Expand description
AIXI planning components, environments, and model abstractions. AIXI Agent Implementations
This module contains:
- Monte Carlo AIXI (MC-AIXI) with pluggable predictive models.
- AIQI (Universal AI with Q-Induction) with phase-indexed return prediction.
AIXI is a theoretical mathematical formalism for universal artificial intelligence, which combines Solomonoff induction with sequential decision theory. This implementation follows the “Monte Carlo” approach (MC-AIXI) introduced by Veness et al., which uses Monte Carlo Tree Search (MCTS) to approximate the optimal policy.
§VM Backend
A high-performance Firecracker-based VM environment is available via the vm feature:
- NyxVmEnvironment: Uses nyx-lite for 10,000+ resets/second (requires KVM).
Modules§
- agent
- The core AIXI agent implementation.
- aiqi
- AIQI (Universal AI with Q-Induction) implementation.
- common
- Common types and utilities for the AIXI implementation.
- environment
- Standard benchmark environments for AIXI.
- mcts
- Monte Carlo Tree Search (MCTS) for AIXI.
- model
- Predictive models for AIXI.
- vm_nyx
- High-performance VM-backed AIXI environment using nyx-lite (Firecracker).