pomdp

From-scratch PPO/PPO-LSTM for partially observable MiniGrid DoorKey, featuring performance-gated adaptive curriculum learning and reproducible 5-seed benchmarks.

Modelo macroeconomico brasileiro como POMDP: divida publica endogena por indexador, credito segmentado e busca de politica multi-objetivo. Resultado contraria a hipotese que motivou o projeto.

A simulation of how people read intent into things other people make, what breaks when nothing made it, and what a provenance label does about it.

A Modern Probabilistic Model Checker

End-to-end empirical implementation of model-based off-policy evaluation and pessimistic policy selection for confounded POMDPs (Hong, Qi & Xu, ICML 2024)

Machine-checked Lean 4 / Mathlib formalization of Ismail's Primitives — six structural primitives proven necessary, mutually independent, and sequentially linked for sequential decision-making under uncertainty. 0 sorry · 0 axiom · ~12,700 lines.

RSSM world model on a POMDP pendulum, actor critic in imagination, and a representation ablation. Reconstruction measurably helps open loop prediction for 5 steps and not after.

Treats Markdown documents as Markov decision processes — parses structure into a NetworkX state space, optimizes it with reinforcement learning policies, tracks POMDP belief updates, and generates text from learned Markov chains. Python 3.8+ with Pandoc conversion; 238/243 tests passing.

Study 1 of the Hebbian Belief-State World Model: does a BDH core's plastic synapse state encode linearly readable beliefs? A preregistered negative result.

A Python implementation of active inference for Markov Decision Processes

A 150-line environment where looking costs $5, fixing $50, breaking $500. When is knowing worth its price? Zero dependencies. Beat the baseline.

Engineering decision environments where information has a price — typed physical/epistemic actions, an information-cost channel in the step API, 3 public adversarial audits.

Active inference in Go 1.24 — variational free energy belief updating and expected free energy policy selection for POMDP agents, with categorical, Gaussian, and Dirichlet belief representations on gonum. Concurrent policy evaluation and benchmarks, for robotics, game AI, and cognitive modeling.

Tool-component framework for an Active Inference agentic mesh — an abstract async ToolComponent interface with typed metadata, parameter and return-value dataclasses, plus two worked tools: a sandboxed MathTool and a discrete-POMDP Active Inference agent (A/B/C/D/E matrices, variational free energy, habit learning) with matplotlib diagnostics.

Constrained deep RL for joint control and resource-aware active sensing under partial observability.

Reinforcement learning agent that decides which gene expression test to order next, trading diagnostic accuracy against measurement cost. Rainbow DQN and PPO on TCGA pan-cancer data, with a browser dashboard that runs the trained policy live on held-out patients

A modular research repository for implementing, reproducing, and extending reinforcement learning algorithms, with a focus on GRPO, PPO, critic-free RL, and partial observability.

Standalone Active Inference and message-passing planning for partially observable environments, written in Mojo in pure 🔥🔥🔥

Knowledge base and Python package for discrete and continuous Active Inference: generative models, expected free energy, POMDP simulation, and an executable manuscript with reproducible validation commands.

NumPy-only pilot study coupling a temperature-scaled classifier calibration instrument to RL/heuristic path planners for energy-constrained aerial crop scouting.