud resources. As cloud infrastructure grows, smart governance will be essential for scalable, secure and cost effective cloud management. The future of cloud security isn’t just about access control - it’s about understanding ownership at scale! The future of cloud security isn’t just about access control - it’s about understanding ownership at scale!

Your Cloud Is Leaking Money and Data—This AI Fixes It Before Hackers Do

The Cosmos (ATOM) SWOT Analysis—Q4, 2023 a SWOT analysis is an evaluation of the fundamental, operational, technical, social, economic, and even to some degree administrative elements of a project. This is not a model to be used for trading purposes. (NFA, DYOR) Note: Composed of four elements, Strengths, Weaknesses, Opportunities, and Threats, a SWOT analysis framework provides excellent insight for establishing a high-level understanding of the state of a project's well-being through the lens of a birds-eye view. It can help formulate decisions around which areas require more attention, set performance goals, and organize a foundational understanding of where a project is headed. Rarely (if ever) used in crypto, it is time to apply this timeless method of evaluation to the digital asset space. Today, the interoperability OG and father of the IBC, will get a SWOT. Cosmos (ATOM), 💪 Strengths (Internal) (Helpful) 1. AEZ The Atomic Economic Zone is at the heart of Cosmos's upcoming techno-economic overhaul. The purpose of the Economic Zone is to tighten the alignment of the ecosystem with the ATOM token. A multitude of networks, including Kujira, Agoric, Osmosis, Stride, Neutron, et al., have already integrated and pledged their diligence. At its most basic, the AEZ's main focus is to have ATOM play a foundational role in the inter-chain activities and unifiy chains with universally applicable governance frameworks. 2. World Class Governance Communityx Either legendary community-building skills, deviously clever teams, or a combination of both, Cosmos has achieved incredible levels of engagement both through its social and governance platforms. The core community is always active and involved in the proceedings, helping steer the project and uphold its interesting commitment to decentralization. 3. Adoption by Chains Comos has seen an influx of projects into its ecosystem. Hosting over 46 networks so far, of which many are considered to be noteworthy (Osmosis, KUJI, Cronos, ThorChain, Injective, Secret, KAVA, Akash, etc.) and a pipeline of new leading projects, namely Celestia and dYdX, the positioning of Cosmos within the context of the greater industry could be tipping into the "Too-Big-Too-Fail" territory. 4. Contribution to Web3 Technology From its consensus mechanism (CometBFT) to the inter-chain communications protocol (IBC) to its appchain thesis and beyond, so many solutions have tapped into the tech pioneered by Cosmos and made available through their SDK. Value in the form of new technological primitives can be found laden throughout their excellent documentation. If a project could truly be valued on its merits, Cosmos would dwarf 99% of other base layers. 5. Shift of Security Model Cosmos has been undergoing tectonic transformations in its architecture. Previously acting as a relay point, the Cosmos Hub will now act as a point of unified security provision. Effectively, the shift signifies an emphasis on shared security (whereby multiple different networks leverage the same validator set). While there has been some valid rebuttal against using economic primitives (price) for security due to the implied relationship between price and security, the conversation has primarily been based in theory and cannot be proven until something happens. The new model just might be enough to help Cosmos stand out from the increasing pool of infrastructure providers in Web3. 6. Stake Rate over 67% Stake rate is a direct expression of an ecosystem's community's willingness to support a project. With over 2/3's of all ATOM staked, Cosmos happens to have one of the highest rates among other networks of the same caliber. Taking into consideration that ATOM was among the worst performers (purely from a price perspective) this last year, the fact that long-term stakeholders did not abandon their positions (in fact, actually grew them from 62% same time last year) shows deep commitment. The higher the stake rate, the more constrained the circulating supply and the sharper the reflexivity between demand and price. If this trend continues, it would be unlikely that ATOM does not reflect it. 😞 Weaknesses (Internal) (Harmful) While a SWOT analysis does not usually account for price action, it is impossible to ignore the fact that ATOM is one of the worst performers among tier-1 crypto projects. Regardless of the parabolic growth taking place across independent Cosmos chains and the market at large, it seems that market participants are simply not interested in ATOM. The expression of interest is similar to liquidity, a self-feeding mechanism where interest begets more interest. 1. Economic Underperformance 2. Tokenomics A contentious point that has long been discussed around Cosmos is the design of its core system; while the technology has been amazing and certainly seen adoption, the ATOM token's economics have been lackluster in their ability to capture value. With its relatively high double-digit perpetual inflation, absence as a requirement for Cosmos chain's fees, and inability to address these issues earlier, has resulted in a prolonged period of sub-optimal performance attributable to the inefficient initial implementation forces the project to ensure invisible costs in the form of early community members. The problem can be (and is currently being) fixed but it does not guarantee that the new model will not be flawed in its own right. 3. IBC's Lack of Asset Interoperability A lesser talked about nuance with the technical design of the Inter-blockchain-Communication protocol is the accounting system that takes place under the hood during the movement of assets between chains. Every transfer instantiates its own trail. ETH on Osmis is (osm)ETH) moving it from Osmosis to Kujira turns it into (osm-kuji)ETH). If moved to Cronos, the asset becomes (osm-kuji-cro)ETH). Therefore, it loses its composability because other ETH will have different trails depending on how it has been transferred. Imagine party A has 10 ETH in a liquidity pool on KAVA, and party B wants to provide liquidity in that same pool. Moving their 10 (osm-kuji-cro)ETH) to KAVA, party B's asset becomes (osm-kuji-cro-kava)ETH) which is different than Party A's (kava)ETH) in the pool; therefore, party B is unable to join the KAVA liquidity pools; unless there have been sophisticated customizations built in to accommodate that or they retrace their ETH's footprint. 4. IBC's Performance Waning The Interblockchain Communications Protocol was once heralded as one of the industry's most promising innovations in the interoperability space and contributed to the blossoming of Cosmos's hosting of over >45 independent chains. However, it seems that over the last year demands for IBC have been dwindling. Total transfers are down over -38%, Unique senders are down -48%, and Unique Receivers down -48%. If this trend continues, it could be a matter of time before we see the once-glorified technology disappear into an oblivion of obsolescence. 🧐 Opportunities (External) (Helpful) 1. ATOM 2.0 Initiatives have been put forth by the Foundation to transform ATOM tokenomics. A multitude of parameter adjustments to address centralization concerns, value accrual, and inflation. Already slowly being rolled out through governance proposal 848, the staking rate has been cut from ~14% to 10%, lowering the real yield from 19.3% to 13.4%. With lower inflation, there is a higher proclivity from users to hold on to their assets and find DEFI applications for it. Value accrual will be imposed through the alignment of independent environments with the use of ATOM tokens to facilitate cross-chain settlements. As it pertains to centralization, Cosmos currently scores 8 on the Nakamoto coefficient, meaning that it only takes 8 of the 177 validators to subvert the network. Under the new regime, there will be disincentives put in place for delegators that over-concentrate their allocation; a tax that increases alongside validator balances and, in turn, economically incentivizes end users to increase their delegations to smaller validators. 2. Widespread adoption of IBC This one is a bit of a wildcard, shot-in-the-dark, roll-of-the-dice potentiality, but one with gargantuan positive implications. In the event that the new ATOM token model delivers on its promise to adequately capture the value, and IBC delivers on its promise to provide seamless interoperability functionality, there is a good chance that Cosmos becomes further engrained into the industry through an acceleration of adoption. 3. Fork Proposal Jae Kwon, founder of Tendermint and co-creator of Cosmos, is known for his activism and radical attitude. He has recently come out and said that given the circumstance of Cosmos currently suffering from inadequate governance, it would be best to fork the network. By and large, forks a notoriously malicious event that can negatively impact the integrity of the original chain and result in communities being split. However, this appears to have an inverse potential for Cosmos as it would put an end to the continual disputes within the network and enrich the holders with new tokens. Be it purely speculative airdrop hunting or some genius economic resolution, talk of a possible fork has actually been accepted relatively well by the crypto markets. This can have very strong short-to-medium-term impacts on price as people try to front-run each other and qualify for the hard fork. 😳 Threats (External) (Harmful) 1. Terra Luna Debacle Built with the Cosmos SDK, Terra was an IBC-compatible network that handled tens of Billions of dollars in UST transactions. After a spectacular blowout that literally wiped hundreds of billions from the Crypto market's capitalization within days, the crippling whiplash crept across the entire industry, with Cosmos suffering a loss of over -75% 2 months post-mortem. Frankly, no direct value correlation between Cosmos and Terra Luna existed; however, based on the fact that their SDK was used, the image of Comsos has been tainted by possibly the most notable failures of any crypto protocol. 2. Declining Activity and Revenue In spite of an extremely positive market sentiment, Cosmos has seen declines across the board in terms of its utilization. The amount of daily active addresses has shrunk from >17,000 to ~16,000, the amount of daily transactions fell by nearly 33% from ~63,000 to ~40,000, and average revenue fell by nearly 50%, from a laughable $4,000 to an even more laughable $2,000. Luckily, the project does not seem to rely on the revenue it generates to sustain itself, otherwise it might have already been dead. 3. Rise of Advanced Alternative Competition Scalability and interoperability have been, are, and likely will continue to be the most aggressively researched sectors of the Web3 economy. With each new generation of blockchain-based technology, new superior solutions come to market with the capacity to outperform and displace predecessors. Rollups, Subnets, Parachains, Payment Channels, Nested side-chains; the growing amount of similar tech that enables similar capabilities currently available will likely continue to dominate mindshare and further squeeze OG's like Comsos out of public consciousness. Takeaway: Jokingly referred to as the "Dogecoin for Intellectuals", Cosmos is floating on the fringes between techno-economic paradise and irrelevance. The architectural pivot towards becoming providers of a validator set for other networks is not something entirely new; other projects, such as Eigen Layer, are ahead in this regard. However, if Cosmos's approach finds better market adoption, it holds a lot of promise for the ATOM token. There remains an elephant in the room; valid rebuttals about a logical gap in Cosmos's economic modeling. If the security provision by validators is based on their stake, then the security itself hinges upon speculative elements; if the price goes down, then so does security. Conclusion: It seems that despite the internal techno-political headwinds, Cosmos continues to be a domineering force on the layer 0/1 scene. Outside of what was mentioned earlier, is the fact that Cosmos has the support and material endorsement from many of the industry's Giga-brains (such as Chris Burniske from Placeholder Ventures) that make the case as to why Cosmos is a better option than its two leading competitors Polkadot and Avalanche. Cosmos is living proof that even with an incredibly talented team, good intentions, innovative technology, and a flourishing community, nothing is ever guaranteed in crypto. A cool piece of information about Cosmos that is hardly ever mentioned is the brilliance in the intentions behind the Cosmos Hub's design. The Hub itself is just an account-based model based on an environment that does not support smart contracts. This minimalism is distinctly reminiscent of Bitcoin, where the product is hyper-focused and purpose-built to mitigate the attack vector surface area and prevent fees from running out of control. Thank you so much for reading,
I hope this serves you well on your journey. Live long and prosper 🥂 Also published here.

