CS 395T - Deep learning seminar - Fall 2018

meets TTh 2:00 - 3:30pm in GDC 4.304

instructor Philipp Krähenbühl
email philkr (at) utexas.edu
office hours by appointment (send email)

email TBD
office TBD
TA hours TBD

Please use canvas for assignment questions.


Class overview

This class covers advanced topics in deep learning, ranging from optimization to computer vision, computer graphics and unsupervised feature learning, and touches on deep language models, as well as deep learning for games. This is meant to be a very interactive class for upper level students (MS or PhD). For every class we read two recent research papers (most no older than two years), which we will discuss in class.

Goals of the class

After this class you should be able to


To map percentages to letter grades we use the following python script

def grade(p):
  from math import floor
  if p < 60: return 'F'
  v = (100-p) / (10 + 1e-5)
  return chr(ord('A')+int(v)) + ['+','','','-'][int((v-floor(v))*4)]


Date Topic Papers Presenters Notes and due dates
Aug 29 Administrative and intro (Linear models)   Philipp  
Sep 3 no class (labor day)      
Sep 5 Gradient based optimization Large-scale machine learning with stochastic gradient descent, Bottou 2010 Philipp  
Sep 10 Deep networks and backpropagation Deep learning, LeCun, Bengio and Hinton 2015

Efficient BackProp, LeCun etal. 1998
TA paper selection Th Sep 10, 1am email TA
Sep 12 Dropout and batch normalization Dropout: A Simple Way to Prevent Neural Networks from Overfitting, Srivastava etal. 2014

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift, Ioffe etal. 2015
  project 1 out
Sep 17 Advanced optimization and initialization Adam: A Method for Stochastic Optimization, Kingma and Ba 2015

On the Convergence of Adam and Beyond, Reddi etal 2018
Sep 19 Convolutional Networks for image classification Gradient-based learning applied to document recognition, LeCun etal. 1998 (sec 4+ optional)

ImageNet Classification with Deep Convolutional Neural Networks, Krizhevsky etal. 2012
Sep 24 Convolutional Networks for object detection and pixel-wise prediction Rich feature hierarchies for accurate object detection and semantic segmentation, Girshick etal. 2014

Fully Convolutional Networks for Semantic Segmentation, Long etal 2015
Sep 26 Advanced deep network architectures Deep Residual Learning for Image Recognition, He etal. 2016

Densely Connected Convolutional Networks
  Project 1 QA
Oct 1 Visualizing deep networks Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps, Simonyan etal. 2014

Inverting visual representations with convolutional networks, Dosovitskiy 2016
Oct 3 Image generation Generative Adversarial Nets, Goodfellow etal. 2014

Wasserstein GAN, Arjovsky etal 2017
Oct 8 Project 1 presentations - Yearbook/Geolocation 8 min per team   Project 1 due 1am
Oct 10 Project 1 presentations - Yearbook/Geolocation 8 min per team   Project 2 out
Oct 15 Image generation II Auto-Encoding Variational Bayes, Klingma etal. 2014

Density estimation using Real NVP, Dinh etal 2016
Oct 17 Image translation High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs, Wang etal 2018

Semi-parametric Image Synthesis, Qi etal. 2018
Oct 22 Unpaired image translation Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks, Zhu etal. 2017
Project 2 flash presentations (2min)
Oct 24 Compression      
Oct 29 Adversarial attacks      
Oct 31 Recurrent models Generating Sequences With Recurrent Neural Networks, Graves 2013

Pixel Recurrent Neural Networks, Oord etal 2016
Nov 5 Language models Sequence to sequence learning with neural networks, Sutskever etal. 2014

Neural machine translation in linear time, Kalchbrenner etal. 2016
Nov 7 Video      
Nov 12 Atari games Playing Atari with Deep Reinforcement Learning, Mhin etal. 2013

Human-level control through deep reinforcement learning, Mhin etal. 2015
Nov 14 Video games      
Nov 19 Alpha GO Mastering the game of Go with deep neural networks and tree search, Silver etal. 2016 AGO-0   Project 2 - QA
Nov 21 no class (Thanksgiving)      
Nov 26 Not RL Learning to act by predicting the future, Dosovitskiy etal. 2016

Evolution Strategies as a Scalable Alternative to Reinforcement Learning, Salimans etal. 2017
Nov 28 Fresh off the press      
Dec 3 Final project presentations 7 min including questions   Project 2 due 1am
Dec 5 Final project presentations 7 min including questions    
Dec 10 Fresh off the press      

Paper assignment

All papers are assigned by a Deep Networks based on student preferences. If you don’t like your paper, please blame deep learning, tensorflow or your preference list (instead of the instructor). The code for the assignment problem will be make available.

Expected workload

Estimates of required effort to pass the class are:

General tips

Tips for reading/reviewing a paper

Tips for presentations


Syllabus subject to change.