Kukeshajanth Kodeswaran

📝🚗 Paper Highlight: Trajectory Prediction in Turnaround Scenarios Using the rounD Dataset! We're happy to share another great research project that leverages our #rounD dataset to improve trajectory prediction in autonomous vehicles! In their paper, researchers George Daoud, Mohamed El-darieby, and Khalid Elgazzar from Ontario Tech University and Assiut University proposed a deep neural network model based on Graph Neural Networks (GNN) to predict vehicle trajectories in complex turnaround scenarios. The #rounD dataset is a collection of three different roundabout locations in Germany using drones. It captures the complexity of real-world driving with almost 14,000 road users distributed across eight classes including pedestrians and bicyclists. This high-resolution dataset offers detailed information on vehicle behavior, allowing the model to learn diverse driving patterns. Key Findings: 🔹 The proposed GNN model parallely predicts the trajectory of surrounding vehicles collectively, avoiding over-smoothing and offering real-time context-aware predictions. 🔹 The model trained on the rounD dataset achieved a final displacement error of less than 1.7 meters for a 3-second horizon, outperforming traditional Bird’s Eye View (BEV) models. 🔹 The research highlights the importance of using context-rich datasets like rounD to understand complex traffic scenarios, especially in situations like entering, driving through, and exiting roundabouts. 📄 Find the link to the paper and the dataset in the comments below! #TrajectoryPrediction #AutonomousDriving #rounD #TrafficData #Innovation #GraphNeuralNetworks #leveLXData #Research #ScenarioData #ADAS

29

1 Comment

Build your machine learning portfolio with our Python for Machine Learning: The Essential Starter Kit course, starting September 9. Register before August 12 and receive a 10 per cent early bird discount with code PMLEB10. This code must be applied at the time of registration and cannot be combined with other discounts or promotions. Perfect for professionals looking to advance into roles that include machine learning, this course is designed to introduce you to Python, a programming language known for its simplicity and readability. You will understand how Python can be used for machine learning tasks and learn to use its popular packages and libraries for data preparation. Learn more and register: https://lnkd.in/gz9wS_fp Mehrdad Pirnia Fuat Can Beylunioğlu Manda (Hongxiu) Li, Ph.D. #Python #Machinelearning

13

DVA is not associated with this job posting DevOps Engineer - ML Ops United States, Europe, Canada, Brazil https://lnkd.in/gEdnfdrC The team Kraken is looking for an experienced Machine Learning Ops engineer to join our AI/ML Team in the centralized Data organization. In this role you will be building infrastructure to support building cutting edge AI/ML technology to solve the most complex and exciting problems in the quickly growing and evolving crypto industry. We are looking for an extremely strong communicator and team-player, who is able to break down large complex problems into smaller more manageable problems-to-solve. You will take initiative to work with engineers across the team and org, exploring different ways to resolve issues. The opportunity Build ML and AI Ops infrastructure to enable the development and deployment of production models running at scale. This includes deployments across multiple cloud infrastructures Lead resource planning and optimization, especially with GPU instances Develop tool(s) to load test various production AI/ML systems Work closely with SREs across the entire organization Support 24/7/365 uptime of services by supporting a partial on-call rotation along with other members of the team Stay up-to-date in machine learning, and artificial intelligence trends and technologies, all while contributing to the growth of AI/ML in the Crypto industry #innovation #management #digitalmarketing #technology #creativity #futurism #startups #marketing #socialmedia #socialnetworking #motivation #personaldevelopment #jobinterviews #sustainability #personalbranding #education #productivity #travel #sales #socialentrepreneurship #fundraising #law #strategy #culture #fashion #business #networking #hiring #health #inspiration

