Paper presentations and projects: schedule & sign up sheet
Deadlines:
Fri, March 22 - project proposals due
Mon, May 6 - final project submission due
Lecturer: Irina Rish
Topic: The Large-scale AI Revolution: Breakthroughs and Challenges (slides)
Papers: The Bitter Lesson, GPT-3 paper: Language Models are Few-Shot Learners
Topic: The Large-scale AI Revolution: Breakthroughs and Challenges - continued (slides, video)
Papers: Training compute-optimal large language models (summary/blog: New Scaling Laws for LLMs), Broken Neural Scaling Laws
CERC-AAI Lab projects on Foundation Models - please join us!
Part 1: Lecturer: Irina Rish
Topic: Introduction to Continual Learning (slides)
Papers: Continual T0, Effect of scale on catastrophic forgetting in neural networks (summary: Effects of Model and Prior Learning Scale on Catastrophic Forgetting), Foundational Models for Continual Learning: An Empirical Study of Latent Replay
Part 2: Lecturer: Irina Rish
Topic: Continual Learning at Scale
Papers: Continual Pre-Training of Large Language Models: How to (re)warm your model.
Part 1: Lecturers: Mohammad Reza Samsami and Arjun Ashok
Topic: State-space Models for Time-series: Call for Collaborations (slides, video)
Papers: Efficiently Modeling Long Sequences with Structured State Spaces
HiPPO: Recurrent Memory with Optimal Polynomial Projections
Combining Recurrent, Convolutional, and Continuous-time Models with Linear State-Space Layers
Part 2: Lecturer: Irina Rish (video)
Announcements: Neural Scaling Laws workshops , reading group on Scaling Laws & Emergent Behaviors, Paper presentations and projects: schedule & sign up sheet
Brief discussion on AI Alignment: AI and the paperclip problem, AI alignment research links, Unsolved Problems in ML Safety, Concrete Problems in AI Safety, Alignment Workshop, Introduction to AI Safety, Ethics and Society.
Brief discussion on objective ethincs: Derek Parfit , On What Matters (vol 1, vol 2, vol 3). See also Why Anything? Why This? Reasons and Persons
Lecturer: Irina Rish
Topic: Phase Transitions in AI &Emergent Behaviors in Large-Scale models (slides, video)
Tutorial on Phase Transitions (by Guillaume Dumas, at the 2nd Workshop on Neural Scaling Laws)
Talk by Pascal Jr. Tikeng Notsawo (University of Montreal/Mila): Is grokking predictable? video Predicting Grokking Long Before it Happens: A look into the loss landscape of models which grok
The Law of Robustness by Sebastien Bubeck
Papers: Hard and Easy Distributions of SAT Problems, Every Monotone Graph Property Has a Sharp Threshold, Approximability of probability distributions, Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets, GPT-3 paper: Language Models are Few-Shot Learners, A universal law of robustness via isoperimetry
Topic: Albert Gu's talk about his paper on S4 (video)
Papers: Efficiently Modeling Long Sequences with Structured State Spaces, other SSM papers
Part 1: Lecturers: Benjamin Therien and Avery Ryoo
Topic : Scaling Laws for Neural Language Models (slides, video)
Part 2: Training Compute-Optimal Large Language Models (Chinchilla Explained: video)
Part 1: Lecturer: Romain Roy
Topic: Technical Report: Large Language Models can Strategically Deceive their Users when Put Under Pressure (slides, video)
Part 2: Lecturer: Irina Rish
Topic: Computational Psychology & Foundation Models papers and ComPsy FoMo Workshop (video)
Part 1: Lecturers: Paria Mehrbod and Vaibhav Singh
Topic: Scaling Laws for Transfer (slides, video) Passcode: +U^C8Zfp
Part 2: Lecturer: Irina Rish
Topic: Computational Psychoilogy and Foundation Models intersection, and related class projects (video)
Papers: Huang et al (2023) Who is ChatGPT? Benchmarking LLMs' Psychological Portrayal Using PsychoBench
Gupta et al (2023) Investigating the Applicability of Self-Assessment Tests for Personality Measurement of Large Language Models.
Explore PsychoBench
Part 1: Lecturers: Thomas Jiralerspong and Dragos Secrieru
Topic: Are emergent abilities of LLMs a Mirage? (slides, video) Video passcode: ^43yC+Ay
Part 2: Lecturers: Maxime Petrenko and Parviz Haggi
Topic: HiPPO: Recurrent Memory with Optimal Polynomial Projections (slides, video)
Part 3: Lecturer: Alexis Roger
Topic: Robin Suite of Open-Source Foundation Models and Alignment: Status Update and Call for Collaborations (slides, video)
Part 1: Lecturers: Tanner Ducharme and Mahmood Hegazy
Topic: Chain of Thought Prompting Elicits Reasoning in Large Language Models (slides, video) Passcode: wpPHJ1H#
Part 2: Lecturer: Moetez Kdayem
Topic: Vision Mamba : Efficient Visual Representation Learning with Bidirectional State Space Model
VMamba: Visual State Space Model (slides, video-part1, video-part2)
Part 1: Lecturers: Arjun Ashok, Andrew Williams
Topic: ForecastPFN: Synthetically-Trained Zero-Shot Forecasting (slides, video) Passcode: .nO6422A
Part 2: Lecturer: Irina Rish
Topic: Computational Psychology & Psychological Computation (slides, video)
Project Ideas: a Discussion (slides, video)
Robin Projects: Call for Collaboration
Part 1: Lecturers: Istabrak Abbes, Megh Thakkar
Topic: Wide Neural Networks Forget Less Catastrophically (slides, video)
Part 2: Lecturer: Paloma Fernandez, Anthony Gosselin
Topic: Unsolved Problems in ML Safety (slides, video)
Part 1: Lecturers: Emiliano Penaloza and Sophie Wu
Topic: Direct Preference Optimization: Your Language Model is Secretly a Reward Model (slides, video)
Part 2: a brief discussion, then youtube talk by Natasha Jaques
Topic: RLHF: How to Learn from Human Feedback with Reinforcement Learning
Part 1: Lecturers: William Callaghan and Humza Wajid Hameed
Topic: Artificial Intelligence, Values, and Alignment (slides, video - 1st hr)
Part 2: Lecturers: Moksh Jain and Vedant Shah (slides, video - 2nd hr)
Topic: Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Part 1: Lecturers: Jonathan Siu Chi Lim and Mina Beiramy
Topic: A Generalist Agent (Gato) (slides, video)
Part 2: Lecturers: Xiaoyin Chen and Xnyu Yuan
Topic: Uncertainty of Thoughts: Uncertainty-Aware Planning Enhances Information Seeking in Large Language Models (slides, video)
Part 1: Lecturers: Shruti Joshi and Aniket Didolkar
Topic: Mamba: Linear-Time Sequence Modeling with Selective State Spaces (slides, video - part 1)
Part 2: Lecturers: Anirudh and Dhuruva Priyan
Topic: Optimal neural representation in neural systems at the edge of chaos (slides, video - part 2)
Scaling & Emergence Reading Group (Mon, March 11, 2pm -3pm)
Lecturer: Irina Rish
Topic: This class is combined with the Scaling & Emergent Phenomena Reading Group
Part 1: Lecturers: Mahsan Abdoli and Prince Immanuel
Topic: Constitutional AI: Harmlessness from AI Feedback (slides, video)
Part 1: Lecturers: Misha Barth and Sammy Sharief
Topic: GENERALIZATION IN DIFFUSION MODELS ARISES FROM GEOMETRY-ADAPTIVE HARMONIC REPRESENTATION ON MEMORIZATION IN DIFFUSION MODELS (slides, video)
Part 2: Lecturers: Yuchen Hui and Yicong Li (slides, video)
Topic: Mixtral of Experts
Part 1: Lecturers: Paolo Cudrano
Topic: Effect Of Model And Pretraining Scale On Catastrophic Forgetting In Neural Networks (slides, video)
Part 2: Lecturers: Mahsan Abdoli, Prince Immanuel
Topic: Constitutional AI: Harmlessness from AI Feedback (slides, video)
Part 1: Lecturers: Nicolas Bernier
Topic: Scaling Data-Constrained Language Models (slides, video)
Part 1: Lecturers: Jon Pilarte Bilbao, Karoline Lippert
Topic: The power of quantum neural networks (slides, video)
Part 2: Lecturers: Aloys Portafaix, Congshu Zou
Topic: UniT: Multimodal Multitask Learning with a Unified Transformer (slides, video)
Part 1: Lecturers: Aniket Saxena, Venkatesh Ramesh
Topic: Retrieval-Augmented Multimodal Language Modeling (slides, video)
Part 2: Lecturers: Jeremy Qin, Jinghan Sun
Topic: In deep reinforcement learning, a pruned network is a good network (slides, video - part 2)
Part 1: Lecturers: Darsh Kaushik, Xavier Morin
Topic: LoRA: Low-Rank Adaptation of Large Language Models (slides, video)
Part 2: Lecturers: Jerome Francis, Raj Ghughare
Topic: Beyond neural scaling laws: beating power law scaling via data pruning (slides, video)
Part 1: Lecturers:
Topic:
Part 2: Lecturers:
Topic:
Part 1: Lecturers: Meng Cao
Topic: The Quantization Model of Neural Scaling (slides, video)
Part 1: Lecturers: Rishika Bhagwatkar, Shubham Gupta
Topic: Perceiver: General Perception with Iterative Attention (slides, video)
Part 1: Lecturers: Andrei Mircea
Topic: Maximal update parameterization / transfer (slides, video)
Part 2: Lecturers: Adel Nabli, Johan Obando
Topic: DiLoCo: Distributed Low-Communication Training of Language Models (slides, video)
Part 1: Lecturers: Maxime Gevers
Topic: Segment Anything (slides, video)
Part 2: Lecturers: Nithya Shikarpur, Subhrajyoti Dasgupta
Topic: Adding Conditional Control to Text-to-Image Diffusion Models (slides, video)
Part 1: Lecturers: Juan David Vargas
Topic: Recurrent Independent Mechanisms (slides, video)
Part 1: Lecturers: Sarvjeet Singh Ghotra
Topic: Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos (slides, video)
Part 2: Lecturers: Sparsha Mishra, Maxime Gevers
Topic: When to Make Exceptions: Exploring Language Models as Accounts of Human Moral Judgment (slides, video)
Part 1: Lecturers: Neeraj Kumar, Paul Janson
Topic: Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (slides, video)
Part 2: Lecturers: Ayoub Echchahed, Nassim El Massaudi
Topic: Stop Regressing: Training Value Functions via Classification for Scalable Deep RL (slides, video)