CSE 254: Seminar on Learning Algorithms

Time
TuTh 11-12.20 in CSE 4140

Instructor:
Sanjoy Dasgupta
Office hours TBA in CSE 4138

This quarter the theme of CSE 254 is deep unsupervised learning.

Class meetings will consist of student presentations. Each student will present a technical paper (or several papers) in detail. All seminar participants will discuss the paper and the issues raised by it.


Date Presenter
Paper/Topic
Slides
Tue Apr 9 Sanjoy Overview of class

 

Thu Apr 11 Sanjoy Overview of word embeddings; brief summary of word2vec

 

Tue Apr 16 Sanjoy Distributed representations of words and phrases and their compositionality
Linguistic regularities in sparse and explicit word representations
here
Thu Apr 18 Ronald Neural word embedding as implicit matrix factorization here
  Sanjoy Rand-Walk: A latent variable model approach to word embeddings here
Tue Apr 23 Bokan Generating sequences with recurrent neural networks
The unreasonable effectiveness of recurrent neural networks
The unreasonable effectiveness of character-level language models
here
  Jiaman On the computational power of neural nets here
Thu Apr 25 Kamran Neural machine translation by jointly learning to align and translate here
  Mehrab Attention is all you need here
Tue Apr 30 Casey Variational inference: A review for statisticians here
  Mary Anne Auto-encoding variational Bayes here
Thu May 2 Vignesh Generative adversarial nets here
  Hammad Abdullah Unsupervised representation learning with deep convolutional generative adversarial networks here
Tue May 14 Andrea Reducing the dimensionality of data with neural networks here
  Owen Stacked denoising autoencoders here
Thu May 16 Shixin NICE: Non-linear independent components estimation here
  Fangchen Connections with robust PCA and the role of emergent sparsity in VAEs here
Tue May 21 Dingcheng Black-box variational inference here
  Abhishek Wasserstein GAN here
Thu May 23 Zhifeng Generalization and equilibrium in generative adversarial nets here
  Shuo On GANs and GMMs here
Tue May 28 Tiancheng GAN dissection: Visualizing and understanding GANs here
  Robi Towards principled methods for training GANs here
Thu May 30 Qiaojun Self-supervised visual feature learning with deep neural networks: a survey here
  Chaitaya Unsupervised feature learning via non-parametric instance-level discrimination here
Thu Jun 6 Devendra Unsupervised image-to-image translation networks  
  Zhiyao Word translation without parallel data  
  Cheng Improved semantic representations from tree-structured LSTMs  
  An Temporal difference variational auto-encoder  

This is a four unit course in which the work consists of (1) oral presentations and (2) final projects.

Presentations (50% of grade)

The procedure for each student presentation is as follows: Please read, reflect upon, and follow these presentation guidelines, courtesy of Prof Charles Elkan.  Presentations will be evaluated, in a friendly way but with high standards, using this feedback form.

The schedule of presentations will be determined as much as possible during the first class.  Here is a preliminary list of papers.

If you want to change your presentation date, please arrange a swap with another student and notify me at least two weeks in advance.
 

Final projects (50% of grade)

Each student will pick a project related to the paper they present. This could, for instance, consist of: careful experiments on toy examples, designed to understand properties of algorithms from the paper; or a tutorial on the paper and all relevant background, pitched at a general audience; or a formulation of some mathematical questions around the paper, along with some preliminary analysis or experiments. There are two items to be turned in: