Task: Process Research Papers and Implement HippocampAIF
Research & Understanding
- Convert PDFs to page images (3 papers → 3 folders)
- Read Lake et al. BPL paper (56 pages) — Bayesian Program Learning
- Read oneandtrulyone Distortable Canvas paper (13 pages) — One-shot learning (and re-analyzed)
- Read Friston Free-energy Principle paper (9 pages) — FEP for brain (and re-analyzed)
- Read hippocampaif.md user preference file
Planning
- Create implementation plan for HippocampAIF framework
- Get user approval on plan
Execution (10 Phases)
Phase 1: Core infrastructure (tensor, free_energy, message_passing, dynamics)
Phase 2: Biological Retina (Days 3-4)
- Photoreceptor log-compression.
- Ganglion ON/OFF center-surround.
- Spatiotemporal energy transients.
- Verify tests pass.
Phase 3: Visual Cortex V1-V5 + HMAX (Days 5-6)
- V1 Simple Cells (Gabor filters).
- V1 Complex Cells (Sparse pooling).
- Extrastriate HMAX Hierarchy.
- Verify tests pass.
Phase 4: Hippocampus (Days 7-8)
[/] Phase 5: Spelke's Core Knowledge (object, agent, number, geometry, social, physics)
Phase 6: Neocortex + Attention (predictive_coding, prefrontal, SC, precision)
Phase 7: One-Shot Learning (distortable_canvas, AMGD, classifier, Hebbian)
Phase 8: Action & Active Inference (motor_primitives, active_inference, reflex)
Phase 9: Integrated Agent (brain, mnist_agent, breakout_agent)
Phase 10: Setup & packaging
Verification
- Per-phase component tests (not stubs)
- MNIST one-shot benchmark (>90% with 1 sample/digit)
- Breakout mastery benchmark (<5 episodes)
- Create walkthrough artifact