HTM Spatial Pooler + Temporal Memory + Reflex Acceleration

Accelerate HTM temporal memory by integrating a reflex LSH caching fast path for frequently observed SDR patterns, enabling real-time sequence learning and rapid predictive lookups. This pipeline combines Spatial Pooler and Temporal Memory for accelerated inference.

Reflex memory is an acceleration mechanism that caches frequent SDR patterns using LSH-based lookup. It works with Spatial Pooler outputs by tracking confidence, verifying baseline snapshots, and promoting observations to bypass standard processing for faster inference.

Configure an end-to-end HTM pipeline using a nested HTMConfig structure with configurable presets. Instantiate a minimal preset to run sequences and observe Spatial Pooler outputs, Temporal Memory predictions, and reflex fast-path results.

Yes, this HTM implementation requires PyTorch and pytest. PyTorch handles tensor operations for the SDR outputs, while pytest supports the included test and benchmark scaffold for rapid development and evaluation of the cognitive pipeline.

Yes, you can checkpoint and restore reflex memory using state_dict functionality. The system supports observation and promoted tables with confidence tracking and baseline snapshot verification, allowing consistent state persistence across sessions.

Avoid using reflex-accelerated HTM pipelines when working with highly novel or sparse data streams, as the LSH caching fast path relies on frequently observed SDR patterns. Standard Temporal Memory without reflex acceleration may handle rare sequences more effectively.