Sequence Models & RL
A live sequence-models foundations book and an authored RL + control-theory guide — the research direction.
ssm-foundations
in progressPublic foundations book derived from the post-transformers research repo. 17 chapters organized lens-first: continuous-time dynamics → discretization → stability theory → structured state-space models → selective scans (Mamba, S4, S4D) → hybrids. Each chapter pairs prose with JAX and Julia companions, reproducible figures, and Julia stability tests. Built on book-scaffold-astro v4 academic preset; deployed via Cloudflare Workers + Static Assets.
Stack: Astro 6 · MDX · JAX · Julia · DifferentialEquations.jl · KaTeX · Cloudflare Workers
What's next
Live at https://ssm-foundations.brandon-behring.dev. Foundations chapters authored; actively writing the structured-SSM layer (HiPPO, S4/S4D, Mamba) and the later 'beyond SSMs' chapters (linear attention, DeltaNet, xLSTM/RWKV).
rl_and_control — RL + Control Theory guide
in progressIn-progress authored guide bridging classical optimal control (Bellman, DP, MPC, LQR/LQG) with modern reinforcement learning (policy gradient, actor-critic, model-based, MPC-RL hybrids). 28-week syllabus organized into 3 parts: RL Foundations (10 weeks) -> Control Theory (8 weeks) -> Convergence (10 weeks). Includes Python (PyTorch/Gymnasium/SB3), JAX (Brax, differentiable MPC), and Julia (ControlSystems.jl, JuMP MPC) experiment tracks. Bibliography: 135 papers indexed across 18 thematic sections with status tracking and 12 method-family dossiers.
Stack: Astro · MDX · Python · PyTorch · JAX · Julia · Gymnasium · SB3 · ControlSystems.jl
What's next
Local-only; publishing surface (Astro/MDX web guide) planned, matching the ssm-foundations book-scaffold-astro pattern. Bibliography (rl_and_control/references/paper_index.md) is the data source for the /lab/research-graph/ demo (shipped 2026-05-24, densified 2026-05-25).