Connor Shorten

This video explores Teacher-Student Curriculum Learning! Interestingly, the Teacher is rewarded for the Student's progress on Sub-Tasks rather than solely optimizing for progress on the hardest task, Intrinsic Motivation! https://youtu.be/GFCujBpTf3k  #100DaysOfMLCode

Maybe you will find this video I made useful, titled: "The Evolution of AlphaGo to MuZero"!https://www.youtube.com/watch?v=A0HX8BgckFI&t=4s …

This video explores "Learning to Execute" using Sequence-to-Sequence LSTMs to predict the output of a Python program without running the code! I was really interested in this paper because of their exploration into Curriculum Learning! https://youtu.be/5rrldwJdDRE  #100DaysOfMLCode

AI Weekly update for February 3rd, 2020! This update covers Google AI's Meena Chatbot, Microsoft's ImageBERT, blog posts on Curriculum Learning in RL and Contrastive Self-Supervised Learning, and more! https://youtu.be/N9jsScucOXs  #100DaysOfMLCode

I highly recommend checking out the lecture series from "Full Stack Deep Learning" on YouTube https://www.youtube.com/watch?v=5ygO8FxNB8c&t=920s … #100DaysOfMLCode

This video explains @GoogleAI 's amazing new Meena chatbot! An Evolved Transformer with 2.6B parameters on 341 GB / 40B words of conversation data to achieves remarkable chatbot performance! "Horses go to Hayvard!" https://youtu.be/STrrlLG15OY  #100DaysOfMLCode

This video covers the Evolved Transformer used in the mind-blowing Meena chatbot from @GoogleAI! This video explains the details of how they encode Transformers for automated architecture search! https://youtu.be/khA-fiC1Wa0  #100DaysOfMLCode

14th Edition of AI Weekly Update! Happy to get back to this after a 2 month break! Covering Point-Goal navigation in AI-Habitat from FAIR, FixMatch for Semi-Supervised Learning, OpenAI's study on scaling language models and more! https://youtu.be/s5wILiRE-6Q  #100DaysOfMLCode

This video explains FixMatch, a new Semi-Supervised Learning algorithm from @GoogleAI!! FixMatch enforces consistent predictions through Data Augmentation and Pseudo-Labeling to achieve remarkable success with limited labeled data! https://youtu.be/nkewn6XGyt8  #100DaysOfMLCode

This video explores the Reformer Efficient Transformer model! Locality-Sensitive Hashing matches inherent sparsity in attention to attend only to similar keys. Reversible layers reduce memory costs from storing intermediate activations! https://youtu.be/Kf3x3lqf9cQ  #100DaysOfMLCode

This video explains the evolution of AlphaGo to MuZero from @DeepMindAI! Going from policy nets trained with supervised learning on expert moves in Go, to an algorithm that does model-based planning in abstract space on 57 Atari games! https://youtu.be/A0HX8BgckFI  #100DaysOfMLCode

This video explains MuZero! MuZero uses hidden states to facilitate Model-Based RL through the self-play MCTS from the AlphaGo series. I also tried to provide an explanation of how Backprop through Time (BPTT) is used to train this! https://youtu.be/szbvm8aNDxw  #100DaysOfMLCode

This video explains AlphaZero from DeepMind in the series: AlphaGo to MuZero! AlphaZero doesn't dramatically change the previous AlphaGo Zero algorithm, but it does show how the algorithm can generalize to Chess and Shogi! https://youtu.be/4FdiTTZPkos  #100DaysOfMLCode

This video explores AlphaGo Zero! AGZero avoids human encoded knowledge: no handcrafted features for state rep. and no supervised learning on expert moves. AGZero uses a really interesting extension to the MCTS self-play for training! https://youtu.be/B1MUfP1qqLs  #100DaysOfMLCode

This video explains how AlphaGo works! This is the first in a series explaining the evolution of AlphaGo to MuZero! AlphaGo uses convolutional policy and value networks to enhance their tree search! https://youtu.be/jgAj8CqcBBs  #100DaysOfMLCode

This is my recap of Artificial Intelligence in 2019! This video covers new developments in understanding Neural Networks, self-supervised learning, language models, Generative Models, Game-playing RL, and many more! https://youtu.be/6SWpN64Ivb4  #100DaysOfMLCode

Upside-Down RL reformats the RL mapping function: taking states, desired rewards and horizon as input to output actions! This video explores the paper, how supervised learning is used to train this, as well as results in the OpenAI Gym! https://youtu.be/ed7QQMG24MM  #100DaysOfMLCode

Mixed Precision Training is 2 lines of code with the Keras API and TF 2.1! This video explores the docs for this as well as details about Mixed Precision Training such as which layers to keep in FP32 and the use of loss scaling! https://youtu.be/pKZs4hllCvI  #100DaysOfMLCode

This video explores a really exciting new Meta-Learning algorithm learning to generate its own training data! These datasets can train classifiers to high real data accuracy extremely quickly, speeding up Neural Architecture Search! https://youtu.be/lmnJfLjDVrI  #100DaysOfMLCode