Harel Lidar

Archive

Everything here, by kind. Each kind is a different sort of writing, so each is shown differently. Titles without a link were read or attended and logged, but not published.

Essays

Longer pieces in my own words.

Evergreen notes

Atomic notes: each argues one claim, and the title is the claim.

Paper summaries

One paper each: the problem, the method, what the results do and don’t show.

Lecture notes

Notes taken in courses, grouped by course.

Statistical Information Theory

  1. Intro
  2. Entropy
  3. Symbol Codes
  4. Joint Entropy
  5. Channel Coding
  6. Probabilistic Inference
  7. Estimating Information
  8. Information Geometry

Intro to ML

  1. Lecture 1
  2. Decision Trees and KNNs
  3. Evaluation
  4. Neural Networks Part 2

Introduction to Symbolic AI

  1. Lecture 1
  2. Semantics
  3. SAT Problem
  4. First Order Logic
  5. Knowledge Representation
  6. Herbrand Models
  7. Problem Solving by Search
  8. Search with full observability, BFS, DFS, UCS, MiniMax, alpha-beta, Heuristic Search - A* and BestFS
  9. Search with Partial Observability

Mathematics for ML

  1. Lecture 1
  2. Lecture 2

SIT Course Summary

  1. SIT Course Summary

Reinforcement Learning

  1. Lecture 2
  2. Model Free Control
  3. Value function approximation
  4. function approximation

Value function Iteration methods

  1. Value function Iteration methods

AI Safety - Fundamentals course — Week 2 - What is alignment

  1. AI Safety - Fundamentals course — Week 2 - What is alignment

Policy Gradient Methods

  1. Policy Gradient Methods

ML Course Summary

  1. ML Course Summary

Deep Learning

  1. CNNs

Quantifying Fairness in ML

  1. Quantifying Fairness in ML

Computational Optimization

  1. First Order Optimization

Reflections From First Session

  1. Reflections From First Session

Reflections From Second Session

  1. Reflections From Second Session

General Concepts to internalize

  1. General Concepts to internalize

Graphical Models

  1. Graphical Models

Mixture of Models

  1. Mixture of Models

NEML

  1. Lecture 3

Flow Matching and Diffusion Models

  1. Lecture 1

Flow Matchings and Diffusion Models

  1. Training the Model to Predict the Vector Field

Concepts

A glossary: one idea per entry, explained briefly.

Dora (Weight Decomposed low-rank adaptation)
Continual learning & optimization
Effect of condition number of hessian matrix on convergence
Continual learning & optimization
Finetuning Base Model Methods
Continual learning & optimization
LoRa (Low Rank Adaptation)
Continual learning & optimization
Metric Spaces
Math & CS foundations
Outer Product of Vectors
Math & CS foundations
SVD intuition
Math & CS foundations

Excerpts

Mostly other people’s words: article summaries, quoted passages, course summaries. Kept as reading, not as my writing.