The Cosmos (ATOM) SWOT Analysis 

ON THE ELEVATION OF THE WESTERN COAST OF SOUTH AMERICA

ST. HELENA

FTX + SBF Chapter 11 (Bankruptcy) Court Filing by John J. Ray III, Nov 17, 2022 is part of . You can jump to any part in this filing . This is part 8/20 HackerNoon’s Legal PDF Series here Feature Image: Stable Diffusion AI, Prompt “Let the adults in” III. ACTION TAKEN SINCE MR. BANKMAN-FRIED’S DEPARTURE A. New Governance Structure 46. Many of the companies in the FTX Group, especially those organized in Antigua and the Bahamas, did not have appropriate corporate governance. I understand that many entities, for example, never had board meetings. 47. The following new independent directors (the “Directors”) have been appointed as directors of the primary companies in the FTX Group: (a) : Mitchell Sonkin is currently a Senior Advisor to MBIA Insurance Corporation in connection with the restructuring of the Firm’s insured portfolio exposure of the Commonwealth of Puerto Rico’s $72 billion of outstanding debt. He is also currently Chairman of the Board of the ResCap Liquidating Trust, successor to ResCap and GMAC Mortgage Corporations. Before joining MBIA, Mr. Sonkin was a senior partner at the international law firm, King & Spalding, where he was co-chair of King & Spalding’s Financial Restructuring Group and a member of the firm’s Policy Committee. He has over 40 years of experience in U.S. and international bond issuances, corporate reorganizations, bankruptcies and other debt restructurings and has served as a bankruptcy-court-appointed examiner. In particular, he has played a significant role in numerous municipal, utility, insurance, airline, healthcare debt and international debt restructurings including the Anglo/French Euro Tunnel debt reorganization. WRS Silo: Mitchell I. Sonkin (b) : Mr. Rosenberg is a Partner at Lincoln Park Advisors, a financial advisory firm that he founded in 2014. He has more than 25 years of restructuring, corporate finance, principal investing, operating and board experience. Prior to founding Lincoln Park Advisors, he was a partner at the restructuring and investment banking firm Chilmark Partners, a partner in two private equity funds, the Zell/Chilmark Fund and Chilmark Fund II, the Chief Restructuring Officer of The Wellbridge Company and a member of multiple corporate boards. His restructuring advisory experience includes such companies as OSG, Supermedia, Nortel, Trinity Coal, USG Corporation, JHT Holdings, Inc., Covanta Energy, Sirva, Lodgian, Inc., ContiGroup Companies, Inc., Fruit of the Loom, Ltd. and Recycled Paper Greetings. Alameda Silo: Matthew R. Rosenberg (c) : Mr. Jain is a Managing Director and CoHead of the Western Region of Accordion, a financial and technology consulting firm focused on the private equity industry. He has more than 25 years of experience supporting management teams and leading finance and operations initiatives in both stressed and distressed environments. Prior to joining Accordion, Mr. Jain was part of Alvarez & Marsal’s corporate restructuring and turnaround practice for over 10 years and served in a variety of senior financial operating roles. His most notable assignments have included helping lead the restructuring, liquidation and wind down of Washington Mutual and its predecessor entity, WMI Liquidating Trust. He also navigated the restructuring of Global Geophysical Services in its chapter 11 and eventually the liquidation and wind down in its second chapter 11 filing. Ventures Silo: Rishi Jain : Mr. Farnan served as a United States District Judge for the District of Delaware from 1985 to 2010. He served as Chief Judge from 1997-2001. During his tenure, Mr. Farnan presided over numerous bench and jury trials involving complex commercial disputes. Prior to his appointment to the federal bench, Mr. Farnan was appointed to several positions in local, state and the federal government returning to private practice in 2010 with the formation of Farnan LLP, a law firm focused on complex commercial matters, including chapter 11 proceedings, securities litigation, antitrust litigation and patent litigation. Additionally, Mr. Farnan serves as an arbitrator, mediator, independent director and trustee of businesses contemplating or filing chapter 11 bankruptcy. (d) Dotcom Silo: The Honorable Joseph J. Farnan (Lead Independent Director) : Mr. Doheny is President of North Country Capital LLC, an advisory and investment firm focused on challenging advisory assignments and investing private investment portfolios in special situation opportunities. He has held this position since January 2011. Mr. Doheny has served on the board of directors or as Chief Restructuring Officer of numerous stressed and distressed companies, including Yellow Corp., MatlinPatterson, GMAC Rescap and Eastman Kodak. He was also Managing Director and Head of Special Situations Investing at HSBC Securities Inc. from 2015 to 2017. Previously, Mr. Doheny served as Portfolio Manager in Special Situations at Fintech Advisory Inc. from 2008 to 2010 and as Managing Director of the Distressed Products Group at Deutsche Bank Securities Inc. from 2000 to 2008. (e) Dotcom Silo: Matthew A. Doheny 48. The appointment of the Directors will provide the FTX Group with appropriate corporate governance for the first time. 49. The Directors intend to hold joint board meetings of the Debtors on matters of common interest, including (a) the implementation of controls, (b) asset protection and recovery, (c) the investigation into claims against the founders and third parties, (d) cooperation with insolvency proceedings of subsidiary companies in other jurisdictions and (e) the maximization of value for all stakeholders through the eventual reorganization or sale of the Debtors’ complex array of businesses, investments and property around the world. The Directors will implement appropriate procedures for the resolution of any conflicts of interest among the Silos and, if necessary, within the Silos as the case progresses, including the potential engagement of independent counsel to represent various Debtors in the resolution of intercompany claims against other Debtors. I expect there to be a multitude of intercompany claims that will benefit from fair resolution under the rules and conventions of U.S. chapter 11 practice in the District of Delaware for complex, multi-Debtor cases. For the time being, my belief is that all stakeholders are best served by a coordinated and centralized administration. Continue reading here

FTX 'did not have appropriate corporate governance', 'never had board meetings', says John Ray

FTX 'no tenía un gobierno corporativo apropiado', 'nunca tuvo reuniones de directorio', dice John Ray

FTX "n'avait pas de gouvernance d'entreprise appropriée", "n'a jamais eu de réunions du conseil d'administration", déclare John Ray

जॉन रे कहते हैं, एफटीएक्स के पास 'उचित कॉर्पोरेट प्रशासन नहीं था', 'कभी बोर्ड मीटिंग नहीं हुई'

FTXは「適切なコーポレートガバナンスを持っていなかった」、「取締役会を開催したことがない」とジョン・レイは言う

FTX 'não tinha governança corporativa adequada', 'nunca teve reuniões do conselho', diz John Ray

John Ray nói FTX 'không có quản trị công ty phù hợp', 'không bao giờ có cuộc họp hội đồng quản trị'

