A Technology that Learns Like the Human Brain - Neuromorphic Computing
BrainChip's technology uses a type of neuromorphic computing called spiking neural networks (SNNs). It has many attractive characteristics, including the ability to be trained rapidly, high accuracy and low compute overhead. This is an important feature in the world outside of the internet, where massive datasets are not available. For instance, a police department looking for a suspect in live video streams does not have thousands of images of that suspect, nor does it have weeks to train a traditional convolutional neural network system.
What Makes Our Technology So Exciting
Our technology learns from experience, autonomously, just like a human. Deep Learning networks are power-hungry and require large GPU-Server clusters and weeks of training. BrainChip's SNNs can learn in an unsupervised manner, without large datasets, and finds patterns that humans may not be aware of. This rapid learning capability opens up new possibilities to find objects in video, patterns in large datasets, and hundreds of other applications.
Because SNNs can be implemented using regular logic functions, they are inherently high-performance and low-power.
How It Works
Spiking neural networks are inherently feed-forward networks, for both training and inference. Our neuron models learn through selective reinforcement or inhibition of synaptic connections and programmable firing thresholds. Ingrained in these models are innovative training methodologies for supervised and unsupervised training. In the supervised mode, the initial layers of the network train themselves autonomously, while in the final fully-connected layers, labels can be applied, enabling these networks to function as classification networks. The Akida NSoC is designed to perform either on-chip training or off-chip training in the Akida Development Environment.