Arpit N. Verma

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

Machine Learning Engineer - Founding Team - London £200k If you answer yes to 4/5 of these questions, then you want to check out this job (link in comments) 1️⃣ Do you have experience with quantization techniques (e.g., PEFT, DeepSpeed, TensorRT) to optimize ML models? 2️⃣  Have you ever managed or worked with a multi-node HPC cluster? 3️⃣ Are you familiar with Deep Learning using PyTorch Library? 4️⃣ Do you have experience with Kubernetes container orchestration? 5️⃣  Have you worked with some of the recent Large Language Models (e.g., OpenAI, LLaMA)? #LLM #ArtificialIntelligence #Quantization

21

3 Comments

Finding better entity embeddings, and capturing the most suited representation of an entity is an area of cutting-edge research. While working on improving the entity embeddings, here are the top areas/ideas one can think upon. Using a tremendous amount of data for training Using more sophisticated neural network architectures. Utilising the side information of the entity along with the context. A better Negative Sampling approach. Positive sample data is available but how does one create negative samples? And if negatives are too many how to better sample them? Utilising the available information on the sequence in which the entities interacted or were interacted with. And training sequence to sequence models around it. Here's a detailed video for Data Science enthusiasts I have created on the topic covering all the theory and a practical exercise training billions of rows to generate high-quality embeddings. Like and subscribe for more such interesting concepts. Also, like and share over here for maximum reach. : ) Video Link: youtu.be/DN4S96oHRhE YT channel Link:  youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

18

Finding better entity embeddings, and capturing the most suited representation of an entity is an area of cutting-edge research. While working on improving the entity embeddings, here are the top areas/ideas one can think upon. Using a tremendous amount of data for training Using more sophisticated neural network architectures. Utilising the side information of the entity along with the context. A better Negative Sampling approach. Positive sample data is available but how does one create negative samples? And if negatives are too many how to better sample them? Utilising the available information on the sequence in which the entities interacted or were interacted with. And training sequence to sequence models around it. Here's a detailed video for Data Science enthusiasts I have created on the topic covering all the theory and a practical exercise training billions of rows to generate high-quality embeddings. Like and subscribe for more such interesting concepts. Also, like and share over here for maximum reach. : ) Video Link: youtu.be/DN4S96oHRhE YT channel Link:  youtube.com/@datatrek #datatrek #datascience #machinelearning #statistics #deeplearning #ai

9

My research into cell-based architecture led me to look into the latest trends in #SoftwareArchitecture So here they are (some of these are already quite common); 1. Data-driven architecture 2. Edge computing 3. Platform architecture 4. Large language models 5. Socio-technical architecture 6. Green software 7. Privacy engineering Fascinating reading up about these and will definitely be weaving them into conversations to add some meat to the bones. #everydayisaschoolday

16

3 Comments

The best tutorial blog on diffusion models Has intuitive explanation with proper maths I’ve followed Lilian’s blogs since I was in college and her explanation style is best suited for anyone who want to dig deep into machine learning

46

1 Comment

The best tutorial blog on diffusion models Has intuitive explanation with proper maths I’ve followed Lilian’s blogs since I was in college and her explanation style is best suited for anyone who want to dig deep into machine learning

31

The best tutorial blog on diffusion models Has intuitive explanation with proper maths I’ve followed Lilian’s blogs since I was in college and her explanation style is best suited for anyone who want to dig deep into machine learning

