Orly Ramichanov, Pavel Kounitsky
Bat Brain Project aims to create a huge artificial intelligence neural network in order to mimic the echolocating function of the bat's brain. Bats use
echolocating ultrasonic calls to scan the environment, detect prey, etc. In this tricky way bats manage to 'see' with their ears. We would like to engineer a system which would resemble this functionality. The challenge is to build an artificial brain which would learn the echolocating patterns and interpret them into a real image. This way we could understand what bats really see.
The neocortex is a part of the bat's brain which is built up of many layers. It is involved in high level functions such as sensory perception. We aim to create a software model based on Hierarchical temporal memory (HTM) algorithms that will provide a large range of machine-learning capabilities, and recreate the orientational functionality of the bat's brain. This is a challenging task which requires high performance computing resources. With the latest breakthroughs in the field of parallel GPU computing, it is now possible to create huge neural networks of ~1 Billion synapses just with a few machines. This kind of tasks were previously made possible only by a multi-million dollar data centers.