Workshop on Foundations of Computational Science

On August 29-31, 2019 the Center of Mathematical Sciences and Applications will be hosting a workshop on Foundations of Computational Science. The workshop will be held in room G10 of the CMSA, located at 20 Garden Street, Cambridge, MA This workshop is organized by David Xianfeng Gu.

Please register here. 

Speakers:

Videos of the talks are contained in the Youtube playlist below. They can also be found through links in the schedule. 

Schedule: 

Thursday, August 29

Time Speaker Title/Abstract

8:30 – 9:30am

Breakfast

 

9:30 – 9:40am

Opening

 

9:40 – 10:05am

Tat Seng Chua

Video

Explainable AI and Multimedia Research

 

Abstract: AI as a concept has been around since the 1950’s. With the recent advancements in machine learning algorithms, and the availability of big data and large computing resources, the scene is set for AI to be used in many more systems and applications which will profoundly impact society. The current deep learning based AI systems are mostly in black box form and are often non-explainable. Though it has high performance, it is also known to make occasional fatal mistakes. This has limited the applications of AI, especially in mission critical applications.

 

In this talk, I will present the current state-of-the arts in explainable AI, which holds promise to helping humans better understand and interpret the decisions made by the black-box AI models. This is followed by our preliminary research on explainable recommendation, relation inference in videos, as well as leveraging prior domain knowledge, information theoretic principles, and adversarial algorithms to achieving explainable framework. I will also discuss future research towards quality, fairness and robustness of explainable AI. 

10:05 – 10:35am

Group Photo and Coffee Break

 

10:35 – 10:50am

Maosong Sun

Video

Deep Learning-based Chinese Language Computation at Tsinghua University: Progress and Challenges

10:50 – 11:05am

Minlie Huang

Video

Controllable Text Generation

11:05 – 11:20am

Jun Liu


Video

Statistics Meets Neural Networks

11:20 – 11:45am

Yike Guo

Video

Data Efficiency in Machine Learning

11:45 – 12:10pm

Zuowei Shen

Video

Deep Approximation via Deep Learning

 

Abstract: The primary task of many applications is approximating/estimating a function through samples drawn from a probability distribution on the input space. The deep approximation is to approximate a function by compositions of many layers of simple functions, that can be viewed as a series of nested feature extractors. The key idea of deep learning network is to convert layers of compositions to layers of tunable parameters that can be adjusted through a learning process, so that it achieves a good approximation with respect to the input data.  In this talk, we shall discuss mathematical foundation behind this new approach of approximation; how it differs from the classic approximation theory, and how this new theory can be applied to understand and design deep learning network.

12:10 – 1:45pm

Lunch

 

1:45 – 2:00pm

Wenwu Zhu

Video

Explainable media and network representation

2:00 – 2:15pm

Wee Sun Lee

Video

Neuralizing Algorithms

 

Abstract: Most interesting AI problems are computationally intractable to solve in the worst case. We argue that we should be solving AI problems for the typical or average case instead, and that machine learning provides a good set of tools to do so. We illustrate the approach on approximate inference algorithms on probabilistic graphical models using our recent works on factor graph neural networks and particle filter recurrent neural networks.

2:15 – 2:30pm

Jun Zhu

Video

Particle-based Inference for Bayesian deep learning 

2:30 – 3:00pm

Coffee Break

 

3:00 – 3:15pm

Yuanchun Shi

Video

From Human Action Data To User Input Intention

3:15 – 3:30pm

Ke Deng

Video

Understanding complicated patterns of Chinese texts with very weak training

3:30 – 4:00pm

David Gu

Video

A Geometric View to Optimal Transportation and Generative Adversarial Models

4:00 – 4:30pm

Donald Rubin

Video

Relevant Statistical Evaluations When Comparing Procedures for Analyzing Data

 

Friday, August 30

Time Speaker Title/Abstract

8:30 – 9:00am

Breakfast

 

9:00-9:25am

Qianxiao Li


Video

A mean-field optimal control formulation of deep learning

 

Abstract: In this talk, we discuss formulating, through a continuous-time approximation, deep supervised learning as a mean field optimal control problem. This allows us to derive necessary conditions for optimality in deep learning in the form of a mean-field Pontryagin’s maximum principle, as well as global characterizations of optimality using Hamilton-Jacobi Bellman equations. This forms a connection between deep learning on the one hand, and partial differential equations and the calculus of variations on the other. We also discuss interesting numerical algorithms and generalization estimates that can be derived from this viewpoint, as well as some results on function approximation using flows of dynamical systems.