20

Why Join a Startup as a Machine Learning or AI Engineer? Are you a Machine Learning or AI Engineer looking for your next opportunity? Joining a startup might be just what you need. Here’s why: 💡 Innovation at its Core Startups thrive on innovation. You’ll get the chance to work on cutting-edge projects and experiment with the latest AI technologies. The fast-paced environment keeps you sharp, learning, and pushing boundaries every day. 🚀 Accelerated Growth In a startup, your work directly impacts the product and the company’s trajectory. You’ll quickly gain experience across the board—from building models to scaling systems—accelerating your professional development far beyond what you’d see in a larger corporation. 👥 Be a Key Player At a startup, every team member is crucial. Your contributions won’t just be a small part of the puzzle—they'll be front and center, shaping the future of the company. 🔍 Ownership & Autonomy Startups offer the opportunity to take ownership of projects, make key decisions, and lead initiatives. It’s a chance to build something from the ground up, with fewer barriers to creativity. 💼 Diverse Skill Development From data science to software engineering, and even leadership—working at a startup offers you a chance to develop a wide array of skills. You’ll be exposed to multiple facets of business, tech, and product development. If you’re ready to take your AI and Machine Learning expertise to the next level, consider joining a startup where the future is being built every day! 🌐🚀here at Generative we got you covered! #AI #MachineLearning #Startups #TechCareers #Innovation #CareerGrowth

5

2 Comments

Machine Learning Engineer - Founding Team - London £200k If you answer yes to 4/5 of these questions, then you want to check out this job (link in comments) 1️⃣ Do you have experience with quantization techniques (e.g., PEFT, DeepSpeed, TensorRT) to optimize ML models? 2️⃣  Have you ever managed or worked with a multi-node HPC cluster? 3️⃣ Are you familiar with Deep Learning using PyTorch Library? 4️⃣ Do you have experience with Kubernetes container orchestration? 5️⃣  Have you worked with some of the recent Large Language Models (e.g., OpenAI, LLaMA)? #LLM #ArtificialIntelligence #Quantization

3

2 Comments

📚 Just finished "Programming PyTorch for Deep Learning" and wow, what a game-changer ((: If you're diving into the world of AI, this book is your new best friend ((= . Here's why I'm geeking out over it, like others : 1️⃣ Hands-on Approach: It's like having a coding buddy right next to you! The book walks you through real-world projects, making complex concepts click. 2️⃣ PyTorch Mastery: From basics to advanced tricks, it covers everything. I particularly loved the section on custom datasets - super useful! 3️⃣ AI Application Bonanza: It doesn't just teach PyTorch; it shows you how to build cool stuff! Think image classifiers, text generators, and even a mini AI artist. 🎨🍕(food is art too, don't judge me) 4️⃣ Performance Hacks: The optimization tips are gold. My models are running so much faster now! 5️⃣ Future-Ready Content: It even touches on cutting-edge topics like transformers and GANs. Hello, next-gen AI! 👋 The author's style is refreshingly clear - no more head-scratching over dense jargon. Plus, the code examples? *Chef's kiss* 👨‍🍳💋 Whether you're a PyTorch newbie or looking to level up, this book is a must-have. It's already earned a permanent spot on my desk! Have you read it? What's your go-to resource for deep learning? Can you tell me more about other resources like this? Let's chat in the comments! 💬 I love new perspective {: #DeepLearning #PyTorch #AIEngineering #MachineLearning #ProgrammingBooks #TechLearning #ArtificialIntelligence #CodingLife #DataScience #AIcommunity

2

1 Comment

What if your OpenAI API calls were 100% reliable? This is the claim from the Structured Outputs announcement yesterday. Here is the excerpt: "On our evals of complex JSON schema following, our new model gpt-4o-2024-08-06 with Structured Outputs scores a perfect 100%. In comparison, gpt-4-0613 scores less than 40%." I have been using libraries like instructor to get structured output, but this might mean that all you need is the API now. What do you think? Link: https://buff.ly/46SD2Cd

2

Glad to see that NGS-IQ's Excel Support has finally caught up with what I've been doing with Excel support in my now free SQL iQuery for the last 10 years. If you're an NGS-IQ user, check out their new features in this itJungle article. https://lnkd.in/geKRGm3Z

