HUJI Machine Learning Club Courses

SGD Exact Dynamics in High-Dimension - Insights for Algorithm Design

Explore SGD dynamics in high-dimensional regimes through theoretical frameworks, covering generalized linear models, adaptive methods, and differentially private optimization insights.

• YouTube
• 1 hour 2 minutes
• Self-Paced
• Free Video

Mitigating the Curse of Detail - A Heuristic for Feature Learning and Sample Complexity

Discover a heuristic framework that bypasses analytical complexity in deep learning theory by focusing on scaling arguments to predict feature learning patterns.

• YouTube
• 1 hour 2 minutes
• Self-Paced
• Free Video

Consistent Amortized Clustering via Generative Flow Networks

Discover a novel framework for amortized clustering using Generative Flow Networks that ensures order-invariant cluster assignments and outperforms existing methods on real-world data.

• YouTube
• 1 hour 2 minutes
• Self-Paced
• Free Video

Fast Agnostic Learners in the Plane

Explore fast agnostic learning algorithms for geometric concept classes in the plane, including triangles, convex polygons, and convex sets with optimal sample complexity.

• YouTube
• 1 hour 11 minutes
• Self-Paced
• Free Video

A Tight Lower Bound for Non-Stochastic Multi-Armed Bandits with Expert Advice

Discover tight lower bounds for non-stochastic multi-armed bandits with expert advice, resolving a long-standing open question in machine learning theory.

• YouTube
• 46 minutes
• Self-Paced
• Free Video

Statistics-Powered ML - Reliable Black-Box Inference from Untrusted Data

Discover how statistical principles enhance ML reliability for black-box inference with limited data and distribution shifts, featuring conformal prediction and test-time training methods.

• YouTube
• 1 hour
• Self-Paced
• Free Video

A Theory of Learning with Autoregressive Chain of Thought

Explore PAC-learning framework for autoregressive chain-of-thought in language models, covering sample complexity and efficient universal learning methods.

• YouTube
• 1 hour 1 minute
• Self-Paced
• Free Video

Leveraging Temperature Scaling and Class Similarity for Conformal Prediction

Discover how temperature scaling and class similarity enhance conformal prediction methods for reliable classification with guaranteed coverage probabilities.

• YouTube
• 1 hour 3 minutes
• Self-Paced
• Free Video

Differentially Private Quasi-Concave Optimization - Bypassing the Lower Bound and Application to Geometric Problems

Explore differentially private optimization of quasi-concave functions, bypassing lower bounds with new algorithms for geometric problems like center point selection and halfspace learning.

• YouTube
• 1 hour 2 minutes
• Self-Paced
• Free Video

Real-World AI: Risk and Robustness in Reinforcement Learning and Kalman Filtering

Explore real-world applications of AI through risk-aware reinforcement learning and robust Kalman filtering, focusing on practical challenges and solutions for handling uncertainties in AI systems.

• YouTube
• 58 minutes
• Self-Paced
• Free Video

Geometry-oriented Measures of Dependence in Statistical Analysis

Explore geometric approaches to measuring statistical dependence, examining axioms and algorithms for evaluating relationships between variables while comparing classical and modern dependence measures.

• YouTube
• 1 hour 1 minute
• Self-Paced
• Free Video

Optimal Prediction Using Expert Advice and Randomized Littlestone Dimension

Delve into optimal prediction strategies and online learning theory through exploration of expert advice systems and the randomized Littlestone dimension, focusing on practical applications in weather forecasting and image classification.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

Efficiency in the Age of Large Scale Models - Designing and Optimizing Deep Learning Systems

Explore the challenges and solutions of large-scale AI models, from theoretical foundations to practical efficiency improvements, with insights on reducing computational costs while maintaining performance in modern machine learning.

• YouTube
• 1 hour 4 minutes
• Self-Paced
• Free Video

Towards Reverse Algorithmic Engineering of Neural Networks

Explore the algorithmic decision-making process of neural networks through NeuroSAT case study, examining how ML models learn and apply combinatorial features and confidence-based variable selection in SAT solving.

• YouTube
• 1 hour 3 minutes
• Self-Paced
• Free Video

Graph Neural Networks: Modeling Interactions Between Vertices Through Walk Index Analysis

Explore the theoretical foundations of Graph Neural Networks' ability to model vertex interactions, focusing on separation rank, walk index, and a novel edge sparsification algorithm for improved performance.

• YouTube
• 1 hour 2 minutes
• Self-Paced
• Free Video