Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

NEURAL NETWORKS

Linking task structure and neural network dynamics

Nature Neuroscience volume 25, pages 679–681 (2022)Cite this article

Subjects

The solutions found by neural networks to solve a task are often inscrutable. We have little insight into why a particular structure emerges in a network. By reverse engineering neural networks from dynamical principles, Dubreuil, Valente et al. show how neural population structure enables computational flexibility.

Artificial neural networks can be trained to achieve human-level performance in games and brain-like performance on neuroscience tasks1,2, but we still do not understand how the magic happens. In ‘backprop’ — the go-to learning algorithm for training deep networks — the error gradient gradually pushes each neuron away from the unfavorable activity it is generating and the network toward better task performance. Slowly edging away from the mistakes, ex negativo, seems to work well. But on the flipside, we still do not understand what units are actually doing once they have converged to a particular solution; it still feels a bit like pulling a rabbit out of a hat.

Your institute does not have access to this article

Access options

Buy article

Get time limited or full article access on ReadCube.

$32.00

Buy

All prices are NET prices.

Fig. 1: Reverse engineering the solution to the perceptual decision-making (DM) and context-dependent decision-making (CDM) tasks with low-rank recurrent neural networks (LR-RNNs).

References

  1. Botvinick, M. et al. Trends Cogn. Sci. 23, 408–422 (2019).

    Article  PubMed  Google Scholar 

  2. Chaisangmongkon, W., Swaminathan, S. K., Freedman, D. J. & Wang, X.-J. Neuron 93, 1504–1517.e4 (2017).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  3. Cunningham, J. P. & Yu, B. M. Nat. Neurosci. 17, 1500–1509 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  4. Gallego, J. A. et al. Nat. Commun. 9, 4233 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  5. Raposo, D., Kaufman, M. T. & Churchland, A. K. Nat. Neurosci. 17, 1784–1792 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  6. Hirokawa, J., Vaughan, A., Masset, P., Ott, T. & Kepecs, A. Nature 576, 446–451 (2019).

    CAS  Article  PubMed  Google Scholar 

  7. Yang, G. R., Joglekar, M. R., Song, H. F., Newsome, W. T. & Wang, X.-J. Nat. Neurosci. 22, 297–306 (2019).

    CAS  Article  PubMed  Google Scholar 

  8. Yang, G. R., Cole, M. W. & Rajan, K. Curr. Opin. Behav. Sci. 29, 134–143 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  9. Márton, C. D., Lajoie, G. & Rajan, K. Preprint at https://doi.org/10.48550/arXiv.2105.14108 (2021).

  10. Duncker, L., Driscoll, L., Shenoy, K. V., Sahani, M. & Sussillo, D. Adv. Neural Inf. Process. Syst. 33, 14387–14397 (2020).

    Google Scholar 

  11. Dubreuil, A., Valente, A., Beiran, M., Mastrogiuseppe, F. & Ostojic, S. Nat. Neurosci. https://doi.org/10.1038/s41593-022-01088-4 (2022).

    Article  Google Scholar 

  12. Vyas, S., Golub, M. D., Sussillo, D. & Shenoy, K. V. Annu. Rev. Neurosci. 43, 249–275 (2020).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  13. Beiran, M., Dubreuil, A., Valente, A., Mastrogiuseppe, F. & Ostojic, S. Neural Comput. 33, 1572–1615 (2021).

    Article  PubMed  Google Scholar 

  14. Mastrogiuseppe, F. & Ostojic, S. Neuron 99, 609–623.e29 (2018).

    CAS  Article  PubMed  Google Scholar 

  15. Stroud, J. P., Porter, M. A., Hennequin, G. & Vogels, T. P. Nat. Neurosci. 21, 1774–1783 (2018).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

Download references

Author information

Authors and Affiliations

  1. Icahn School of Medicine at Mount Sinai, New York, NY, USA

    Christian David Márton, Siyan Zhou & Kanaka Rajan

  2. Department of Neurobiology, Harvard Medical School, Boston, MA, USA

    Siyan Zhou

Authors
  1. Christian David Márton
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Siyan Zhou
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Kanaka Rajan
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Kanaka Rajan.

Ethics declarations

Competing interests

The authors declare no competing interests.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Márton, C.D., Zhou, S. & Rajan, K. Linking task structure and neural network dynamics. Nat Neurosci 25, 679–681 (2022). https://doi.org/10.1038/s41593-022-01090-w

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41593-022-01090-w

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing