Random/incomplete thoughts on various ML topics

  • A Few Notes on attention mechanisms
  • A Few Notes on Fisher Information (WIP)
  • Some Notes on Kolmogorov Complexity, the Algorithmic Markov Condition, and Causality
  • Causal Graphs and Sources of Bias
  • Likelihood Ratio Policy Gradients for Reinforcement Learning
  • Notes on Policy Gradients (Trust Region Policy Optimization, under construction)
  • Notes on Policy Gradients (under construction)
  • Notes on Noise Contrastive Estimation (NCE)
  • Notes on Maximization of Inner Products over Norm Balls
  • Notes on Adversarial Examples
  • Basic probability stuff that everyone likely should know
  • Notes on Variational Autoencoders
  • Notes on policy gradient and the log derivative trick for reinforcement learning (under construction)
  • LDA Intro/overview (also under construction)
  • Brief intro to auto-encoders (I would explain this differently/use different notation these days)
  • Pretty old random idea
  • How exactly does word2vec work?
  • Why is minimizing error the same thing as maximizing likelihood is the same thing as finding a low energy state...
  • A bit on sequence to sequence learning (e.g Google Inbox smart reply)

    Code Snippets

  • VariationalAutoencoder (VAE) and other examples written in tensorflow. An example reconstruction of MNIST digits by the VAE is shown below. .
  • lr.py: Very simple linear regression in tensorflow (made up dataset)
  • k-prototypes.py: Categorical Clustering for Netflow data
  • Softmax regression on MNIST in tensorflow
  • PCA + K-Means (Spark/MLlib/Scala) on the KDDCUP99 data set

    Last Update: 05.02.2021 by dmm@1-4-5.net