Animated presentation for my PhD Defense, which focused on my work on fast sketching on GPUs (ICML 2026 Spotlight): FlashSketch: Sketch-Kernel Co-Design for Fast Sparse Sketching on GPUs

Defense recording: YouTube

An illuminating perspective on floating point numbers that I haven't seen emphasized before.

My attempt at distilling the ideas behind the neural tangent kernel that is making waves in recent theoretical deep learning research.

The factor graph is a beautiful tool for visualizating complex matrix operations and understanding tensor networks, as well as proving seemingly complicated properties through simple visual proofs.

I experiment with Neural ODEs and touch on parallels between adversarial robustness and equilibria of dynamical systems.

I train a character level decoder RNN to generate words, conditioned on a word embedding which represents the meaning of the word.

I train autoencoders to identify components of doodles using a synthetic dataset, and use them to create nifty animations by interpolating in latent space.

Google brain recently published a paper titled Adversarial Reprogramming of Neural Networks which caught my attention. I explore the ideas of the paper and perform some of my own experiments in this post.