200 — Closed-loop control of a modular neuromorphic biohybrid

Keren et al (1802.07905)

Read on 08 March 2018
#neuroscience  #neuromorphic-computing  #hybrid  #live-culture  #neurons  #simulator  #biofidelic-computation  #artificial-neural-network  #biological-neural-network  #neural-network  #LIAF  #simulation  #control-system  #cell-culture  #BCI  #biohybrid 

It’s my 200th paper! Only 165 left to go.

Oh man.

But today’s paper is a really exciting one. In this research, Keren et al explore the concept of running neural simulators side-by-side with closed loop actual neural cultures in vitro.

The idea is this: Designing a neurofidelic model of computation is hard. There are a lot of moving parts in a neuron and it’s not obvious which of these features we need to model and which of these things are really unimportant for circuit-level simulation.

So instead of trying to simulate the system from scratch, why not learn the responses of a system to a given stimulus by recording from the system itself?

This technique uses a neural culture in vitro next to a computational model of a biological neural network to train the artificial neural network to behave more like the biological one.

The simulated Leaky-Integrate-And-Fire (LIAF) modeled neurons run in parallel with the microelectrode array that records and stimulates the real, biological neurons. Using this control system, it is possible to simulate many thousands of cells using the FPGA’s on-chip neuromorphic computing framework (similar in spirit to systems like TrueNorth et al), and to validate these results with the true neurons.

Additionally, it is possible to use the true neurons as a separate step in the processing pipeline, where the on-chip digital “neurons” act as a control algorithm for the biological cells in culture.

This biohybrid is extremely cool and extremely promising both for BCI work as well as for better understanding the control systems of the brain.