FTX“没有适当的公司治理”，“从未召开过董事会会议”，John Ray 说

FOMO is a one-time event, whereas product adoption is a long-lasting commitment. Before we officially start, let's take a step back and think about one question: what do DeFi farmers want? When evaluating the marketing strategy, we can simply look at the big picture and then discover the wants of users. Therefore, we build up the user journey map by eliminating users’ concerns before moving to the next step. Actual net earnings -APY/APR on your actual position -Rewards in stables or natives token -Enough of a buffer to avoid liquidation -Time and efforts consumed -Gas fees and swapping fees Ease of use & UI/UX -Straightforward navigation -Smooth onboarding flow (wallet, swap, bridge, porfolio tracking) Risk -Contract security -Rug-pull -Bank run risk Trustworthiness -Protocol narratives -Market and potential -Team background -Open source code Backers/investors -Reputation of VC -Ecosystem resources Market traction -Visibility of project -Partnership and accessibility Tokenomics -Token distribution & emission -Token utility -Token liquidity Phase 1 Main task: Building content assets and social assets from day 0 Finding the product-market fit is always the key to building a new product. If you’re looking for true protocol/product users, content marketing and community building will be the best leverage when doing effective marketing under an affordable budget. You can imagine all the docs, files, and even instant communication in your community are selling your product to the public, then how would you prefer to do the introduction? Content Marketing Crypto is a typical content-driven space. Identify how users get value from your product, which is yield in DeFi space. Adding a hook to explain how your protocol maximizes users’ yields by taking a bold claim response to likely skepticism. Once value props are set, it’s important to elaborate the mechanics behind the scenes and know your users understand where the yields are coming from. All in all, Twitter and Medium are the major platforms you can leverage your writing skills to impact the DeFi community effectively. Acquisition channels: Twitter, Gitbook, Medium, Youtube Community Building Community assets can help drive awareness, adoption, and retention. Building a clear, systematic, and rewarding community guild is key in the early stages. Manage the seed/early members and lead them to produce organic UGC content. Besides, Founder is the best community manager; talks to the users, answer their questions, figures out the user demand, executes, and improves. Acquisition channel: Discord, Telegram Security Protection Security is the paramount factor when it relates to DeFi. Getting a third-party secuirty company to audit the code is a basic move. Multi-sig of the smart contracts also leads to decentralized and lowers the concern of rug-pulled. Launching a bug bounty program helps involve the border DeFi community in security efforts, particularly in cases involving white hats. Security Company: Trait of Bits, OpenZepperlin, Quanstamp, Certik, etc Community audit platform: Core4rena, Sherlock Bug bounty platform: Github, ImmuniFi Once laid the solid foundation of content and social assets, it’s time to move to the next stage: get the word out. Phase 2 Main task: Executing the Acquisition strategy Partnership Synergize Fostering key relationships with many prominent projects throughout the space can help you gain a lot of advantages, including but not least new users, visibility, and credibility. It usually comes with protocol/product level features (collateral support, LP tokens usage) so that you may drag users directly from the existing user pool that has been tested out by your partners. Frankly speaking, that’s the priority acquisition channel in crypto. Some projects choose to conduct the public sales with CEX/DEX launchpad before launching so that the token can be partially distributed and generate buzz, however, this may be linked to CEX/DEX listing and MM on the secondary market for overall ideal effects. When it comes to projects that are about to debut without a token, leveraging the partnership/integration would be an effective weapon. Reference: Benqi on Avalanche Viral campaign/UGC Management Finding the gem(alpha) in crypto is the universal meme and good marketers should make good use of this. Expectations can be manipulated. Early adopters campaign, OG NFT, testnet users, and initial LP are on the top of the reward list when building a brand-new community. There’re many ways (e.g. meme contests, content challenges, trading competitions, ambassador programs) to incentivize the community to follow your social, interacted with protocol, and even referral their friends to join the game. NFT Platforms: POAP, Project Galaxy Community Platform: Gitcoin, Layer3 Influencer Marketing Strategy Compared to web2, web3 and crypto are way too niche at the current moment. Instead of the TV and billboards, users rely heavily on the content creators on Twitter and Youtube. Working with legit influencers in the sectors (NFT, DeFi, GameFi, CEX) your project specialized would accelerate your awareness, create valuable content for targeting audiences from the third-party dependencies, then enhance brand assets overall. Geographically, English-speaking countries/regions would undoubtedly gain ground, while regional countries such as China, Russia, Japan, and Vietnam are among the most active cryptocurrency users. Implementing the KOL strategy can be hugely effective - add so much value to the TVL at the very beginning. SEO Optimization Google search consists of a significant part of where those users are coming from. It’s typical crypto users’ behavior that they would do their own research over Google. It’s important what kind of Brand image you would to provide with them and how compelling the message is conveyed through Google indexed. Solution: keywords optimization, reputable media coverage, own-media operation Until now you should have a clear launching strategy ready for the show. Phase 3 Main task: Trigger the positive usage cycle Attractive APY for product adoption and arbitrage opportunity APR/APY is variable, however for DeFi users, it's a significant signal. Many projects launch aggressive liquidity mining programs to kickoff and inject three-digit APY for assets pair to bootstrap initial liquidity. However, it’s worth considering the balance between emissions and rewards to avoid a death spiral once the liquidity mining ends. Reference: Benqi on Avalanche, Aave on Polygon The protocol leverages bAsset rewards to catalyze a positive usage cycle: subsidies incentivize new stablecoin deposits, lowering the borrowing rate, which incentivizes more bAsset-collateralized loans, and enables more bAsset rewards to be collected. Building TVL foundation TVL is the no.1 metric in DeFi and having a steadily growing TVL chart suits long-term projects because you know exactly where your users are coming from and gradually raise the numbers. Building TVL (say $10m for a new project) can bring greater opportunity and trust from the broader audience, thus leading to a healthy circle. There’s also some useful tracking websites, DeFi yield charting, and wallet management to foster growth during the new users’ onboarding journey. Integrations: DeFillama, Coindix, Debank, Zapper. Keep Shipping Launching is just the beginning and it’s important to deliver a roadmap following up to enhance the hype, ranging from product features, token ignition, and partnerships to community grants. Integrating the launching strategy with good metrics into a good narrative to earn more collaboration opportunities. Rather than simply hearing about the launch, the community is hoping for a successful story. At this point, a large amount of traffic and users usually flood in. Promotions: flash news, content creator release, mention/featured in the media Closing Thoughts Web3 is a profit-focus space and DeFi has always competed for APY, but how to offer competitive, safe, and sustainable yield farming products in the market decides how much liquidity you can bootstrap.  FOMO is a one-time event, whereas product adoption is a long-lasting commitment. Cheers!

Web3.0 Marketing Playbook #1 | Bootstrap the Initial Liquidity for a New DeFi Startup

