Session

DescriptionThe recent progress in artificial intelligence is inspired by early insights into brain function. However, the fundamental interaction between neurons and the architecture of the mammalian brain differ radically from present AI hard- and software. Neurons communicate by point-like events, called spikes, where mainly the timing carries information and each neuron emits only a few spikes per second. The brain is hierarchically organized but at the same time highly recurrent on multiple scales. While a neuron receives input from about ten thousand others, the vast network size leads to a very low connection probability. Thus, computation is extremely sparse in space and time; the opposite of computationally dense matrix operations. The field studying the dynamics and function of neuronal networks is computational neuroscience. The field of neuromorphic computing investigates how a computer can be built based on the fundamental interactions and architecture of the brain. Systematic integrated funding has brought Europe into a world-leading position. The contributors to this session work together in the Human Brain Project (HBP). After a brief introduction to the topic by Markus Diesmann, Susanne Kunkel will discuss the state of the art in the simulation of brain-scale neuronal networks on conventional supercomputers on the example of the NEST code. Subsequently, Steve Furber will demonstrate the design and capabilities of the SpiNNaker neuromorphic hardware, highlighting the one-million-core installation in Manchester. Finally, Mihai Petrovici will present progress on spike-based computation. Together, the talks show how computational neuroscience and neuromorphic computing create the foundations for the next-generation AI.