Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper proposes replacing dense vector representations in the Thousand Brains Theory (TBT) with rank-order spike packets to encode the temporal sequence of sensor contacts during object recognition. By using the timing of neural events to implicitly encode sensor displacement and applying STDP for learning, the model captures spatial relationships between features. Experiments on synthetic data demonstrate that this temporal coding approach achieves superior discrimination accuracy, especially for objects with identical features but different spatial arrangements, and maintains robustness under noise.
Spatial relationships between object features can be learned by encoding the *order* in which they are sensed, boosting recognition accuracy where traditional methods fall flat.
The Thousand Brains Theory (TBT) and its open-source Monty framework model object recognition through sensorimotor inference -- identifying objects by actively moving a sensor across their surface and building evidence contact by contact. The current implementation encodes each contact as a dense floating-point vector. While Monty tracks inter-step displacement and accumulates evidence across contacts, it treats the feature activation pattern at each contact as an unordered set - the directional sequence in which features are encountered carries no representational weight. In TBT, the sequence of contacts carries spatial meaning: knowing that feature A was felt before feature B during a left-to-right sweep tells you something about where A and B sit on the object. Dense vectors discard this ordering. We propose replacing dense vectors with rank-order spike packets: each contact produces a brief burst of neural events where the most strongly activated neuron fires first. The time gap between successive bursts implicitly encodes sensor displacement without explicit coordinate calculations. A biologically motivated learning rule (STDP) encodes traversal direction into synaptic weights. A learnable parameter lambda adjusts reliance on earlier versus recent contacts, adapting to each object's geometry. We derive three testable predictions and specify an implementation of four components in approximately 450 lines of NumPy. Three synthetic experiments confirm the core claims: temporal coding achieves perfect discrimination accuracy on objects with identical features in different spatial arrangements, where dense accumulation performs at chance; temporal coding maintains a 30-50 percentage point advantage across all tested noise levels; the adaptive lambda converges to distinct values, reflecting object geometric complexity. End-to-end evaluation on Monty's YCB benchmark is left for future work.