Implementing the first Deepmind papers on deep reinforcement learning and realizing that the Sutton-Barto textbook is a masterpiece in technical pedagogy.

3

Why does BatchNorm work?

5

Forward-mode and reverse-mode autodiff in python featuring dual numbers and graph traversals

11

A note on how Gaussians are convolved to make the reparameterization trick work in the diffusion forward process.

12

Andrew Barto and Richard Sutton for pioneering Reinforcement Learning

20

Learning about where diffusion models break and collecting attempts at solutions

21

My thoughts on the "Grandmaster-Level Chess Without Search" paper from Google DeepMind

27

Notes and some ideas while reading the DreamCoder paper

29

A half-baked idea I had while checking my equations

53

Paper reading notes from w2v-BERT by Google Brain

62

Some ideas on improving semantic segmentation with SAM

65

Live Coding Seam Carving Part 2

76

there is a country layer in the llama 2 transformer

79

Three implementations of self attention, one for visualizing the self attention between two tensors, one for language, and one in PyTorch

88

Exercises in faster and more intuitive tensor notation for deep learning.

89

Helper code that has different ways to visualize audio files, mostly for future reference.

102

Testing whether Newton's method decays updates exponentially

104

Live Coding Seam Carving Part 1

107

13th International (Online) Conference on Transportation Planning and Implementation Methodologies for Developing Countries

138

Official Work Product Submission for the 2020 edition of Google Summer of Code with TensorFlow.

142

Hardcore History, Kobe Bryant and new adventures.

150