Blog
Writeups of projects and some miscellaneous notes. (Slowly) adding interactive figures and better visualizations to my posts.
Technical posts
2026
- ENPIRE for SO-101 ENPIRE for SO-101.
- Offline RL: SAC+BC, IQL, and flow Q-learning on OGBench SAC+BC, IQL, and flow Q-learning on OGBench and navigation task.
- A RAG system for the Berkeley EECS directory with 4 GB of RAM Crawling ~15K eecs.berkeley.edu pages, writing a 138-question QA set, and building a pipeline of dense retrieval, cross-encoder reranking, and Llama-3.1-8B.
- Ideal flow machines: where does a diffusion model's creativity come from? Reimplemented and analyze analytic score machines (IS, LS, ELS, bbELS).
- From DQN to SAC Double DQN on CartPole, LunarLander, and MsPacman, SAC with clipped double-Q and auto-tuned temperature on HalfCheetah and Hopper.
- Policy gradients, one variance reduction at a time REINFORCE on CartPole, HalfCheetah, LunarLander, and InvertedPendulum, adding reward-to-go, advantage normalization, learned baseline, and GAE.
- Facial keypoints: regression vs. transfer learning vs. heatmaps Predicting 68 facial landmarks with a CNN coordinate regressor, fine-tuned ResNet-18 and DINOv2 backbones, and a U-Net outputting Gaussian heatmaps.
- Behavioral Cloning with Flow Matching Policies Behavior cloning on Push-T with an MSE action-chunking policy and a flow matching policy.
Notes
2026
- RL for VLAs Notes from Seohong Park's lectures in Sergey Levine's VLA seminar CS 294-318.
- clanker My 3D-printed SO-101 arm.
- Noise Optimization for Collapse Recovery in Trained Diffusion Models Some notes on a cool paper.
CS 180 projects · Fall 2025
2025
- Diffusion and Flow Matching Sampling, CFG, SDEdit, inpainting, and visual anagrams with DeepFloyd IF, then training a flow-matching UNet on MNIST.
- Neural Radiance Fields Calibrating my own camera, fitting a 2D neural field, then a NeRF from scratch: ray sampling, volume rendering, novel views.
- Autostitching Photo Mosaics Homographies, projective warping, and fully automatic stitching with Harris corners, ANMS, and RANSAC.
- Filters and Frequencies Convolution, unsharp masking, hybrid images, and Laplacian-pyramid blending (the oraple).
- Coloring the Prokudin-Gorskii photo collection Aligning the 3 glass-plate exposures with NCC and an image pyramid.
Misc: course reviews · study streams