Research Readings
A curated list of research papers I’m reading or have read, organized by topic. Each entry includes links, brief notes. And I plan to port summaries from my ipad in the near future.
Mechanistic Interpretability
Title | Authors | Year | Notes |
---|---|---|---|
A Mathematical Framework for Transformer Circuits | Olah et al. | 2021 | |
Toy Models of Superposition | Elhage et al. | 2022 | |
Scaling and Evaluating Sparse Autoencoders | Gao et al. | 2024 |
(Causal)Representation Learning
Title | Authors | Year | Notes |
---|---|---|---|
Beta-VAE: Learning Basic Visual Concepts | Higgins et al. | 2017 | KL Divergence Penalty term to a standard VAE loss func |
Streaming / Encoding
Title | Authors | Org | Year | Notes |
---|---|---|---|---|
Complexity-Based Consistent-Quality Encoding In The Cloud | De Cock et al. | Netflix | 2016 | Tradeoffs between per title encoding vs per chunk encoding; bitrate ladder per title |
Fast algorithm for HDR video pre-processing | Andrey Norkin. | Netflix | 2016 | Better performance over iterative method; Root Cause: non linear transfer func in HDR10 in low luminance ranges; Still computes pixel to pixel (ig pre DL) |
Film Grain Synthesis for AV1 Video Codec | Andrey Norkin. | Netflix | 2018 | Autoregressive Model for modelling film grain. Properties of grain: non temporal, spatial correlations, only on smooth regions. Color Componenets unclear. Is it slow because of autoregressive nature? Potential for DL? |