How to fool a 27M-parameter model with a bit of Python Do you think it is impossible to fool the vision system of a self-driving Tesla car? Or that machine learning models used in malware detection software are too good to be evaded by hackers? Or that face recognition systems in airports are bulletproof? Like any of us machine learning enthusiasts, you might fall into the trap of thinking that deep models used out there are perfect. Well, you are WRONG. There are easy ways to build that can fool any deep learning model and create security issues. In this post, we will cover the following: adversarial examples What are adversarial examples? How do you generate adversarial examples? Hands-on example: let’s break Inception 3 How to defend your models against adversarial examples All the source code shown in this article is in ** **👇🏽 this Github repo Let’s start! 1. What are adversarial examples? 😈 In the last 10 years, deep learning models have left the academic kindergarten, become big boys, and transformed many industries. This is especially true for computer vision models. When hit the charts in 2012, the deep learning era officially started. AlexNet Nowadays, computer vision models are as good or better than human vision. You can find them in myriad places, including… self-driving cars face recognition medical diagnosis surveillance systems malware detection Until recently, researchers trained and tested machine learning models in a environment, such as in machine learning competitions and academic papers. Nowadays, when deployed in real-world scenarios, security vulnerabilities coming from model errors have become a real concern. laboratory Imagine for a moment that the state-of-the-art-super-fancy-deep learning vision system of your self-driving car was not able to recognize as a stop sign. this stop sign Well, . This stop sign image is an . Think of it as an optical illusion for the model. this is exactly what happened adversarial example Let us look at another example. Below you have two images of a dog that are indistinguishable for us humans. The image on the left is an original picture of a puppy taken by . The one on the right is a slight modification of the first, that I created by adding the noise vector in the central image. Dominika Roseclay Inception v3 correctly classifies the original image as a dog breed (a redbone). However, this same model thinks, with very high confidence, that the modified image I created is a paper towel. In other words, I created an adversarial example. And you will too, as this is the example we will work on later in section 3. An for a computer vision model is an input image with small perturbations, imperceptible to the human eye, that causes a wrong model prediction. adversarial example Do not think these 2 examples are rare edge-case examples found after spending tons of time and computing resources. There are easy ways to generate adversarial examples, and this opens the door to serious vulnerabilities of machine learning systems in production. Let’s see how you can generate an adversarial example and fool a state-of-the-art image classification model. 2. How to generate adversarial examples? Adversarial examples 😈  are generated by taking a clean image 👼 that the model correctly classifies, and finding a small perturbation that causes the new image to be misclassified by the ML model. White-box scenario Let’s suppose you have complete information about the model you want to fool. In this case, you can compute the loss function of the model where is the input image X is the output class, y and is the vector of the network parameters. θ This loss function is typically the negative log-likelihood for classification methods. Your goal is to find a new image that is close to the original and that produces a big change in the value of the loss function. X’ X Imagine you are inside the space of all possible input images, sitting on top of the original image . This space has dimensions , so I will excuse you if you cannot visualize it well 😜. X width x height x channels To find an adversarial example, you need to walk a little bit in some direction in this space until you find another image with a remarkably different loss. You want to choose the direction that maximizes the change in the loss function for a fixed small step . X’ J epsilon Now, if you refresh a bit your Maths Calculus course, the direction in the space where the loss function changes the most is precisely the gradient of with respect to the . X J X The gradient of a function with respect to one of its variables is precisely the direction of maximal change. And by the way, this is the reason people train Neural Networks using Stochastic Gradient Descend and not Stochastic Random-Direction Descend. Fast gradient sign method An easy way to formalize this intuition is as follows: We take only the sign of the gradient and scale it using a small parameter epsilon, to guarantee that the distortion between and is small enough to be imperceptible to the human eye. This method is called the . X X fast gradient sign method Black-box attack In most scenarios, it is very likely you will not have complete information about the model. Hence, the previous method is not useful as you cannot compute the gradient. However, there exists a remarkable property called of adversarial examples that malicious agents can exploit to break a model even if they do not know its internal architecture and parameters. transferability Researchers have repeatedly observed that quite well between models, meaning that they can be designed for a target model A, but end up being effective against any other model trained on a similar dataset. adversarial examples transfer Adversarial examples can be generated as follows: Query the targeted model with inputs X_i for i = 1 … n and store the outputs y_i. Use the training data (X_i, y_i) to build another model, called the substitute model. Use a white-box algorithm like the fast gradient sign to generate adversarial examples for the substitute model. Many of them are going to transfer successfully and become adversarial examples for the target model as well. A successful application of this strategy against a commercial Machine learning model is presented in . this Computer Vision Foundation paper 3. Hands-on example: let’s break Inception 3 Let’s get our hands dirty and implement a few attacks using Python and the great library PyTorch. It always comes in handy to know how the attacker thinks. You can find the complete code in . this Github repo Our target model is going to be Inception V3, a powerful image classification model developed by Google, which has around 27M parameters and was trained on around 14M images belonging to 20k categories. # Load pretrained model from the PyTorch hub
# https://pytorch.org/hub/pytorch_vision_inception_v3/
from torchvision.models import inception_v3
model = inception_v3(pretrained=True)
model.eval()

# Count model parameters: 27,161,264
n_params = sum(p.numel() for p in model.parameters())
print(f'{n_params:,} parameters') We download the list of classes the model was trained on, and build an auxiliary dictionary that maps class ids to labels. # Download the txt file with the list of ImageNet classes the model was trained with
#     "!" magic to run shell commands from a Jupyter notebook
!wget https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt

# id2label maps classes ids to their human-readable names: e.g. id2label[1] = 'goldfish'
with open("imagenet_classes.txt", "r") as f:
    categories = [s.strip() for s in f.readlines()]
    
id2label = {}
for idx, category in enumerate(categories):
    id2label[idx] = category Let us take the image of an innocent redbone dog as the starting image we will carefully modify to build adversarial examples: import requests
import io
from PIL import Image

url = 'https://previews.123rf.com/images/meinzahn/meinzahn1211/meinzahn121100339/16350068-cute-tiger-cat-isolated-on-white.jpg'

response = requests.get(url)
img = Image.open(io.BytesIO(response.content))
img Inception V3 expects images with dimensions 299 x 299 and normalized pixel ranges between -1 and 1. Let us preprocess our beautiful dog image: import torch
from torch import Tensor
from torchvision import transforms

def preprocess(img) -> Tensor:
    """
    Inception V3 model from pytorch expects input images with pixel values between -1 and 1
    and dimensions 299 x 299
    """
    mean = [0.485, 0.456, 0.406]
    std = [0.229, 0.224, 0.225]

    preprocess_fn = transforms.Compose([
        transforms.Resize((299,299)),  
        transforms.ToTensor(),
        transforms.Normalize(mean, std)
    ])
    image_tensor = preprocess_fn(img)

    # add batch dimension: C x H x W ==> B x C x H x W
    image_tensor = image_tensor.unsqueeze(0)
    
    return image_tensor
        
x = preprocess(img) and check that the model correctly classifies this image. from easydict import EasyDict
import torch.nn.functional as F

def get_predictions(img: Tensor) -> EasyDict:

    output = model.forward(img)

    class_idx = torch.max(output.data, 1)[1][0].item()

    label = id2label[class_idx]

    output_probs = F.softmax(output, dim=1)
    confidence =  round(torch.max(output_probs.data, 1)[0][0].item(), 4)
    
    return EasyDict(
        id=class_idx,
        label=label,
        confidence=confidence,
    )

get_predictions(x) # {'id': 168, 'label': 'redbone', 'confidence': 0.8861} Good. The model works as expected and the dog is classified as a dog :-). redbone redbone Let’s move to the fun part and generate adversarial examples using the Fast Gradient Sign method. from typing import Tuple
from torch.autograd import Variable

def fast_gradient_sign(x: Tensor, eps: float) -> Tuple[Tensor, Tensor]:
    """"""
    # convert tensor into a variable, because we will need to compute gradients
    # of the loss function with respect to the image pixels
    img_variable = Variable(x, requires_grad=True)
    
    # forward pass on the original image
    output = model.forward(img_variable)
    
    # get predicted class
    y_true = torch.max(output.data, 1)[1][0].item()   
    target = Variable(torch.LongTensor([y_true]), requires_grad=False)
    
    # backward pass to compute gradients
    loss_fn = torch.nn.CrossEntropyLoss()
    loss = loss_fn(output, target)
    
    # this will calculate gradient of each variable (with requires_grad=True)
    # that you can later access with "var.grad.data"
    # PyTorch does the heavy lifting, computing the gradient of the cross-entropy
    # with respect to the input image pixels.
    loss.backward(retain_graph=True)
    
    # sign of gradient of the loss func (with respect to input X)
    x_grad = torch.sign(img_variable.grad.data)
    
    # fast gradient sign formula
    x_adversarial = img_variable.data + eps * x_grad
        
    return x_adversarial, x_grad

# keep epsilon small to generate slight changes to the original image
epsilon = 0.02
x_adv, grad = fast_gradient_sign(x, epsilon) I have created an auxiliary function to both the original and the adversarial image. You can see the full implementation in this . visualize GitHub repository Well. It is interesting how the model prediction changed for the new image, which is almost indistinguishable from the original one. The new prediction is a , which is another dog breed with very similar skin color and big ears. As the puppy in question could be a mixed breed, the model mistake seems to be small, so we want to work further to really break this model. bloodhound One possibility is to play with different values of and try to find one that clearly gives a wrong prediction. Let’s try this. epsilon epsilons = [0.02, 0.2, 0.9]
for epsilon in epsilons:
    x_adv, grad = fast_gradient_sign(x, epsilon)
    print('epsilon: ', epsilon)
    visualize(x, x_adv, grad, epsilon) As epsilon increases the change in the image becomes visible. However, the model predictions are still other dog breeds: and . We need to be smarter than this to break the model. bloodhound basset Remember the intuition behind the Fast gradient sign method, i.e. imagine yourself inside the space of all possible images (with dimension 299 x 299 x 3), right where the original image X is. The gradient of the loss function tells you the direction you need to step to increase its value and make the model less certain about the right prediction. The size of the step is epsilon. You step and you check if you are now sitting on an adversarial example. An extension of this would be to take many smaller steps, instead of a single one. After each step, you re-evaluate the gradient and decide the new direction you are going to walk towards This method is called the Iterative Fast Gradient method. What an original name! More formally: Where , and denotes clipping of the input in the range of [X−ϵ, X+ϵ]. X0 = X Clip X,ϵ An implementation in PyTorch is the following: def iterative_fast_gradient_sign(x_: Tensor, epsilon: float, n_steps: int, alpha: float):
    """"""
    # copy to avoid modifying the original tensor
    x = x_.clone().detach()
    
    for step in range(n_steps):
         
        # one step using basic FGSM
        x_adv, grad = fast_gradient_sign(x, alpha)
        
        # total perturbation
        total_grad = x_adv - x_
        
        # force total perturbation to be lower than epsilon in
        # absolute value
        total_grad = torch.clamp(total_grad, -epsilon, epsilon)

        # add total perturbation to the original image
        x_adv = x_ + total_grad
               
        x = x_adv
        
    return x_adv, total_grad Now, let us try again to generate a good adversarial example starting with our innocent puppy. : bloodhound again Step 1 : beagle again Step 2 : mousetrap? Interesting. However, the model confidence on this prediction is only 16%. Step 3 Let’s go further. : One more dog breed, boring… Step 4 : beagle again.. Step 5 : Step 6 : redbone again. Step 7 Keep calm and continue walking in the image space. : Step 8 : BINGO! 🔥🔥🔥 Step 9 What an exotic paper towel. And the model is pretty confident about its prediction, almost 99%. If you compare the initial and the final image we found in step 9 you see they are essentially the same, a , but for the Inception V3 model they are two very different things. puppy I was very surprised the first time I saw such a thing. How small modifications to an image can cause such model misbehavior. This example is funny, but if you think about the implications for a self-driving car vision you might start to worry a bit. Deep learning models deployed in critical tasks need to properly handle adversarial examples, but how? 4. How to defend your models against adversarial examples? As of November 2021, there is no universal defense strategy you can follow to protect against adversarial examples. In other words, attacking a model with adversarial examples is easier than protecting it. The best universal strategy, that can protect against any known attack, is to use adversarial examples when training the model. If the model adversarial examples during training, its performance at prediction time will be better for adversarial examples generated in the same way. This technique is called . sees adversarial training You generate them on the fly, as you train the model, and adjust the loss function to look both at clean and adversarial inputs. For example, we could eliminate the adversarial examples we found in the previous section if we added the 10+ examples to the training set and labelled all of them as . redbone This defense is very effective against attacks that use the Fast Gradient Sign method. However, there exist more powerful attacks, that we did not cover in this post, that can bypass this defense. If you wanna know more I encourage you to read by Nicholas Carlini and David Wagner. this amazing paper This is the reason why adversarial defense is an open problem in Machine Learning and cybersecurity. Conclusion Adversarial examples are a fascinating topic at the intersection of cybersecurity and machine learning. However, it is still an open problem waiting to be solved. In this post, I gave a practical introduction to the topic with code examples. If you would like to work further I suggest , a Python library for adversarial machine learning developed by the great Ian Goodfellow and Nicolas Papernot, and currently maintained by the . cleverhans University of Toronto All the source code I shown in this article is in . this Github repo Wanna become a PRO in Machine Learning? I offer hands-on content on real-world Machine Learning, that helps you get a top job in the Data world. 👉🏽  Subscribe to the . datamachines newsletter 👉🏽  Follow me on and . LinkedIn Twitter Have a great day 🧡 Pau

Adversarial Examples In Machine Learning Explained

In 2021, hosted the first-ever awards!! HackerNoon Startup of the Year The goal of this campaign was to give recognition to the hardworking startups who pulled through the pandemic -- growing and thriving (despite all of the challenges). More than 44,000 startups were nominated all across the globe! They were selected from over 4,553 cities, with a total number of * cough * “4 billion” votes. Insane, right? To see who is a winner in each respective city, visit and pick a city. The name at the top of the list is the winner. (Compromised awards are under review. Cities with tie votes will be updated to have two winners!) startups.hackernoon.com The will announce the winners per region in their respective cities. Startups Winners Announcement series Here are the winners in the : Asia region Top Startup Cities in East Asia Seoul The Startup of the Year in is with 377 votes! Seoul dable Central is the Startup of the Year! It came out ahead with 121 votes. UpNow Central Sheung Wan came out as the Startup of the Year in with 154 votes. Votee Sheung Wan Tokyo The Startup of the Year in is with 41 votes! Tokyo ACES Wan Chai The Startup of the Year in is with 54 votes! Wan Chai Zetl Osaka, Japan In , is the winner! Osaka Nanophoton Taiwan came out as the Startup of the Year in ! Tico Taiwan Dongcheng, China In , is declared the Startup of the Year! Dongcheng Waiju Taipei came out as the Startup of the Year in ! Tico Taipei Akasaka In , is the winner! Akasaka WAKUWAKU Chuo In , is declared the Startup of the Year! Chuo GO-NET Ginza came out as the Startup of the Year in ! Netyear Ginza Kanazawa In , is the winner! Kanazawa Granzella Minami-aoyama In , is declared the Startup of the Year! Minami-aoyama Tokyo6 Entertainment Shibuya came out as the Startup of the Year in ! Hupro Japan Shibuya Toranomon In , is the winner! Toranomon Small Worlds Toshima In , is declared the Startup of the Year! Toshima Altplus Guangzhou, China came out as the Startup of the Year in ! Leyaoyao Guangzhou Hong Kong, Hong Kong In , is declared the Startup of the Year! Hong Kong Episode Six Kunming In , is the winner! Kunming Green A Sapporo In , is declared the Startup of the Year! Sapporo Polar Star Space Shirokanedai came out as the Startup of the Year in ! Nupp1 Shirokanedai Wuhan In , is the winner! Wuhan CSICEE Baoan, China In , is declared the Startup of the Year! Baoan Jumper Baoshan, China In , is declared the Startup of the Year! Baoshan Renren Games Chengdu, China came out as the Startup of the Year in ! Ju You Xi Chengdu Fujisawa, Japan In , is the winner! Fujisawa Scohia Pharma Fukuoka, Japan In , is declared the Startup of the Year! Fukuoka Kyulux Haidian, China came out as the Startup of the Year in ! 51vv Haidian Hongkou, China In , is the winner! Hongkou Kidstone Jiading, China In , is declared the Startup of the Year! Jiading IMRobotic Kawasaki, Japan came out as the Startup of the Year in ! QD Laser, Inc. Kawasaki Kyoto, Japan In , is declared the Startup of the Year! Kyoto Hacarus Minato, Japan In , is the winner! Minato Visionary Fun Nagoya, Japan came out as the Startup of the Year in ! Prodrone Nagoya Ningbo, China The winner in is . Ningbo OlymTech Corporation Aomori In , is declared the Startup of the Year! Aomori Forte Japan Bandai-cho In , is the winner. Bandai-cho Delta-Fly Pharm Changchun In , the Startup of the Year is . Changchun Kaidi Animation Fuchu In , is declared the Startup of the Year! Fuchu TMS Fukuoka-shi The winner in is . Fukuoka-shi SkyDisc Fukushima In , the Startup of the Year is . Fukushima Matsuri Technologies Gotanda In , is the winner. Gotanda Basset Hachioji In , is declared the Startup of the Year! Hachioji Adriakaim Harbin The winner in is J . Harbin oining Technology Iwaki In , is the winner. Iwaki Tao Engineering Izumo In , the Startup of the Year is . Izumo PuREC Kagoshima In , is declared the Startup of the Year! Kagoshima Farmers Support Kamakura In , the Startup of the Year is . Kamakura KAYAC Kami-meguro The winner in is . Kami-meguro Buzzreach Kanagawa In , is the winner. Kanagawa LoiLo Kashiwa In , the Startup of the Year is . Kashiwa Intelligent Surfaces Kita In , is declared the Startup of the Year! Kita Ax Robotix Kobe The winner in is . Kobe Coconomi Koriyama In , the Startup of the Year is . Koriyama Tocoton Kumamoto In , is the winner. Kumamoto Bloom Technology Kunitachi In , is declared the Startup of the Year! Kunitachi Trys Liaocheng The winner in is . Liaocheng Jinzheng Animation Minami In , is declared the Startup of the Year! Minami ATLAB Miyazaki In , is the winner. Miyazaki Caster Nagano In , is declared the Startup of the Year! Nagano Precision Forestry Nagaoka The winner in is . Nagaoka Plant Form Nihombashi-hamacho In , the Startup of the Year is . Nihombashi-hamacho Shizen Energy Niigata In , is declared the Startup of the Year! Niigata UniBio Japan Nishi-shinagawa In , is the winner. Nishi-shinagawa Play Heart Okayama In , is the winner! Okayama Arch Farm Okinawa In , is declared the Startup of the Year! Okinawa BlogNinja Otemachi In , the Startup of the Year is . Otemachi beBit, Inc Qingdao Shi In , is the winner! Qingdao Shi Manyoubang Quanzhou In , is declared the Startup of the Year! Quanzhou KungFu Animation Group Roppongi The winner in is . Roppongi Holo-X Saitama In , the Startup of the Year is . Saitama GameTector Sendai In , is the winner! Sendai Sendai Smart Machines Setagaya In , is declared the Startup of the Year! Setagaya Skyestate Shenzhen In , the Startup of the Year is . Shenzhen Aiator Shinjuku In , is declared the Startup of the Year! Shinjuku Scalar Shizuoka In , is the winner! Shizuoka PiPhotonics Sumida The winner in is . Sumida Holotch Tachikawa In , the Startup of the Year is . Tachikawa PlusOne Takamatsu In , is declared the Startup of the Year! Takamatsu Ken Robotec Tokushima In , is the winner! Tokushima DennoKotsu Toyonaka In , the Startup of the Year is . Toyonaka FISH BIOTECH Tsukuba In , is declared the Startup of the Year! Tsukuba WARPSPACE Wakayama In , the Startup of the Year is . Wakayama 4D Sensor Xiangzhou In , is the winner! Xiangzhou ALLiN Edutech Group Xianyang The winner in is . Xianyang TONLY Xiaoshan In , is declared the Startup of the Year! Xiaoshan GONGHUI Yamaguchi In , the Startup of the Year is . Yamaguchi Archis Yangpu The winner in is . Yangpu Tmxmall Yau Tsim Mong In , is the winner! Yau Tsim Mong Airnugget Yeoksamdong In , the Startup of the Year is . Yeoksamdong WizSchool Yichang In , is declared the Startup of the Year! Yichang Hubei Kento Electronic Yokohama In , is the winner! Yokohama Next Innovation Yubei In , the Startup of the Year is . Yubei Ophiux Network Yueqing The winner in is . Yueqing Yueqing Easythink Media Zhuhai In , is the winner! Zhuhai Aijuba Hongo In , the Startup of the Year is . Hongo Smile Robotics, Inc Admiralty In , is declared the Startup of the Year! Admiralty BUNCH Southern In , the Startup of the Year is . Southern Retykle Yongdungpo In , is the winner! Yongdungpo Push News Taichung In , is declared the Startup of the Year! Taichung Wondercise Cyberport, Hong Kong In , the Startup of the Year is . Cyberport Animoca Brands Hefei, China In , is the winner! Hefei Tsinglink New Territories, Hong Kong The winner in is . New Territories, Hong Kong Zeek Dezhou In , the Startup of the Year is . Dezhou Dezhou Baji Kitakyushu In , is declared the Startup of the Year! Kitakyushu Reif Seongnam In , is the winner! Seongnam EOFlow Yanshi In , the Startup of the Year is . Yanshi CASI Vision Top Startup Cities in South Asia Hyderabad, India The Startup of the Year in is with 962 votes Hyderabad Quixy Lahore, Pakistan is the Startup of the Year! It came out ahead with 186 votes. SmartWindows Lahore Bangalore, India When it comes to , is the Startup of the Year with 210 votes! Bangalore vBridge Hub Pune, India couldn’t be beaten for Startup of the Year in , it won with 129 votes! InfraCloud Pune Jaipur, India The Startup of the Year in is with 166 votes Jaipur Emizen Tech Lucknow, India is the Startup of the Year! It came out ahead with 279 votes. Keeros Superfoods Lucknow Noida, India came out as the Startup of the Year in with 156 votes! Bagisto Noida Mumbai, India When it comes to , is the Startup of the Year with 96 votes! Mumbai Sequretek Sylhet couldn’t be beaten for Startup of the Year in , it won with 122 votes! DataXpie Sylhet Mylapore, India In , is the winner! Mylapore Yalimart Goa, India In , the Startup of the Year is Goa Filter Coffee Delhi, India In , is declared the Startup of the Year! Delhi EkAnek Networks Karachi, Pakistan The winner in is Karachi Bykea Ahmedabad, India In , is the winner Ahmedabad MageComp Dhaka, Bangladesh The winner in is Dhaka SOLshare Islamabad, Pakistan In Islamabad, is declared the Startup of the Year! TAG Gurgaon, India In , the Startup of the Year is Gurgaon ixigo Bengaluru, India In , is the winner! Bengaluru Prakshep Private Limited Haryana, India couldn’t be beaten for Startup of the Year in ! Footprints Childcare Haryana Kolkata, India In , is the winner! Kolkata Wow! Momo New Delhi, India In , is declared the Startup of the Year! New Delhi Rusk Media Indore, India The winner in is Indore Chattar Patar Kormangala In , is the winner Kormangala Sanctum Technologies Ghaziabad came out as the Startup of the Year in Craftfurnish.com Ghaziabad! Mohali In , the Startup of the Year is Mohali DataKund Mysore The Startup of the Year in is Mysore Excelsoft Karnal came out as the Startup of the Year in ! Dunar Karnal Kolhapur In , is the winner! Kolhapur Zanvar Group Patiya In , the Startup of the Year is Patiya Kloud Project Kathmandu In , is declared the Startup of the Year! Kathmandu Vairav Technology Bangalore City, India In is the winner! Bangalore City, PagarBook Ernakulam The winner in is Ernakulam Epixel Solutions Colombo, Sri Lanka couldn’t be beaten for Startup in the Year in MAs Foods Colombo Andheri, India When it comes to , is the Startup of the Year! Andheri DoorMint Kanpur, India came out as the Startup of the Year in ! ServiceWalaa Kanpur Faisalabad In , is the winner! Faisalabad DigiKhata Nasik The Startup of the Year in is Nasik Sahyadri Farms Patna came out as the Startup of the Year in ! Asvin Tech Patna Chennai, India In , is the winner! Chennai Agnikul Cosmos Faridabad, India In , is declared the Startup of the Year! Faridabad UClean Panjim, India The winner in is Panjim TempoGO Ranchi, India In , is the winner Ranchi Osam Dairy Surat, India In , is declared the Startup of the Year! Surat Wishbook Thane, India In , is the winner! Thane Delight Foods Trivandrum, India came out as the Startup of the Year in ! GenRobotic Trivandrum Vijayawada, India In , the Startup of the Year is Vijayawada Eruvaka Technologies Visakhapatnam, India In , Ltd is declared the Startup of the Year! Visakhapatnam Greenline Labs Pvt Ahmadabad In , is the winner! Ahmadabad KiwiQA Services Bandra came out as the Startup of the Year in ! Car Club Bandra Bombay In , is the winner! Bombay Locobuddy Mobile Technology (Sportobuddy) Cochin In , the Startup of the Year is Cochin Tuttifrutti Interactive Gandhinagar In , is declared the Startup of the Year! Gandhinagar Kroop AI Guwahati In , the Startup of the Year is Guwahati GNRC Hospitals Jamshedpur In , is declared the Startup of the Year! Jamshedpur Intelliber Kabul The winner in is Kabul Miracle Shop (Vocinno Technology) Karur In , is the winner Karur Karur Vysya Bank Kochi The winner in is Kochi BestDoc Kozhikode In , is declared the Startup of the Year! Kozhikode OfficeKit HR Ludhiana In , the Startup of the Year is Ludhiana Trendy Beat Network Madhapur In , is the winner! Madhapur Head Digital Works Mangalore When it comes to , is the Startup of the Year Mangalore Hangyo Nagpur came out as the Startup of the Year in ! UPrudence Agro Nagpur Rawalpindi The Startup of the Year in is Rawalpindi Pro Options Marketing Rishikesh came out as the Startup of the Year in ! Descatuk Rishikesh Secunderabad couldn’t be beaten for Startup of the Year in Prakriti Biosystems Engineering Secunderabad Tirupur In , is the winner! Tirupur Tubeknit Fashions Udupi The winner in is Udupi Krishitantra Top Startup Cities in Southeast Asia Singapore, Singapore The Startup of the Year in is with 6556 votes! Singapore Cake DeFi Hanoi, Vietnam is the Startup of the Year! It came out ahead with 683 votes. DeFi For You Hanoi Kuala Lumpur, Malaysia came out as the Startup of the Year in with 109 votes! Trambellir Kuala Lumpur Pasig, Philippines When it comes to , . is the Startup of the Year with 110 votes! Pasig Quickreach Inc Bangkok, Thailand couldn’t be beaten for Startup in the Year in , it won with 60 votes! Perception Codes Bangkok Johor Bahru In , the Startup of the Year is . Johor Bahru AGR1 Bandung, Indonesia In , is declared the Startup of the Year! Bandung Prosa Portugal The winner in is . Portugal SCEMAI Tangerang, Indonesia In , is the winner Tangerang Yummy Corp Jakarta, Indonesia In , is the winner! Jakarta Tokocrypto Dki Jakarta, Indonesia In , is declared the Startup of the Year! Dki Jakarta Limakilo Bangkok Metropolis In , is the winner! Bangkok Meticuly Medan The winner in is ! Medan DIKA Manila, Philippines In , is declared the Startup of the Year! Manila PawnHero Phnom Penh, Cambodia In , the Startup of the Year is Phnom Penh vKirirom Bali In , is the winner! Bali Radius App Penang When it comes to , is the Startup of the Year Penang EasyParcel Surabaya came out as the Startup of the Year in ! Super Surabaya Setiabudi, Indonesia The Startup of the Year in is Setiabudi Chilibeli Cebu City came out as the Startup of the Year in MiCab Cebu City! Jawa couldn’t be beaten for Startup in the Year in DAUR Jawa Ho Chi Minh City, Vietnam In , is the winner! Ho Chi Minh City Pharmacity Selangor, Malaysia is the Startup of the Year! BabyDash Selangor Jakarta Pusat, Indonesia In , is the winner! Jakarta Pusat StickEarn Semarang When it comes to , is the Startup of the Year! Semarang Food Bear Petaling Jaya, Malaysia The winner in is . Petaling Jaya Deemples Kebayoran Baru, Indonesia couldn’t be beaten for Startup in the Year in ! Yogrt Kebayoran Baru Kuningan The winner in is . Kuningan Prinzio Bukit Merah came out as the Startup of the Year in ! DC Alliance Bukit Merah Huai Khwang In , the Startup of the Year is . Huai Khwang AppMan Co.,Ltd Kuta In , is declared the Startup of the Year! Kuta Gringgo Mandaluyong came out ahead as the Startup of the Year! Work Inspire Mandaluyong Midview City In , is the winner! Midview City BioMers Phuket The Startup of the Year in is ! Phuket Sapahn Yangon, Myanmar The winner in is . Yangon Wave Money Denpasar When it comes to , is the Startup of the Year! Denpasar Revasi Nakhon Ratchasima, Thailand The winner in is . Nakhon Ratchasima Siam Lottery co., Ltd Taguig, Philippines In , is the winner! Taguig PayMongo Bogor came out as the Startup of the Year in ! TIRANYX Bogor Senai, Malaysia won as the Startup of the Year. Nutrition Technologies Senai Bekasi Kota couldn’t be beaten for Startup in the Year in ! 8villages Bekasi Kota Top Startup Cities in Middle East Tel Aviv, Israel The Startup of the Year in is with 1,065 votes! Tel Aviv hyro Beirut, Lebanon is the Startup of the Year! It came out ahead with 1,276 votes. Hovi Digital Lab Beirut Riyadh, Saudi Arabia With 350 votes, came out as the Startup of the Year in ! Flicron Riyadh Tel Aviv-yafo, Israel When it comes to , is the Startup of the Year with 167 votes! Tel Aviv-yafo Cymbio Haifa, Israel couldn’t be beaten for Startup in the Year in , it won with 183 votes! Pico Haifa Amman, Jordan In , with 141 votes, is the winner! Amman Edexy Dubai, United Arab Emirates In , the Startup of the Year is with 74 votes. Dubai Ziina Doha, Qatar In , is declared the Startup of the Year! Doha SkipCash Tehran, Iran The winner in is . BeyondBoard AR Tehran Jerusalem, Israel In , is the winner! Jerusalem SolidBlock Rishon Le Zion The Startup of the Year in is ! Rishon Le Zion Passparto Ness Ziona, Israel is the Startup of the Year! Soapy Ness Ziona Kuwait, Kuwait came out as the Startup of the Year in ! Brapper Shop Kuwait Manama, Bahrain couldn’t be beaten for Startup in the Year in ! Doctori Manama Lod, Israel is the winner in ! Phinergy Lod Agudat Israel, Israel When it comes to is the Startup of the Year with [vote ] votes! Agudat Israel, Aqaurius Engines Beersheba, Israel In , the Startup of the Year is . Beersheba Visual Estate Ltd Giv'atayim, Israel In , is declared the Startup of the Year! Giv'atayim WSC Sports Modi'in, Israel The winner in is . Modi'in Loginno Nazareth, Israel In , is the winner! Nazareth AR Guide In Medical Ramat Gan, Israel The Startup of the Year in is . Ramat Gan Datos Health Baghdad When it comes to , is the Startup of the Year! Baghdad Mishwar Khobar In , the Startup of the Year is . Khobar Garage Misgav couldn’t be beaten for Startup in the Year in ! Eden Shield Misgav Sharjah came out as the Startup of the Year in ! Bookends Sharjah Affula In , the Startup of the Year is . Affula eMotion Technologies Bene Beraq, Israel In , is the winner Bene Beraq Reflectiz Caesarea, Israel In , the Startup of the Year is . Caesarea CardiacSense Dubayy In , is declared the Startup of the Year! Dubayy North Ladder Muscat couldn’t be beaten for Startup in the Year in ! iLab Marine Muscat Tirat Carmel In , the Startup of the Year is . Tirat Carmel PassCall Advanced Technologies Hod Hasharon, Israel The winner in is . Hod Hasharon Idomoo Holon, Israel came out ahead as the Startup of the Year! OptimalPlus Holon Jeddah, Saudi Arabia When it comes to , is the Startup of the Year! Jeddah Arabian Combustion Technologies Ramat Hasharon, Israel is the Startup of the Year! Lumen Ramat Hasharon Al Khobar came out as the Startup of the Year in ! Swyft Al Khobar Bet Shemesh In , is the winner! Bet Shemesh Degania Medical Binyamina When it comes to , is the Startup of the Year! Binyamina OncoHost Makkah In , the Startup of the Year is Makkah Scimagine Or Akiva Last but not least, couldn’t be beaten for Startup in the Year in the city of ! WMT Or Akiva More Information: If you’re a winner, please answer ! If you’re a reader, participant, just a curious individual, consider answering one of our interviews . this interview here Read the FAQs about or . Startups here here To see all of the winners you may visit , pick any city and view the name that appears at the top. this site If you’re interested in partnering with us, we invite you to look through some of these or reach out to our business department. unique sponsorship options Perhaps creating your very own tickles your fancy or maybe you have a whole new idea for how we can partner! writing contest Thank you to all of our sponsors! We truly appreciate your contribution to the tech world and thus the future! The Startups Awards are sponsored by: , , , , , , , , , , Blockster Brex FreeCodeCamp LaunchDarkly Coil Udacity Udemy Vercel Asia Pacific Transgender Network Bybit Everscale Network The post was written by Kushal Tamrakar and Valerie Ejegi, edited by Hang Ngo. If you have any questions or concerns or would like to report a discrepancy please email ellen@hackernoon.com.

Meet the Winners of Startups in Asia

Identifying patterns and extracting features on images using deep learning models SPCA takes in 7,000 to 9,000 animals each year in Singapore. Half of them are abandoned pets like cats, dogs, rabbits, and guinea pigs. It takes time for the pet to be identified and hence prolongs the wait to be listed for adoption. NParks has a group of volunteers who meet regularly for bird watching activities. At the same time they help collect data on the avian population in Singapore, but not all of them can identify the birds species correctly. My project goal is to develop a tool for these 2 organisations: to identify animal breeds for SPCA; and to identify avian species for NParks. This goal can be translated into an image classification problem for deep learning models. So I explored a simple neural network, and then progressed to convolutional neural network and transfer learning. Simple Neural Network First I started with image classification using a simple neural network. The dataset is from , which has 3 classes: cat, dog, and panda. There are 3000 images in total, ie, 1000 for each class. pyimagesearch I set up a simple neural network model with only 1 dense layer in the middle and took about 4 minutes to train the model. The accuracy achieved was 61% and I was ready to test the model with new images. I input these images to the model, and the simple neural network model was able to make to classify them according to the highest probability. For example, the model was 58% sure that this is a panda. But it has legs, so there is a small chance it could be a cat or a dog as well. I started to be adventurous, however the simple neural network model was unable to make the correct classification. So I trained a model which was able to classify them correctly. convolutional neural network (CNN) I had to explore further with more challenging images, and the CNN model is well known to be good at image classification. Convolutional Neural Network What is a CNN model then? CNN stands for , where each image goes through a series of convolution and max pooling for features extraction. I explored using the dataset which has 60,000 images divided into 10 classes. Convolutional Neural Network CIFAR-10 With so many images, it took almost 4 hours to train the model, and achieved an accuracy of 75%. So I was ready to test the model, using unseen images from Google search. The CNN model was able make the correct prediction most of the time, for example the model was quite sure that this is an airplane, and this is a ship with 72% probability. And also this is a deer and this is a horse based on the highest predicted probability. Incidentally there is some chance that this horse could be a deer or a frog, because of certain features picked up by the model. To improve classification accuracy, I need more data. I need to train the model on a larger data set. Transfer Learning Next I explored a huge dataset of over a million images. However, this model would take a long time to train with my limited resources. So I did to avoid reinventing the wheel. I used the developed by , which has 1000 classes ranging from animals to things and food. Oxford has spent a lot of GPU processing power, time and resources to train this model. Transfer Learning VGG16 pre-trained model University of Oxford Let’s test the model by feeding these images which I have downloaded from Google search (so I know the answers). These are quite similar images, but the model was able to classify them according to their . Notice that the Hush Puppies dog (Basset) on the left has more distinct features, the model was also more certain in its classification. breed How about to identify these cats? Oxford has already trained the VGG16 model on many cat species, and the model has no problem classifying them. How about these birds? Birds seem to have very distinct features, and the model was able to identify their species with very high certainty. However, not all these birds are found in Singapore. My next step is to look for many images of common birds and animals found in Singapore to train the model, so as to append to the “knowledge database” of the model. This would help to improve the classification tool for these 2 organisations (SPCA and NParks). Conclusion Image classification can be done using neural network models. Identifying patterns and extracting features on images are what deep learning models can do, and they do it very well. “The model is as intelligent as you train it to be” codes for the above analysis are available on my , do feel free to refer to them here: Python GitHub https://github.com/JNYH/Project_Kojak Video presentation: Thank you for reading.

