Modern Hopfield associative memory on a hypercube graph. Sparse local-attention retrieval with exponential capacity. C++23 and Python SDKs.

Volumetric language model with Triangle Cross-Scan State Modelling. Without Attention. With Neural Turing Machines (NTM) & Differentiable Neural Computers (DNC) smells. And maybe with bulk emit without TBT AR loop.

A non-parametric memory system inspired by hippocampal architecture. Using a three-stage training curriculum, the model learns to store images in a compressed latent space and retrieve them using corrupted or partial "cues".

VHDL implementation of CAM (Content Addressable Memory)

Build a sparse Hopfield network on a hypercube graph for fast local-attention retrieval with lower update cost and high capacity

Lara Networks, Inc. (founded as Lara Technology) was a fabless semiconductor company headquartered at 110 Nortech Parkway, San Jose, California, that architected, designed, and marketed silicon-based packet-processing solutions for high-speed Internet and networking infrastructure.

Fault-tolerant associative memory via 3D Neural Cellular Automata. 100% retrieval at N=10,000 patterns, zero catastrophic forgetting, 27K parameters.

Distributed Content Addressable Storage for Agentic Harnesses

A bit-native predictive machine: next-bit (0/1) prediction via a learned binary-address content-addressable memory. No tokens, no embeddings — vocabulary of 2. An honest, adversarially-verified research log.