This is a 1-credit seminar class in which we will read and discuss papers on "deep learning". We will focus on critical view of such models, reading papers that highlight their limitations and focus on the societal implications of these moreso than technical content. This is a great way to learn about active topics of research in deep learning.
Students will take turns presenting a paper each week, which we will then discuss together critically during class meetings. Students should also post questions and/or reactions/comments about readings to Piazza prior to class.
Students will rotate responsibility for presenting (one of the) assigned papers and leading the ensuing discussion. The presentation should provide necessary background, the core contribution of the paper, (perceived) strengths and weaknesses, and ideas for improving the work.
Additionally, prior to every class (for which there are papers listed), all students are to write brief summaries ("reviews") of the paper(s) to be discussed that week. These can be short; enough to communicate the gist of the paper(s), and to raise any points one would like to make about it. In addition, you must raise at least one question about each paper for a given week -- these will be discussed (selectively) during meeting, but I will not reveal who asked what. This is mandatory. The idea is basically to ensure robust discussion during class period. Grades will reflect the consistency and quality of the execution of these activities. Summaries and questions must be submitted via blackboard (as one uploaded file).
Given the nature of this class, attendence is mandatory.
| Meeting | Paper(s) | Lead/Presenter (tentative, in some cases!) |
| 09/04 W | Introductions, aims, expectations, logistics, etc. | |
| 09/11 |
An overview of deep learning Deep Learning. LeCun, Bengio, Hinton, 2015 | Byron |
| 09/18 |
Representation learning Representation Learning: A Review and New Perspectives, Bengio et al., 2012. | Casey Pancoast and Levi Kaplan |
| 09/25 |
Representations of words (word2vec) Efficient Estimation of Word Representations in Vector Space, Mikolov et al., 2013. | Duk Hwan Kim |
| 10/02 |
Bias in word representations Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings, Caliskan et al., 2016 Semantics derived automatically from language corpora contain human-like biases, Caliskan et al., 2017 | Iain Methe, Sarah Wessel |
| 10/09 |
Bias in sentence encoders, too On Measuring Social Biases in Sentence Encoders | Daniel Melcer |
| 10/16 |
Bias in face classification Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification, Joy Buolamwini, Timnit Gebru, 2018 Turning a Blind Eye: Explicit Removal of Biases and Variation from Deep Neural Network Embeddings | Skye Nygaard, Theresa Nina Pelosi |
| 10/23 |
Reducing representation bias in vision / cognitive psych (?!) REPAIR: Removing Representation Bias by Dataset Resampling, Li and Vasconcelos, 2019 Cognitive psychology for deep neural networks: A shape bias case study, Ritter et al., 2017 | Eric Lehman, Garrett Tucker |
| 10/30 |
Fairness Learning Fair Representations, Zemel et al., 2013 Learning Adversarially Fair and Transferable Representations, Madras et al., 2018 | Shane Timmerman, Michael Delmonaco |
| 11/06 |
Interpretable Machine Learning Towards A Rigorous Science of Interpretable Machine Learning, Doshi-Velez and Kim, 2017. The Mythos of Model Interpretability, Lipton 2016. | An Vu, Charles Denhart |
| 11/13 |
Robustness Intriguing properties of neural networks, Szegedy et al., 2013 Adversarial Examples Are Not Bugs, They Are Features (blogpost), Engstrom et al., 2019 | Cameron Bracco, Raymond Huang |
| 11/20 |
Limitations of deep learning Deep Learning: A Critical Appraisal, Marcus, 2016 | Joshua Rodriguez |
| 11/27 | Thanksgiving recess | |
| 12/04 |
Defining and measuring intelligence The Measure of Intelligence, Chollet, 2019; Part I and II | Charles Forrester Welch |