Aligned and oblique dynamics in recurrent neural networks

HIGHLIGHTS

• What: The authors show that the magnitude of the readout weights can serve as a control knob between the regimes. The authors aim for a theoretical understanding of neural representations: What determines how strongly activity and behavioral output variables are related? The authors show that these networks can operate between two extremes: an "aligned" regime in which the output weights and the largest PCs are strongly correlated. The authors moreover assume that the magnitude of the activity ∥x∥ is constrained which the authors show below is the case for robust RNNs in presence of noise.
•  

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or E-mail:

  If you don't have an account, you can create one here.