Structure

This research seminar is intended to discuss recent advances and trends in machine learning. We will be presenting and discussing 1-2 recent related technical papers each week. The focus will be on modern techniques, ideas, and trends in machine learning. The aim is to understand the fundamental ideas, tricks, and concepts involved with the aim of using them in practice and stimulating research. The stress will be on reading and grasping maximum out of some seminal papers. Whenever necessary, some concepts will be introduced for clarification and to make connections.

This year we will start with basic SVMs, Kernels, and Random Features for Large Scale Learning. Then we will go from random features to learned features (a.k.a deep learning). Look into tricks and trades of how to train large-scale deep network. We will then switch to reinforcement learning and finally deep-reinforcement learning, which is where we will get some insight into the coolest AI engines of our times: the self-driving car and the Alpha-go system.

Grading and Logistics

Class participation (5 min quiz), one paper presentation, and one paper summarization for 1 credit. In addition students can undergo a semester long research project for 3 credits. There will be a quiz on the readings in the first 5 minutes. It is important to read the listed papers (as much as you can) before coming to the class.

Prerequisite

A rigorous course in machine learning is required. We will be discussing advanced papers in ML papers every week.

Piazza For Discussions

Please signup at: piazza page

Presentations and Scribe Logistics

Each student should sign up for 1 class to present (2-3 students per class) and 1 class to scribe the discussions(2-3 students per class). You cannot scribe the same class that you presented. You should give a dry run of your presentation to the instructor a week before the class in the office hours (or some other scheduled time). Several rounds of modification may be needed before a presentation is ready for the class, so make sure to schedule early. The scribe should be submitted no later than a week of the presentation.

Please sign-up for scribe and presentation assignment at Google Spreadsheet

Schedule

08/21 : Introduction, Logistics. slides
08/28 : SVMs, Kernels and Random Projections Class Cancelled Due to Hurricane

Support-Vector Networks pdf
Random Projection, Margins, Kernels, and Feature-Selection pdf
Optional : An Elementary Proof of a Theorem of Johnson and Lindenstrauss pdf

09/04 : Labor Day
09/11 : SVMs + Random Features for Large-Scale Learning (Learning with Millions or Higher Features)

Support-Vector Networks pdf
Random Features for Large-Scale Kernel Machines pdf
Hash Kernels pdf
Hashing Algorithms for Large Scale Learning pdf

09/18 : From Random Features to Learned Features (The Need for Deep Learning)

Scaling Learning Algorithms towards AI pdf
How to Scale Up Kernel Methods to Be As Good As Deep Neural Nets pdf

09/25 : Introduction to Different Deep-Learning Architectures (Training Old Pipelines End-to-End)

Convolution Network pdf
LSTMs pdf
Sequence to Sequence Learning with Neural Networks pdf

10/02 : Tranfer Learning and Multi-task learning (Can combine Pipelines. Deep-nets may be the future)

Multi-Task Learning pdf
One Model to Rule Them All pdf

10/03 : Ken Kennedy Institute Lecture Jeff Dean, Google Senior Fellow (4pm to 5pm)
10/09 : Midterm Recess: No Class (Look for Data Science Conference at Rice)
10/16 : Scaling up Deep Learning (Computation is the most valuble resource in AI. Use it wisely)

Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1
Compressing Neural Networks with the Hashing Trick
Scalable and Sustainable Deep Learning via Randomized Hashing

10/23 : Useful Tricks and Hacks for training deep networks (Some tricks always seem to work)

On the difficulty of training Recurrent Neural Networks pdf
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift pdf

10/30 : Adversarial Training (Interplay of Networks and Losses)

Generative Adversarial Networkspdf
EXPLAINING AND HARNESSING ADVERSARIAL EXAMPLES pdf

10/06 : Reinforcement Learning 1

11/13 : Deep Reinforcement Learning (Towards Human Like AI)

blog
Lecture

11/20 : Self Driving Cars and AI Games (AI Seems to Take Over)

End-to-End Learning for Self-Driving Cars pdf
Playing Poker pdf and related references

11/27 : Final Project Presentations

Students with Disability

If you have a documented disability that may affect academic performance, you should: 1) make sure this documentation is on file with Disability Support Services (Allen Center, Room 111 / adarice@rice.edu / x5841) to determine the accommodations you need; and 2) meet with me to discuss your accommodation needs.