3

How are Graph Convolutional Neural Networks (GCNNs) a generalisation of Convolutional Neural Networks (CNNs)? My next blog in the Graph series on "Deep Learning with Graphs" is on the way (well, it's already written on paper) and it's going to be a blockbuster. The beauty of Graph Machine Learning is that the more you read about it, the more you feel that your neurons (real ones!) are learning to think better! And you're already in love with it! Coming to the point: One might recall the convolution operations in a CNN. You have a grid of pixels representing an image. You have a 3 x 3 filter. You slide the filter over the image and carry out aggregation to obtain a new pixel value. You proceed in the same fashion over the whole grid, sliding towards the right , bottom, carrying out the convolution operation and finally getting a new transformed, reduced image through the series of aggregations. You repeat this through the different convolutional layers which together form a Convolutional Neural Network Now think of a Graph. It is not simple grid like an image. The topology of a Graph is arbitrary. There is no notion of left, bottom, right in a Graph like in an image which is represented through a grid of pixels. You cannot slide over a Graph as simplistically you did and performed the convolutional operations in an image. How do you approach the problem? How do you carry out convolutional operations in a Graph? Now take a node in the Graph. Take all the nodes which are the immediate neighbours of this node. The convolutional operation here is: You aggregate the information from all the neighbours and pass it to the node. The information is nothing but the emedfings of the neighbour nodes. You add together the emedding of the node under consideration. You transform the aggregated embedding by multiplying with a weight Matrix and transform the node's own embedding by multiplying with a different weight matrix. This is the typical linear transformation operation in a NN. You apply non linearity - the activation function like ReLu. Now think about a CNN taking the same grid and considering a central pixel. Consider the rest of the pixels around the central pixel. You are essentially doing the same thing: taking information from the neighbour pixels along with the central pixel, aggregating, as I described above for the node in the Graph. Hence a Graph Convolutional Network is a generalisation of CNN The story goes, you do the same for every node on the Graph. Each node in the Graph represented through a different neural network architecture! This is amazing 🔥🔥🔥💓💓💓 #graphs #machinelearning #deeplearning My Technical Blogs Library: 1) https://lnkd.in/ejyWacXn 2) https://lnkd.in/bfrkRjq Follow Ajay Taneja for insights on: Data Science, Data Engineering, Finite Element Analysis, Non Linear Finite Element Analysis, Fatigue, Fracture Mechanics

22

Learn how you can convert a PyTorch model, trained on the web, to a mobile-optimized format. Madhur Zanwar describes the process using a few lines of code. He also discusses his journey and how one approach didn't yield satisfactory results while trying to run it in a React Native app. #machinelearning #pytorch #computervision #edgeml #reactnative #objectdetection #yolov5

14

Where did the regularization term in loss function originate from? Experimentally, it’s pretty easy to verify the effectiveness of Regularization. As the regularization parameter increases, the model tends to create simpler decision boundaries. But very few know the origin of regularization. More specifically, Regularization comes from a slight modification to the maximum likelihood estimation (MLE) procedure, resulting in maximum a posteriori estimation (MAP). The difference is that: - MLE always attempts to find the parameters that are most likely to generate the data. - MAP also considers an initial belief about the parameters themselves. As a result, the solution provided by MAP is typically more acceptable. If it isn’t clear, let me give you a bit of a hypothetical but intuitive explanation of this. Imagine you are finding the weather extremely hot this year and intolerable compared to previous years. You want to figure out the reason so you come up with the following three possibilities: - Personal sensitivity to heat. - The earth has moved closer to the sun relative to how far it was last year. - Climate change, wildfires in nearby states, etc. If you think about it, explanation #2 is guaranteed to result in extremely hot weather, right? So, following a maximum likelihood estimation (MLE) approach would result in explanation #2 being the most likely reason. In other words, in MLE, you just accept the explanation that will most likely generate the observation (or data); that’s it. But there’s something that tells you that explanation #2 is not a perfect fit, right? This is because such cosmic events happen over lakhs/millions of years. Thus, the explanation itself (the earth moving closer to the sun) is highly unlikely to occur in the first place. You took a maximum a posteriori estimation (MAP) by doing this. Simply put, in addition to the most likely explanation, MAP also considers an initial belief about the explanation itself. The “explanation,” in the machine learning modeling case, is the parameters. A similar idea is utilized in regularization: > Don’t just look at which parameters most likely generated the data, but also consider how likely the parameters are to appear themselves. I covered this in much more mathematical detail here: https://lnkd.in/gAEQs7JB Overall, it turns out that there is a concrete probabilistic justification for using regularization and it just does not come from thin air.

134

2 Comments

Are you grappling with the complexities of deploying Large Language Models? TitanML's CEO, Meryem Arik, shared 7 tips to streamline your AI infrastructure at the Toronto Machine Learning Society (TMLS) Summit. 🎥 Watch the full talk: https://lnkd.in/ece3KYGm 📝 Deep dive into the details: https://lnkd.in/etfQn4Gu Key takeaways: * Why self-hosting LLMs matters * Quantization strategies for optimal performance * Inference optimization techniques * Infrastructure consolidation best practices * Future-proofing your LLM deployments Don't miss out on this opportunity to level up your LLM deployment skills. Watch, read, and transform your approach today! #AIInfrastructure #LLM #MachineLearning #TechInnovation #MLOps

16

1 Comment

Unlock the full potential of your Large Language Models (LLMs) with Intel® Extension for PyTorch (IPEX) and the Intel® LLM Library for PyTorch (IPEX-LLM). Download this whitepaper to explore the optimization of LLMs, including their performance, resource utilization, and response times in real-world applications. Link - https://intel.ly/3BBJ4ey

