||This workshop is intended for people interested in working with SpiNNaker. Tutorials for using SpiNNaker: https://spinnakermanchester.github.io/workshops/eighth.html/.
Our particular aim is to implement BitBrain on SpiNNaker neuromorphic platform and learn event-based stimuli from sensors. This work was presented during NICE conference 2022: https://flagship.kip.uni-heidelberg.de/jss/HBPm?mI=235&publicVideoID=8944.
BitBrain is a learning algorithm based upon a novel synthesis of ideas from sparse coding, computational neuroscience and information theory that support single-pass learning, accurate and robust inference, and the potential for continuous adaptive learning. They are designed to be
implemented efficiently on current and future neuromorphic devices as well as on more
conventional CPU and memory architectures.
The SBC memory stores coincidences between features detected in class examples in a training
set, and infers the class of a previously unseen test example by identifying the class with which it
shares the highest number of feature coincidences. A number of SBC memories may be combined
in a BitBrain to increase the diversity of the contributing feature coincidences. The resulting
inference mechanism is shown to have excellent classification performance on benchmarks such
as MNIST and EMNIST, achieving classification accuracy with single-pass learning approaching
that of state-of-the-art deep networks with much larger tuneable parameter spaces and much
higher training costs. It can also be made very robust to noise.
BitBrain has some similarities to kernel-based classification methods, and this relationship is
discussed in detail. However, BitBrain has much lower computational requirements than kernel
methods (and much lower than deep networks) both in training and in inference, and is therefore
well-suited to edge applications where it could also interface efficiently with event-based sensors
such as silicon retinas.