CS 672: Bayesian Deep Learning (Fall 2021)
|Time:||Friday, 1:00-4:00 pm|
|Office Hours:||Thursdays 3:00-4:00 pm or by appointment||Office Hour Link:||Please use this Zoom link.
Mask Requirement in Class
In order to protect the health and well-being of all members of the University community, masks must be worn by all persons on campus when in the presence of others (within six feet) and in buildings in non-private enclosed settings (e.g., common workspaces, workstations, meeting rooms, classrooms, etc.). Masks must be worn during class meetings; any student not wearing a mask will be asked to leave.
Masks should conform to CDC guidelines and should completely cover the nose and mouth: https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/about-face-coverings.html
Each day before you arrive on campus or leave your residence hall, you must complete the brief survey on the My Campus Pass symptom checker self-screening app.
A comprehensive artificial intelligence system needs to not only perceive the environment with different “senses” (e.g., seeing and hearing) but also infer the world’s conditional (or even causal) relations and corresponding uncertainty. The past decade has seen major advances in many perception tasks, such as visual object recognition and speech recognition, using deep learning models. For higher-level inference, however, probabilistic graphical models with their Bayesian nature are still more powerful and flexible. In recent years, Bayesian deep learning (BDL) has emerged as a unified probabilistic framework to tightly integrate deep learning and Bayesian models. In this general framework, the perception of text or images using deep learning can boost the performance of higher-level inference and, in turn, the feedback from the inference process is able to enhance the perception of text or images.
In this course, we will study the state-of-the-art methodology and theory in this emerging area, as well as applications including recommender systems, computer vision, natural language processing, graph learning, forecasting, healthcare, domain adaptation, speech recognition, etc.
Tips and More Details on Paper Presentations
Below are some tips and details.
Team Forming: Students can form a team of at most two. For example, if Alice and Bob form a team, their team will need to present two papers, each with a 60-min presentation.
Paper Reading: Read the paper many times to really understand the content in depth. Find any resources available (e.g., online talks, animation, demos) to help understand the paper. Read the key references recursively to gain better background knowledge. Discuss with your fellow students and friends on the paper.
Tips for Presentation: Here are some tips that can make the 60-min presentations smoother and more effective:
Note that this is a tentative syllabus to give you an idea of what topics this course will cover. This syllabus is subject to change as the course progresses.
|1||Sep 3||Course Introduction and Bayesian Deep Learning Basics||Read Bayesian Deep Learning Survey, Collaborative Deep Learning, Natural Parameter Networks||Slides 1, Slides 2|
|Probabilistic Neural Networks as Building Blocks|
|2||Sep 10||Variational Autoencoders (VAEs) and Applications||Read VAE, VAE for Reaction Prediction, Causal Effect VAE|
|3||Sep 17||Flow-based Models||Read Glow, Flow++|
|Concrete BDL Models with Applications|
|4||Sep 24||BDL for Computer Vision (1): Generation||Read Attend, Infer, Repeat (AIR), Diffusion Probabilistic Models|
|5||Oct 1||BDL for Computer Vision (2): Robustness||Read GenInt, ReverseAttack|
|6||Oct 8||BDL for Natural Language Processing||Read DEM-VAE, Sequence to Better Sequence|
|7||Oct 15||BDL for Healthcare||Read Sampling-free GRU, Neural Pharmacodynamic State Space Models|
|8||Oct 22||BDL for Control||Read Embed to Control, Deep Planning Network|
|9||Oct 29||BDL for Graphs||Read SBM Meets GNN, BetaE|
|10||Nov 5||BDL for Forecasting||Read Deep Factor Models, RNN with Particle Flow|
|11||Nov 12||BDL for Speech||Read Factorized Hierarchical VAE, Grad-TTS|
|12||Nov 19||BDL for Domain Adaptatoin and Meta Learning||Read Bayesian Domain-Invariant Learning, Probabilistic Meta-Learning|
|13||Nov 26||No Class (Thanksgiving Recess)|
|Mini Research Conference: Final Project Presentation|
|14||Dec 3||Final Project Presentation|
|15||Dec 10||Final Project Presentation|
Rutgers CS Diversity and Inclusion Statement
Rutgers Computer Science Department is committed to creating a consciously anti-racist, inclusive community that welcomes diversity in various dimensions (e.g., race, national origin, gender, sexuality, disability status, class, or religious beliefs). We will not tolerate micro-aggressions and discrimination that creates a hostile atmosphere in the class and/or threatens the well-being of our students. We will continuously strive to create a safe learning environment that allows for the open exchange of ideas while also ensuring equitable opportunities and respect for all of us. Our goal is to maintain an environment where students, staff, and faculty can contribute without the fear of ridicule or intolerant or offensive language. If you witness or experience racism, discrimination micro-aggressions, or other offensive behavior, you are encouraged to bring it to the attention to the undergraduate program director, the graduate program director, or the department chair. You can also report it to the Bias Incident Reporting System.