University of Cambridge > Talks.cam > Machine Learning Reading Group @ CUED > Circuits and Interpretability

  Browse A-Z

Log in

Googledetails Microsoftdetails

🔐 Log In

Information on

Subscribing to talksdetails Finding a talkdetails Adding a talkdetails Disseminating talksdetails Help and Documentationdetails

ℹ️ Information

Circuits and Interpretability

Download to your calendar using vCal

If you have a question about this talk, please contact Elre Oldewage .

In this talk we will look at methods that aim to make the internal computations of neural networks visible (‘interpretable’) to humans. This is useful for a) making deep learning models robust / fair / safe and b) in order to come to an empirical, scientific understanding of why deep learning works. We will cover various methods from the literature, and focus in particular on the study of circuits, i.e. modular subnetworks that serve a particular function.

Recommended reading:

The Building Blocks of Interpretability (https://distill.pub/2018/building-blocks/)

Optional reading:

  1. Adversarial Examples Are Not Bugs, They Are Features (https://arxiv.org/abs/1905.02175)
  2. Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Masks (https://arxiv.org/abs/2010.02066)

Our reading groups are live-streamed via Zoom and recorded for our Youtube channel. The Zoom details are distributed via our weekly mailing list.

This talk is part of the Machine Learning Reading Group @ CUED series.

This talk is included in these lists:

Note that ex-directory lists are not shown.

Other lists

Centre for Mathematical Sciences, MR4 Babraham Events Lecture Room, Institute for Manufacturing

Other talks

Complete and Submit Project Choice Forms Why people with schizophrenia die earlier? Mortality and accelerated ageing Title and abstract tba Coming soon Talk title TBC Clonal haematopoiesis, early detection and pre-leukemia to leukemia development
 

© 2006-2025 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity