Convergent Learning: Do different neural networks learn the same representations?

author: Jason Yosinski, Department of Computer Science, Cornell University
published: May 27, 2016,   recorded: May 2016,   views: 10044


Related Open Educational Resources

Related content

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
Lecture popularity: You need to login to cast your vote.


Recent success in training deep neural networks have prompted active investigation into the features learned on their intermediate layers. Such research is difficult because it requires making sense of non-linear computations performed by millions of parameters, but valuable because it increases our ability to understand current models and create improved versions of them. In this paper we investigate the extent to which neural networks exhibit what we call convergent learning, which is when the representations learned by multiple nets converge to a set of features which are either individually similar between networks or where subsets of features span similar low-dimensional spaces. We propose a specific method of probing representations: training multiple networks and then comparing and contrasting their individual, learned representations at the level of neurons or groups of neurons. We begin research into this question using three techniques to approximately align different neural networks on a feature level: a bipartite matching approach that makes one-to-one assignments between neurons, a sparse prediction approach that finds one-to-many mappings, and a spectral clustering approach that finds many-to-many mappings. This initial investigation reveals a few previously unknown properties of neural networks, and we argue that future research into the question of convergent learning will yield many more. The insights described here include (1) that some features are learned reliably in multiple networks, yet other features are not consistently learned; (2) that units learn to span low-dimensional subspaces and, while these subspaces are common to multiple networks, the specific basis vectors learned are not; (3) that the representation codes show evidence of being a mix between a local code and slightly, but not fully, distributed codes across multiple units.

See Also:

Download slides icon Download slides: iclr2016_yosinski_convergent_learning_01.pdf (10.3┬áMB)

Help icon Streaming Video Help

Link this page

Would you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !

Reviews and comments:

Comment1 Charly Rich, February 10, 2021 at 2:49 a.m.:

Interesting Technik

Comment2 John, February 10, 2021 at 2:53 a.m.:

A very interesting technique. I tried to learn how to write using many methods, but I still couldn't. I always turned to this service for help, and they write the student paper for me. I'm not experiencing any problems, but I think I'll try your technique for fun.

Comment3 Carol, March 25, 2021 at 1:09 p.m.:

Convergence or interdisciplinarity is a reflection of the process of combining scientific knowledge and the integrative nature of modern scientific knowledge. And the school should teach exactly this new scientific knowledge, but an important component of the convergent approach is the continuity of the subject and the ways of its presentation. Thank you very informative. If you need to write an essay, you can always contact the experts - to better cover the topic.

Write your own review or comment:

make sure you have javascript enabled or clear this field: