Calendar of Events
Representing Manifolds Representation by Bioinspired Neural Nets
Speaker: Anirvan Sengupta (Rutgers University)
I will give a brief introduction to the problem of nonlinear dimension reduction or manifold learning. As the name suggests, this is an unsupervised method for discovering nonlinear continuous latent spaces. After the brief introduction, I will describe how some neural networks describe manifolds: not by setting up local "cartesian' coordinates, but by having different neurons being active at different but overlapping neighborhoods. In some cases, autoencoders describe data manifolds this way. So do place fields in the hippocampus.
For orientation, see the scikit-learn page on manifold learning: https://scikit-learn.org/stable/modules/manifold.html#:~:text=Manifold%20learning%20is%20an%20approach,sets%20is%20only%20artificially%20high
Our paper on manifold tiling: https://proceedings.neurips.cc/paper/2018/file/ee14c41e92ec5c97b54cf9b74e25bd99-Paper.pdf