How to Classify Animal Images via a Convolutional Neural Network

founders are often “hailed as superhumans,” but the truth is they’re not. They’re everyday people like you and me. But the thing that makes them different is their fearlessness to take risks. Startup In saying that, there is a lot that can be learned from successful entrepreneurs that have made it. So, I decided to spend a good chunk of my weekend researching billion dollar tech companies to see if there were any unique patterns or key learnings that I could distill. Here are a few things that I discovered. Lemonade Insurance 2015 Founded: Fintech Industry: $2 billion Approximate valuation: Using artificial intelligence, a mobile app and other tech-centric methods, Lemonade founders Daniel Schreiber and Shai Wininger are turning the centuries-old business of property insurance into a . millennial-friendly consumer product Every homeowner and renter needs insurance, but many distrust insurance providers. The two concluded that insurance, with its market value of $4.6 trillion, was ripe for disruption. The differs from that of a typical insurance company – it keeps a flat 25% fee of a customer’s premium while setting aside the remaining 75% to pay claims and purchase insurance. Unpaid premiums go to a nonprofit of the users opted for the annual “give back” deal. This builds trust with millenials. Lemonade business model Unlike fintech competitors Hippo and Jetty, Lemonade doesn’t sell policies backed by established insurers. It took the risk to become a licensed carrier itself, retaining claim liability on its own balance sheet. This risk has paid off, allowing them to create a social-compact pitch that pays claims fast. Lemonade can’t profit from denying legit claims, and customers making bogus claims are cheating charity, not some greedy insurer. Lemonade puts users first, building its product with the in mind. One of the biggest pain points of traditional insurance companies is the fact that interactions can be intimidating and time-consuming, leaving consumers powerless. Not the case with Lemonade. They started by thinking about the users and built from there. user experience Entrepreneurs can sometimes make their lives harder than they need to be. Often falling victim to things like and paralysis by analysis. By focusing purely on the most essential things I.e. the problem you’re solving and your solution to that problem. shiny object syndrome Lemonade, as a name, is brilliant in its simplicity. Each step along the customer journey? Also simple. Their content? Simple to digest. Netflix 1997 Founded: Entertainment Industry: $141 billion Approximate valuation: For nearly twenty years Blockbuster sat at the top of the US video-rental industry. In the early 2000s a small disruptive company called Netflix offered to partner with them to help run their online brand in exchange for promoting Netflix in their stores. Reed Hastings (co-founder of Netflix) got laughed out of the room. Fast forward to 2020, Blockbuster is now merely a distant memory after filing for bankruptcy in 2010 while Netflix is currently worth $166B (Feb 2020). Netflix is one of the best examples of transforming your business to stay ahead of the curve. When Netflix first launched in 1997 it was purely a website-based movie rental service — this was groundbreaking at the time. It took them 10 years before they introduced their streaming services which changed the entertainment industry forever. Like many other innovative companies, Netflix is not driven by pressure to do things first, but by delivering for the customer. The opportunities to do this evolve with technology, making it a never-ending process of testing, implementation and testing again. good experiences Netflix also shows that it pays to take risks. It took a decade for broadband internet to catch up with the founder’s vision, and Netflix’s decision to switch to "original content" and divert budgets away from licensing film releases was another famously big bet. The political drama House of Cards, cost a reported to produce the first two seasons. This bet paid off, when the show went on to win multiple awards and attract millions of viewers every episode. $100 million Airbnb 2007 Founded: Travel Industry: $35 billion Approximate valuation: Long before persuaded strangers to sleep in one another's homes and became a $35 billion company, it was just an idea to earn a bit of extra money to make rent. After their first guests, the three founders realized they were onto something bigger. Airbnb They presented a final version in 2008 but investors weren’t convinced. Out of 15 angel investors, eight said no and seven ignored them completely. It was 2009 when they finally picked up $600,000 in seed funding and the rest is history. Airbnb shows how important it is to understand your users and update your product according to their requirements. The company realized early on that receiving payments was crucial for those who are giving out their space on rent, so they promptly added the advance payments feature. Marketplace models such as Airbnb have become common today, but back in 2007, this model was mostly unproven. To apply this to your own startup, you need to have an idea that not only challenges existing solutions but is also able to create value. lesson Airbnb used disruptive technology by creating a marketplace at a time when it was rare. It created ripples in the hotel industry by providing a that was different, endearing, and reasonably priced. Airbnb used innovative and new-user discounts to build loyalty for its product. service referral programs Their focus is on providing more and incremental value to their customers. If you can build a product that adds value while providing your users with an incentive to spread the word, you can create a win-win situation for your business as well as your users. Canva 2013 Founded: Internet software + services Industry: $4.7 billion Approximate valuation: For every second that you’re reading this article, Sydney-based startup Canva (fun fact: they were originally known as ) — have helped millions of people create beautiful designs and documents using its online graphic tools. Canvas Chef More than have been created since its 2013 launch, which makes you wonder why it took three years of rejections by more than 100 potential investors before it got started. These knockbacks worked in favor of the founders, allowing them to polish their offering before it hit the market. 100 million designs Canva’s purpose is to democratise graphic design – to create a tool that is imbued with such simplicity, it allows amateurs to do things that previously were only in the domain of experienced design professionals. “Simplicity” is at the core of Canva’s brand, and it’s an attribute that the company tries to weave through its entire customer experience, from an easy-to-use drag-and-drop interface, to its polished design templates – even its . privacy notice The strongest attribute Canva’s founders Melanie Perkins, Cliff Obrecht and Cameron Adams have is . Canva’s success didn’t happen overnight, with their journey starting all the way back in 2007. persistence It takes time to build a unicorn — with it almost being guaranteed you’ll experience rejection, failure and frustration along the way. A great example of the founder’s persistence is the fact that it spent three years from meeting their first investor to actually landing the first investment. Focus is another strong attribute. One of the sole reasons Canva has been a big success is that the founders focused in the early days on finding one highly influential person in their space and getting them to help with their startup. Avoid trying to meet hundreds of people that aren’t really related to your business or what you need. Concentrate on the few names that can have the most impact instead. SEEK Limited 1997 Founded: Employment Industry: $7 billion Approximate valuation: SEEK Limited (stylised as SEEK) and its subsidiary companies, known as the SEEK Group, is one of the largest . SEEK focuses on facilitating the matching between jobseekers and employment opportunities and helping hirers find candidates for advertised roles. job sites in Australia SEEK was founded by brothers Paul and Andrew Bassat and Matthew Rockman, essentially as an online version of print employment classifieds. Today SEEK makes a positive impact on a global scale, with exposure to , more than 51 million students and learners and a presence in 18 countries. 2.9 billion people While a recruitment giant and disruptive force, Seek went through “a scary period” when it found itself on the other side of the fence, with new disruptors coming in and threatening its place in the market. they were not “pushing or challenging ourselves”. Andrew Basset admits Seek started to re-focus, considering what makes a person choose one platform over another, given the way technology has evolved. Bassat and SEEK’s leadership team realized that if they didn’t innovate they would be in “a lot of trouble.” “The reality is, today, everyone is either disrupting or being disrupted,” he said. “We always need to be vigilant. There’s always someone coming to take our business from us.” Key takeaways It’s important to remember that although some of the ideas and technology leveraged by these companies had never been seen before and completely changed the game. The principles and values used by the founders have been around for a very long time. Here are 3 of the main takeaways: Canva and Netflix are great examples of two companies that took nearly a decade to really hit their stride. That’s over 120 months of riding the emotional rollercoaster that comes with building a company. Consistency: All of the companies I’ve covered in this article have benefitted from simplifying processes, tasks and experiences that were traditionally painful for their customers. Simplicity: It can be easy for founders to become distracted by vanity metrics and superficial product features. By focusing on your customers' problems and how you can solve them — your company is more likely to create real value for your customers. Be obsessed with your customers: These ideas might not be as exciting as the latest trends or tactics. But they are almost guaranteed to continue to be  essential building blocks for the billion dollar companies of the future. Summing up Establishing a unicorn startup is no easy task and it’s not meant to be. You need to become comfortable with the feeling of failure and remind yourself the journey of an entrepreneur is a marathon not a sprint. Once upon a time, unicorn startups were a rare breed. Today, not so much. Lastly, if you are looking to take the dive into startup land then: be curious, be problem focused, but above all, take action.

Stories, News & Mentions about BASSET

Basset’s Stories on HackerNoon

Basset's News & Mentions