9:25-10:15am

Sarah Adel Bargal

Video

Grounding Deep Models for Improved Decision Making

 

Abstract: Deep models are state-of-the-art for many computer vision tasks including object classification, action recognition, and captioning. As Artificial Intelligence systems that utilize deep models are becoming ubiquitous, it is becoming crucial to explain (ground) why they make certain decisions, and utilize such explanations (grounding) to further improve model performance. In this talk, I will present: (1) Frameworks in which grounding guides decision-making on the fly at test time by questioning whether the utilized evidence is ‘reasonable’, and during learning through the exploitation of new pathways in deep models. (2) A formulation that simultaneously grounds evidence in space and time, in a single pass, using top-down saliency. This visualizes the spatiotemporal cues that contribute to a deep recurrent neural network’s classification/captioning output. Based on these spatiotemporal cues, segments within a video that correspond with a specific action, or phrase from a caption, could be localized without explicitly optimizing/training for these tasks.    

10:15-10:40am

Xiaoqin Wang

Video

Encoding and decoding auditory information by the brain

10:40-11:00am

Coffee Break

 

11:00-11:15am

Hang Su


Video

Adversarial attacks in deep learning

11:15-11:30am

Bin Xu

Video

AI Practice for Gaokao: Knowledge Graph Construction for Chinese K12 Education

11:30 – 11:45am

Liu Hanzhong

Video

Penalized regression-adjusted average treatment effect estimates in randomized experiments

11:45 – 1:30pm

Lunch

 

1:30 – 2:15pm

Cengiz Pehlevan

Video

Building sensory representations through learning

2:15 – 3:00pm

Sergiy Verstyuk

Modeling economic time series using deep learning

3:00 – 3:30pm

Coffee

 

3:30 – 4:15pm

Xiao-Li Meng

Video

Artificial Bayesian Monte Carlo Integration: A Practical Resolution to the Bayesian (Normalizing Constant) Paradox

 

Abstract: Advances in Markov chain Monte Carlo in the past 30 years have made Bayesian analysis a routine practice. However, there is virtually no practice of performing Monte Carlo integration from the Bayesian perspective; indeed, this problem has earned the “paradox” label in the context of computing normalizing constants (Wasserman, 2013). We first use the modeling-what-we-ignore idea of Kong et al. (2003) to explain that the crux of the paradox is not with the likelihood theory, which is essentially the same as for a standard non-parametric probability/density estimation (Vardi, 1985); though via using group theory, it provides a richer framework for modeling the trade-off between statistical efficiency and computational efficiency. But there is a real Bayesian paradox: Bayesian analysis cannot be applied exactly for solving Bayesian computation, because to perform the exact Bayesian Monte Carlo integration would require more computation than needed to solve the original Monte Carlo problem. We then show that there is a practical resolution to this paradox using the profile likelihood obtained in Kong et al. (2006) and that this approximation is second-order valid asymptotically. We also investigate a more computationally efficient approximation via an artificial likelihood of Geyer (1994). This artificial likelihood approach is only first-order valid, but there is a computationally trivial adjustment to render its second-order validity. We demonstrate empirically the efficiency of these approximated Bayesian estimators, compared to the usual frequentist-based Monte Carlo estimators, such as bridge sampling estimators (Meng and Wong, 1996). [This is a joint work with Masatoshi Uehara.]

 

Saturday, August 31

Time Speaker Title/Abstract

8:30 – 9:00am

Breakfast

 

9:00 – 9:45am

Brian Kulis

Video

New Directions in Metric Learning

 

Abstract: Metric learning is a supervised machine learning problem concerned with learning a task-specific distance function from supervised data.  It has found numerous applications in problems such as similarity search, clustering, and ranking. Much of the foundational work in this area focused on the class of so-called Mahalanobis metrics, which may be viewed as Euclidean distances after linear transformations of the data.  This talk will describe two recent directions in metric learning: deep metric learning and divergence learning. The first replaces the linear transformations with the output of a neural network, while the second considers a broader class than Mahalanobis metrics. I will discuss some of my recent work along both of these fronts, as well as ongoing attempts to combine these approaches together using a novel framework called deep divergences.

9:45 – 10:30am

Justin Solomon

Video

Linking the Theory and Practice of Optimal Transport

 

Abstract:  Optimal transport is a theory linking probability to geometry, with applications across computer graphics, machine learning, and scientific computing.  While transport has long been recognized as a valuable theoretical tool, only recently have we developed the computational machinery needed to apply it to practical computational problems.  In this talk, I will discuss efforts with my students to scale up transport and related computations, showing that the best algorithm and model for this task depends on details of the application scenario.  In particular, we will consider settings in representation learning using entropically-regularized transport, Bayesian inference using semi-discrete transport, and graphics/PDE using dynamical Eulerian models.

10:30 – 11:00am

Coffee

 

11:00 – 11:45am

Mirac Suzgun

Towards Understanding the Limitations of Deep Learning Models for Language

11:45 – 1:15pm

Lunch

 

1:15pm – 2:00pm

Jiafeng Chen &

Suproteem Sarkar

Video

Robust and Extensible Deep Learning for Economic and Financial Applications

2:00 – 2:45pm

Scott Kominers

 

Related Posts