Deep Learning for Artificial Intelligence
Master Course at Universitat Politècnica de Catalunya (Autumn 2017)
Deep learning technologies are at the core of the current revolution in artificial intelligence for multimedia data analysis. The convergence of large-scale annotated datasets and affordable GPU hardware has allowed the training of neural networks for data analysis tasks which were previously addressed with hand-crafted features. Architectures such as convolutional neural networks, recurrent neural networks or Q-nets for reinforcement learning have shaped a brand new scenario in signal processing. This course will cover the basic principles of deep learning from both an algorithmic and computational perspectives.
Lectures - Room: D5-010
Course will be divided in modules of one hour covering the following topics (draft):
- 19/09 D01L1 (XG) Welcome
- 19/09 D01L2 (XG) Perceptron [PDF] [Slideshare] [Video]
- 26/09 D02L1 (ES) Multi-Layer Perceptron [PDF] [Slideshare] [Video]
- 17/10 D03L1 (ES) Backpropagation [PDF] [Slideshare]
- 17/10 D03L2 (JT) Supercomputing [Slideshare]
- 24/10 D04L1 (VV) Optimizers [PDF] [Slideshare]
- 24/10 D04L2 (JR) Loss functions [PDF] [Slideshare]
- 31/10 D05L1 (VV) Convolutional Neural Networks [PDF] [Slideshare]
- 31/10 D05L2 (RM) Transfer learning [PDF] [Slideshare]
- 07/11 D06L1 (RM) Incremental learning [PDF] [Slideshare]
- 07/11 D06L2 (JR) Methodology [PDF] [Slideshare]
- 14/11 D07L1 (MC) Recurrent Neural Networks [PDF] [Slideshare] [Video]
- 14/11 D07L2 (XG) Reinforcement learning [PDF] [Slides] [Video]
- 20/11 (Public) Oriol Vinyals (Google Deepmind) & Yannins Kalantidis (Facebook Research) - RSVP here
- 21/11 D08L1 (MC) Attention-based models [PDF] [Slideshare] [Video (2nd part)]
- 28/11 D09L1 (XG) Unsupervised & Predictive Learning [PDF] [Slideshare] [Video]
- 28/11 D09L2 (SP) Generative models: PixelCNN, WaveNet, VAE (SP) [PDF] [Slideshare] [Video]
- 05/12 D10L1 (SP Generative Models: GANs (SP)[PDF] [Slideshare] [Video]
- 19/12 (Public) Cristian Canton (Facebook, Seattle, WA, USA) & Amaia Salvador (UPC) - RSVP here
- 22/12 (Public) Adrià Recasens (MIT), Lluís Castrejón (MILA) & Ramon Sanabria (CMU) - RSVP here
Labs (15%) - Rooms: D5-004, D5-005 & D5-007
The course will contain guided hands on lab that will lead the students in their first steps in deep learning frameworks. A summary can be found on this site by Prof. Jordi Torres.:
- 19/09: Getting started (JT)
- 26/09: Visit to MareNostrum at Barcelona Supercomputing Center (BSC) (JT)
- 24/10: First steps with Keras and TensorBoard(JT)
- 31/10: First steps with PyTorch (JT)
- (home) First steps with TensorFlow
- 05/12: Generative Adversarial Networks (SP)
Project (40%) - Rooms: D5-004, D5-005 & D5-007
Students will work in teams to develop a machine learning research project that will be presented both in an oral presentation and as a poster during the final session open to the general public.
- 26/09: Guidelines for Project (XG)
- 17/10: Project Proposals (XG)
- 07/11: Project Development (XG, AC, NC)
- 14/11: Project Critical Review (XG, NC, AC)
- 21/11: Project Development (XG, NC, AC)
- 12/12: Project oral presentations (XG, NC, ES, AC)
- 19/12: Poster presentations (open session) - RSVP here
Project pages with source code, slides and contact information for recruiters:
- Study Programs: Master MET at ETSETB TelecomBCN and Masters FIB, from the Universitat Politecnica de Catalunya.
- Course code and official guide: 230706 - DLAI
- ECTS credits: 5 ECTS
- Teaching language: English
- Semester: Fall 2017
- Class Schedule: Tuesdays 3-6pm (3-5pm Lectures, 5-6pm Lab)
- Capacity: 40 MSc students
- Location: Campus Nord UPC, Module D5, Room 010
Registration procedure depends on the student profile:
- Master students at ETSETB and FIB: Follow the regular schedule from your academic office.
- Mobility students: If your host institution has signed an agreement with UPC ETSETB Telecom BCN school, you can request a mobility from your host institution and sign up for the course under the same conditions as ETSETB students.
- Students at UPC but not in ETSETB: Contact the your academic office and request being allowed to take this course. If accepted, contact ETSETB academic office and request more details.
- Non UPC nor mobility students: You must apply for being accepted in the course and cover the 100% cost of the ECTS credits, without the support of the public funds. This corresponds to 143,08 € per ECTS credit (Summer 2016). If you are interested in this option, please contact the ETSETB Telecom BCN academic office, with an e-mail to firstname.lastname@example.org or calling at 93 405 4174 / 93 401 6772 / 93 401 5966 or 93 401 6750 in the morning (Monday to Thursday from 11 to 14 and Fridays from 11 to 13) or noons (Wednesdays and Thursdays from 16 to 17h).
- Deep Learning for Computer Vision UPC TelecomBCN: 2016 2017
- Deep Learning for Speech and Language. UPC TelecomBCN 2017.
- Deep Learning for Multimedia. Insight Dublin City University 2017.
- Amaia Salvador and Santiago Pascual. “Hands on Keras and TensorFlow”. Persontyle 2017.
- Xavier Giro-i-Nieto, “Deep learning for computer vision: Image, Object, Videos Analytics and Beyond”. LaSalle URL. May 2016.