Akida – Low-power Event-based Neural Networks for Fast Inference at the Edge

Peter van der Made will talk about Akida, BrainChip's Neuromorphic System on Chip (NSoC), that represents the first in a new breed of AI neuromorphic computing. Akida uses spiking neural networks (SNNs) which has many attractive characteristics, including the ability to be trained rapidly, with high accuracy and low compute overhead. This is an important feature in the world outside of the Internet, where large datasets are not available.

Akida learns from experience, autonomously, just like a human and consumes very little power. Akida can learn rapidly in an unsupervised manner, without large datasets, without the Internet, and finds patterns that humans may not be aware of. This rapid learning capability opens up new possibilities to very rapidly find objects in video, surveillance, drones, automotive and robotics as well as acoustic analysis, cybersecurity, and the industrial IoT. Because SNNs can be implemented using regular logic functions, Akida is inherently high-performance and low-power. The Akida NSoC is designed to perform either on-chip training or off-chip training in the Akida Development Environment.

Akida is an Edge Network device that includes all necessary functions to implement a complete neural network and supports both event-based Spiking Neural Networks (SNN) and Convolutional Neural Networks (CNN). Akida is trained and performs inference on-chip. The device can learn autonomously as new patterns emerge without re-training the entire network. This unsupervised autonomous learning capability establishes Akida as a first in Neuromorphic AI Edge solutions.

Please arrive before 6 pm to ensure entrance