Blog
A place for me to post thoughts on a broad spectrum of topics.
On AI and Investment
- AI / ML
- 2025 final: Simplicity Scales
- 2024 final: A Minimal Introduction to Reinforcement Learning
- 2024 midterm: Ocean Deep - A Glimpse of Intelligence
- 2022 final: ML Trends and Projections
- Investment
- 2024 final: Going Long: A Glimpse into Value Investing (TBD)
- 2023 final: Keep Believing in Futures
- 2023 midterm: Left, Left, Right
- 2022 midterm: Stay Positive
- 2021 midterm: Trading Reflections
Tech Stack
- LLM Infra (2025)
- Session 5: Docker + K8s + Ray: cloud infra for SFT, RL (GRPO) and Serving
- Session 4: GPU Programming
- Session 3: Optimizations and LLM Inference
- Session 2: Parallelism and LLM Training
- Session 1: Transformer Model
Research Stack
Diffusion Models: A Very Short Introduction to Diffusion Models
On The Equivalence Between AUC and Contrastive Loss: TBD
Comparing MPC and PID controllers: a simple case.
There is a trend that some of the PID controllers are being replaced by more advanced one, i.e., Model predictive controller (MPC). However, in what aspects that MPC would make improvement is not theoretically clear. As a starting point, the following dicussion aims to reveal the difference between the two controllers on first order systems. Follow-up: convert MPC (linear-quadratic) as SP/SQP optimization problem (link).Useful Math Tools for ML Proofs
useful math tools for ML proofs: link