Argmax

We talk about Low Rank Approximation for fine tuning Transformers. We are also on YouTube now! Check out the video here: https://youtu.be/lLzHr0VFi3Y

In this episode we discuss the paper "Training language models to follow instructions with human feedback" by Ouyang et al (2022). We discuss the RLHF paradigm and how important RL is to tuning GPT.

This week we talk about Whisper. It is a weakly supervised speech recognition model.

We talk about AlphaTensor, and how researchers were able to find a new algorithm for matrix multiplication.

In this episode we talked about "Implicit Neural Representations with Periodic Activation Functions" and the strength of periodic non-linearities.

In this episode we discuss this video: https://youtu.be/jPCV4GKX9DwHow Tesla approaches collision detection with novel methods.

We discuss Sony AI's accomplishment of creating a novel AI agent that can beat professional racers in Gran Turismo. Some topics include:- The crafting of rewards to make the agent behave nicely- What is QR-SAC?- How to deal with "rare" experiences in the replay bufferLink to paper: https://www.nature.com/articles/s41586-021-04357-7

Today we talk about recent AI advances in Poker; specifically the use of counterfactual regret minimization to solve the game of 2-player Limit Texas Hold'em.

Today we talk about GATO, a multi-modal, multi-task, multi-embodiment generalist agent.

We start talking about diffusion models as a technique for generative deep learning.

We discuss NeurIPS outstanding paper award winning paper, talking about important topics surrounding metrics and reproducibility.

We talk about QMIX https://arxiv.org/abs/1803.11485 as an example of Deep Multi-agent RL.

Todays paper: Can Neural Nets Learn the Same Model Twice? Investigating Reproducibilityand Double Descent from the Decision Boundary Perspective (https://arxiv.org/pdf/2203.08124.pdf)Summary:A discussion of reproducibility and double descent through visualizations of decision boundaries.Highlights of the discussion:Relationship between model performance and reproducibilityWhich models are robust and reproducibleHow they calculate the various scores

Todays paper: VICReg (https://arxiv.org/abs/2105.04906)Summary of the paperVICReg prevents representation collapse using a mixture of variance, invariance and covariance when calculating the loss. It does not require negative samples and achieves great performance on downstream tasks.Highlights of discussionThe VICReg architecture (Figure 1)Sensitivity to hyperparameters (Table 7)Top 5 metric usefulness

Todays paper: data2vec (https://arxiv.org/abs/2202.03555)Summary of the paperA multimodal SSL algorithm that predicts latent representation of different types of input.Highlights of discussionWhat are the motivations of SSL and multimodalHow does the student teacher learning work?What are similarities and differences between ViT, BYOL, and Reinforcement Learning algorithms.

This is the first episode of Argmax! We talk about our motivations for doing a podcast, and what we hope listeners will get out of it.Todays paper: Reward is Enough Summary of the paperThe authors present the Reward is Enough hypothesis: Intelligence, and its associated abilities, can be understood as subserving the maximisation of reward by an agent acting in its environment.Highlights of discussionHigh level overview of Reinforcement LearningHow evolution can be encoded as a reward maximization problemWhat is the one reward signal we are trying to optimize?