1

Myths about Large Language Models (LLMs) like #Chatgpt #Gemini #llama and others - LLMs can understand like humans: LLMs work on pattern matching using algorithms. LLMs cannot understand the questions as humans do and so the response might not be always 100% as expected. - Accuracy of LLMs:While models try for accuracy, the information given might be incorrect or misleading. Model does not verify facts of the response - More data will improve response of LLMs:While more data can improve quality to some extent, fine tuning, transfer learning, tunning some parameters, modifying architecture of data flow, etc. can improve performance of LLMs. - LLMs can be creative like humans: LLMs can generate text, images or videos. While LLMs can augment human creativity but it cannot replace complexity of human creativity. - LLMs are invulnerable to errors or manipulation: LLMs can make mistakes and can be manipulated with adverse inputs to generate misleading information. - LLMs will surpass human intelligence: Artificial general intelligence (AGI) is still far off from reality and hence LLMs are far from achieving human-like intelligence. * views expressed are personal

42

3 Comments

AEs (Variational Autoencoders) and AEs (Autoencoders) are neural network architectures used for dimensionality reduction and data compression. They share a similar structure with an encoder and a decoder, but they differ in how they handle the latent space, which is a compressed representation of the input data. Here is a good article explaining the difference between AEs and VAEs. #AE #VAE #neuralnetworks #deeplearning #computervision I like to share my Data Science journey with my LinkedIn friends… I believe learning everyday is part of life. Follow me if you like my posts. keep reading. keep learning.

9

This is an invaluable resource! The ability to benchmark the speed and cost efficiency of different LLM API providers is crucial for developers optimizing real-time applications. By focusing on factors like quality versus price, comparison of abilities, and balancing quality with output speed... this tool addresses essential aspects that impact decision-making.

5

Heterogeneous Graphs and Relational Graph Convolutional Networks (RGCNs): How can we prevent overfitting? In my post of last week [https://lnkd.in/d5cZn5ur], I spoke about Heterogeneous Graphs - examples of Heterogeneous Graphs being Knowledge Graphs which might have several node types or relationship types. I had mentioned that training such complex networks can become computationally intensive because each relationship type has to be associated with different weight/parameter thus increasing the number of  overall parameters to be learnt for a RGCN model and thus resulting in overfitting of the model. I had then highlighted about "Block Diagonal Matrices" And "Basis Learning" to reduce the number of parameters of a RGCN model. Let us try and put some more insight into "Block Diagonal and Basis Learning" and understand what these actually mean! Block Diagonal Matrices: The key idea with "Block Diagonal Matrices" is to make the Transformation matrices corresponding Wr as sparse as possible- that is: to enforce a block diagonal structure as shown in the Figure 1 below so that the non-zero elements lie only on the specific blocks of the transformation matrix "Wr". This reduces the number of non-zero elements which means that during computation you will have to only estimate the non-zero elements. This reduces the training time considerably What do you lose by using Block Diagonal Matrices? Well, we do lose some information by using Block Diagonal Matrices - the embedding dimension far from one another will not be able to interact with each other It’s only the embedding dimension that are along the non-zero entities that will be able to interact with each other that is: only the neurons along the non-zero entities will be able to interact with each other Basis Learning: The second idea to make the RCNs more scalable is to share the weights amongst the relationship types. That is: we do not want the relationships to be independent of each other, but we want the relations to share the information based on some magnitude of weighting. This can be achieved by elegantly representing the Transformation matrices Wr as a linear combination of "Basis" matrices. These "Basis" matrices are also termed "Dictionaries" In this approach, the weight matrix Wr will be represented as a linear combination i.e. a weighted sum of Dictionary matrices Vb as highlighted in the Equation below In the equation, arb is the "importance" weight corresponding to the relation for the matrix "b". Thus, every relation will have to learn its importance weight for each of the basis matrices. This significantly reduces the number of parameters - as the basis matrices may be 10 or a larger number whereas number of relation types may be thousands and arb are scalar numbers will have to be learnt #graphs #machinelearning #deeplearning My Technical Blogs Library 1) https://lnkd.in/ejyWacXn 2) https://lnkd.in/bfrkRjq Follow Ajay Taneja for insights on Engineering

6

How to Build a Custom Optimizer in PyTorch: 5 Simple Steps via #TowardsAI → https://bit.ly/3Y2UXTB

1

How to Build a Custom Optimizer in PyTorch: 5 Simple Steps via #TowardsAI → https://bit.ly/3Y2UXTB

2