CS291K - Advanced Deep Learning home | schedule


1. No class on May 7

Abstract: This is a graduate-level advanced course on deep learning.  We will discuss/present newest and important publications on deep learning, specifically in natural language processing (main focus) and computer vision.

Each student is expected to read papers before lecture, write paper reviews,  and complete a course project (e.g., implement an existing algorithm or solve a new problem creatively using deep learning. One team could have two students).  Projects that simply apply CNN and LSTM are not encouraged.

Prerequisites: Neural network building experience or CS292F/CS291A: Deep Learning for NLP or successfully finished an online deep learning course

Enrollment Code: 08813 Instructor: Prof. Xifeng Yan , Email: xyan at cs.ucsb.edu

Time: Monday/Wednesday 1:00- 2:45pm, Location: PHELP 3526   Office Hour:  Monday 3:00-4:00pm (HFH 1111)

Make-Up Lecture Time: Friday 1:00- 2:45pm, Location: PHELP 3526  (Due to meeting/travelling, I will frequently use this slot for make-up lectures, you are required to attend.

TA: N/A.  Research Project Coordinator: Xiaoyong Jin (jxy9426 at gmail dot com);  PoQaa Website Coordinator: Keqian Li (keqianlicg at gmail dot com)

Grading: Your grade will be derived from paper reading (30%),  paper review (10%),  active participation (10%), project presentation (15%), project quality (20%), project report (15%)  + Extra Credit 5% per paper presentation volunteered by you.

Paper Reading: You are required to read every paper carefully before lecture. We will randomly select six lectures and give a 30-minute quiz in each lecture.  Each quiz has a few questions to check if you have read the paper or not.  Please bring your notebook every lecture.  If you forget, you are allowed to handwrite your answer.

Paper Review: Find two deep learning related papers that are not presented in lecture and write a detailed 2-page review for each of them.  Your review shall include your understanding of the work, discuss its strength and weakness, and mostly important, propose new ideas to improve it.   The first review is due on April 30;  the second one is due on May 23.  A review will be graded by the quality of the paper (20%) and the quality of your review (80%).

Twitter Following: We created a few twitter accounts (poqaa_ai, poqaa_cv, poqaa_nlp, and poqaa_cs) to track the frequently mentioned papers in the Twitter space (arxiv.org paper only).  You are welcome to follow and find valuable papers to review.  You can also check the Poqaa  Trending (same to the above accounts),  and Reddit.com machine learning forum to find good papers.  You are also welcome to join us to improve the automatic discovery of important papers mentioned in social media.

Text Books (not required, but you'd better read it)

Deep Learning, An MIT Press book, Ian Goodfellow and Yoshua Bengio and Aaron Courville

Lecture notes (will be available later)