University of Cambridge > Talks.cam > Computational Neuroscience > Computational Neuroscience Journal Club

Computational Neuroscience Journal Club

Add to your list(s) Download to your calendar using vCal

  • UserYul Kang and Wayne Soo
  • ClockTuesday 20 October 2020, 15:00-16:30
  • HouseOnline on Zoom.

If you have a question about this talk, please contact Jake Stroud.

Please join us for our fortnightly journal club online via zoom where two presenters will jointly present a topic together.

Zoom info: https://us02web.zoom.us/j/81395647267?pwd=YW9Ub1YzTUpBbndzZXl4c0loU2pqUT09 Meeting ID: 813 9564 7267 Passcode: 839088

The next topic is ‘recurrent neural network (RNN) models of spatial navigation’.

RNNs are suited for the study of circuits involved in spatial navigation because (1) unlike feedforward networks, they are capable of maintaining an internal state (e.g., current location of the agent/animal in the arena) and updating it, which is necessary for navigation, and (2) the brain regions involved in spatial navigation (hippocampal-entorhinal system) are known to have recurrent connectivity that is important for maintaining their spatial representation.

In Part 1, we will look at some early and straightforward approaches that directly use RNN models. Kanitscheider et al. trained their network to perform simultaneous location and mapping. Banino et al. used an RNN to perform path integration, and investigated the efficiency of its resultant grid-like representations. Cueva et al. tackled a similar path integration task with their own RNN model, which gave rise to various spatial-selective units such as grid and band cells.

In Part 2, we will cover recent proposals that push the boundary of the field by studying unsupervised training or incorporating more biological structure into the model. Recanatesi et al. trained their network without supervision using predictive learning, and offered an explanation why predictive learning gives rise to low-dimensional representation of latent variables. Evans et al. incorporated known hippocampal-entorhinal structure into their model and explained the observed pattern of the hippocampal-entorhinal activity that deviates from what would be expected from a simple rule of the physical space.

Papers:

Kanitscheider, I., Fiete, I. Training recurrent networks to generate hypotheses about how the brain solves hard navigation problems. NeurIPS (2017). http://papers.nips.cc/paper/7039-training-recurrent-networks-to-generate-hypotheses-about-how-the-brain-solves-hard-navigation-problems

Banino, A., Barry, C., Uria, B., Blundell, C., Lillicrap, T., Mirowski, P., Pritzel, A., Chadwick, M., Degris, T., Modayil, J., Wayne, G., Soyer, H., Viola, F., Zhang, B., Goroshin, R., Rabinowitz, N., Pascanu, R., Beattie, C., Petersen, S., Sadik, A., Gaffney, S., King, H., Kavukcuoglu, K., Hassabis, D., Hadsell, R., Kumaran, D. (2018). Vector-based navigation using grid-like representations in artificial agents. Nature https://dx.doi.org/10.1038/s41586-018-0102-6

Cueva, C., Wang, P., Chin, M., Wei, X. Emergence of functional and structural properties of the head direction system by optimization of recurrent neural networks. ICLR Spotlight (2020). https://openreview.net/forum?id=HklSeREtPB

Recanatesi, S., Farrell, M., Lajoie, G., Deneve, S., Rigotti, M., Shea-Brown, E. Predictive learning extracts latent space representations from sensory observations. bioRxiv (2019). https://dx.doi.org/10.1101/471987

Evans, T., Burgess, N. (2020). Replay as structural inference in the hippocampal-entorhinal system. bioRxiv https://dx.doi.org/10.1101/2020.08.07.241547

This talk is part of the Computational Neuroscience series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2024 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity