We have merged the decay operation into the read and write operations in our new 2-1 two-phase circuit. The result is 33% faster and lower power than previous three-phase designs and has twice the synaptic density! The design supports all
Via our new spike logic encoding we have demonstrated that the AHaH attractor states can access all logic functions without external thresholding logic gates (NAND). In other words, the AHaH NPU is provably computationally complete and can support any algorithm.
We have demonstrated stable AHaH attractor states in hardware simulation using our meta-stable switch model and the 2-2 node configuration. This result paves the way for demonstration of clustering.
We have sucessfully ported the fractal-flow-fabric concept for combinatorial optimization to our AHaH node functional model and used it optimize solutions to the traveling salesman problem. This result adds continuous real-time optimization to growing list of capabilities of an AHaH
We have simulated at the hardware level a small collective of 2-2 AHaH nodes operating as a clusterer! Our results show pairity with functional models.
We have simulated at the hardware level with MemSim a 2-2 AHaH node operating as a supervised classifier! We achieved a classification rate of 98.9% on the Wisconsin Breast Cancer test data set. Although small, this is an important real-world
We have demonstrated that the AHaH Classifier is capable of unsupervised learning. If no supervised labels are given but the classifier is able to output labels with high confidence, the output can be assumed to be correct and used as
We have demonstrated a thermodynamic bit formed of a AHaH node in a 2-1 configuration. This result paves the way for demonstrations of generic combinatorial optimization.
After demonstrating the feasibility of the AHaH node as a building block for adaptive neuromorphic we are moving to phase-II! We are extreamly excited to take the work to the next level, where we will be carefully constructing the bridge
We have demonstrated that collectives of AHaH nodes each controlling one fiber of opposing muscles will actuate a multi-jointed robotic arm toward a target. This result validates our hypothesis that the two phase “read/write” operation of the AHaH rule can