Add SIML FEPGate (one-bit surprise gating) to V-JEPA 2 planning loop #98
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation
Short-horizon latent planning can waste samples in spurious minima. We add a one-bit surprise gate (FEPGate) that prunes high-surprise elites during CEM/MPC, improving both accuracy and efficiency—no model retraining.
Paper Link
What’s in this PR
Optional SIML sidecar integration (surprise, step CLI) via .npy + JSON IPC
Elite-set gating hook in the CEM loop (hard/soft)
τ-calibration script (obs-driven 5–10th percentile)
Repro harness for OFF vs FEP comparisons (100 eps reaching)
Plots (violin, latency, energy) and hero schematic
Key results (frozen V-JEPA 2, same budget)
Final error 0.193 m → 0.109 m
Monotonicity 0.12 → 0.43
Latency 2.28 s → 1.13 s
Energy/episode 0.058 Wh → 0.029 Wh
EDP ~4× lower
How to run