Mahadevan Venkataraman’s Post

We all love Andrew Ng. We love his posts. 👏 But have you read his 167 page Ph.D. Dissertation from the Graduate Division of the University of California, Berkeley entitled "Shaping and policy search in Reinforcement Learning"? I was reading "The Alignment Problem - Machine Learning and Human Values" by Brian Christian. On page 168, I read this: ✴ Andrew Ng was ambitious. He would do his PhD thesis on using reinforcement learning to fly an actual un-simulated, nine-foot-long, one-hundred-pound, seventy-thousand-dollar-Yamaha R-50 helicopter. ✅ Page 126 of Dissertation - be amazed by "Model selection and incorporating prior knowledge" ✅ Page 133 of Dissertation - get ready for "Learning to Hover" Armed with the simulator for the helicopter, Andrew proceeded to use reinforcement learning to make a helicopter hover stably. He was looking for the hardest control problem he could find at that time. I'm already impressed with Andrew Ng. But now I'm speechless. 🚁

Interesting watch! Really unpacks attention, working memory & long-term memory as key players in learning, successful knowledge transfer, application and long-term retention 🧠 While "Learning = a change in LTM" might be a very familiar mantra, this really showcases just how important this is! #ScienceOfLearning #TLA #FE #Memory #Psychology

In the Science of Learning Programme, we explore what we know about learning in the brain 🧠 Here, Prof Kime talks with Hayley Vlach about the role of attention, working memory and long-term memory in learning.

In the Science of Learning Programme, we explore what we know about learning in the brain 🧠 Here, Prof Kime talks with Hayley Vlach about the role of attention, working memory and long-term memory in learning.

Visualizing 65,000 ArXiv Papers: A Deep Dive into Research Similarities 📊 Using NeuralWork's recently released dataset of ArXiv papers from January-October 2023, I created this visualization to map the relationship between research papers across different disciplines. The plot represents paper similarities using a custom-trained version of e5_large_v2, with each point representing a paper and colors denoting research categories. What's fascinating is how certain fields cluster together, while others show interesting overlaps in their research approaches. I'm especially intrigued by economics which seem to cluster at opposite ends of the dimensions showing they probably cross over with seperate disciplines at some point! The x and y axes measure research similarity across two dimensions, revealing natural groupings and interdisciplinary connections between Computer Science, Mathematics, Physics, Economics, and other fields. Interactive version: https://lnkd.in/eJG_u-wJ #MachineLearning #ResearchVisualization #DataScience link to dataset: https://lnkd.in/ecy8rKmN

Back in 1999 it was already totally clear that we could expect increased and extended use of all types of parallel computing technologies to solve compute-intensive problems, especially in emerging areas such as artificial intelligence, financial modelling, simulations etc. I wrote this academic study for my dissertation (graduation thesis) and I’m happy to see that it is still available inside the book “Parallel Computing - Fundamentals and Applications”. https://lnkd.in/dpbwHh8S

In the Science of Learning Programme, we explore what we know about learning in the brain 🧠 Here, Prof Kime talks with Hayley Vlach about the role of attention, working memory and long-term memory in learning.

In Education we look to simple solutions to complex problems, while these seems coherent the reality is much more sophisticated. By simplifying the cognitive process down to the easy to understand analogy of a computer’s ROM and RAM are we really doing justice to the complex systems involved in teaching and learning?

🎓 Abdullah Mana M Al Farwan is defending his Doctor of Philosophy dissertation titled "An Experimental Study of Supervised Machine Learning Techniques for Minor Class Prediction Utilizing Kernel Density Estimation: Factors Impacting Model Performance." 📅 Wednesday, April 24, 2024, from 1 to 3 p.m. at 3120 Sangren Hall 🤖 Dive into the world of data science and discover how kernel density estimation shapes predictive models. 🌟 Come and let's support Abdullah as he shares his contributions to educational leadership and technology research. #PhDDefense #EMR # Measurement #MachineLearning #DataScience #EducationalResearch 📊🔬📚

🎓Thrilled to announce the submission of my thesis at the Computer Science's Computer Vision Lab, Reichman University! 🎉📚 Thesis Title: Advancing Fine-Grained Classification by Structure and Subject-Preserving Augmentation. My research explored the application of advanced generative models like Stable Diffusion and ControlNet to enhance fine-grained classification datasets. The developed method, SaSPA, generates realistic and class-consistent image augmentations, significantly boosting classification accuracy. Grateful for the support and excited for what’s next! 🚀 Checkout a TL;DR version of the paper (in the comments)