9

1 Comment

The average machine learning engineer salary in the USA is $120K per year or $76.44 per hour. Best courses to become a Machine Learning Engineer 👇👇 1. Machine Learning (Coursera - Stanford University) Why it's great: This course is considered the gold standard for introductory ML. It provides a strong foundation in core concepts, algorithms, and best practices. Key topics: Supervised learning, unsupervised learning, practical advice for applying ML, best practices. Tools: Octave/MATLAB (though you can adapt the code to Python). 🔗 https://lnkd.in/gNXTg8aP 2. Deep Learning Specialization Why it's great: This specialization dives deep into neural networks, covering CNNs, RNNs, LSTMs, and more. It's essential for anyone interested in deep learning. Key topics: Neural networks, convolutional networks, recurrent networks, deep learning best practices. Tools: Python, TensorFlow. 🔗 https://lnkd.in/gSUchfqm 3. Machine Learning Engineering for Production (MLOps) Specialization Why it's great: This specialization focuses on the practical aspects of deploying and maintaining ML models in production environments. Key topics: MLOps fundamentals, building ML pipelines, model deployment, monitoring, and maintenance. Tools: Cloud platforms (AWS, GCP, Azure), CI/CD tools. 🔗 https://lnkd.in/dP3285Ry 4. Natural Language Processing Specialization Why it's great: This specialization provides a comprehensive introduction to NLP, covering a wide range of techniques and applications. Key topics: Word embeddings, sequence models, transformers, sentiment analysis, machine translation. Tools: Python, TensorFlow, Trax. 🔗 https://lnkd.in/g4D6RmVX 5. Introduction to Computer Vision (Udacity - Georgia Tech) Why it's great: This course provides a solid foundation in computer vision concepts and techniques. Key topics: Image formation, feature extraction, object recognition, motion estimation. Tools: Python, OpenCV. 6. Data Analysis with Pandas and Python (Udemy - Boris Paskhaver) Why it's great: This hands-on course teaches you how to use Pandas effectively for data analysis and manipulation. Key topics: Data wrangling, cleaning, aggregation, visualization, time series analysis. Tools: Python, Pandas, Jupyter Notebook. . . . Follow for more. 🙏 Join OpenAILearning: [https://lnkd.in/gaaR8jxM] . . .

135

22 Comments

PyTorch 2.X Compiling Modes: 1️⃣ Compiling a model means changing the execution mode of forward and backward phases from eager to graph mode. In doing this, the machine learning framework evaluates all operations and operands involved in these phases in advance to compile them into a single piece of code. Therefore, notice that when we use the term “compiling a model,” we are referring to compiling the processing flow that’s executed in the forward and backward phases. 2️⃣ But why would we want to do that? We compile a model to accelerate its training time since the compiled code tends to run faster than the code that’s executed in eager mode. As we will see in the next few sections, performance improvement depends on diverse factors, such as the computational capability of the GPU used to train the model. 📕 From: Accelerate Model Training with PyTorch 2.X by Maicon Melo Alves 📻 Stay tuned for the book review!

18

3 Comments

Are you struggling to minimise the number of classification errors in your machine learning model? 💻 Optimise Your Classification Model with these Loss Functions! Let's break it down: 0-1 Loss: A non-convex function that aims to minimise errors, but is hard to optimize. Convex Approximation: A smoother alternative that's easier to minimise, but still has degenerate solutions. Hinge Loss: A convex function that prevents degenerate solutions, but is not differentiable. Logistic Loss: The smooth, differentiable solution that combines the best of both worlds. It's a game-changer for optimising your classification model! Note however, this may not be the best optimiser and may not give the widest margin between the opposing classes even in binary classification so be careful and be sure to try different available options. Dive deeper into the world of loss functions and learn how to optimise your model for better performance. Share your thoughts and experiences in the comments! #machinelearning #lossfunctions #classification

5

If you know how LLMs work but want to take your technical understanding to the next level, this is the best place to start: Ben Bycroft’s site LLM Guide is one of the best free resources I’ve come across for anyone working with large language models. It’s packed with: • Easy-to-digest explanations of complex LLM topics • Detailed insights into fine-tuning and optimization • Practical resources for hands-on experimentation And the best part: interactive visualizations that activate as you learn each concept—making complex ideas easier to understand and remember. Whether you’re building AI products, exploring LLMs for the first time, or just curious about how they work, this guide is a goldmine. Check it out here: https://bbycroft.net/llm What’s your favorite resource for learning about LLMs?

7

1 Comment