OptoRoboRat III

Creation of a simulation environment for virtual behavioral experiments with rats, learning and explanation of animal behaviour, and realisation of a robotic rat


Relevant for Research Area

C - Applications

The project builds on

DeepDecode

PIs

Prof. Thomas Brox

Jun.-Prof. Joschka Boedecker


Summary and state of the project

This project develops new machine learning and simulation tools to investigate the influence of neural brain activity on the behavior of rodents. During the funding period, we were able to establish a new paradigm for neural decoding and behavior prediction in animals. Our method called “NeuRL” [1] builds on recently developed inverse reinforcement learning algorithms, and allows extracting a latent reward function from observed animal behavior to which we fit features derived from neural recordings. Our models enable significantly more accurate behavior prediction than state-of–the-art methods. In addition, in contrast to other methods, they are able to successfully capture trends in reaction time changes in ontogenetic perturbation experiments. This new method for neural decoding and behavior prediction will significantly speed-up and facilitate the identification of suitable target neurons for optogenetic stimulation in future experiments within OptoRoboRat.

[1] Gabriel Kalweit, Maria Kalweit, Mansour Alyahyay, Zoe Jaeckel, Florian Steenbergen, Stefanie Hardung, Ilka Diester and Joschka Boedecker. NeuRL: Closed-form Inverse Reinforcement Learning for Neural Decoding. ICML 2021 Workshop on Computational Biology, https://icml-compbio.github.io/2021/papers/WCBICML2021_paper_11.pdf