【青年论坛】The 5th PKU CFCS Youth Forum Agenda

  北京大学前沿计算研究中心第五届青年论坛将于2020年4月5日通过线上方式举办,旨在为计算理论、人工智能等领域以及相关交叉领域的海内外青年学者提供高水平学术交流平台,吸引国际优秀人才加盟中心。同时,中心将邀请目前活跃在工业界的技术精英和领袖,分享当前国内相关领域最新技术。我们诚邀海内外优秀学者和校友相聚燕园,交流学术前沿热点,探讨未来科技发展。

 

直播说明

论坛将全程直播:http://live.bilibili.com/22051279

 

论坛日程

 

 

特邀报告人


Daniel Povey  Chief Voice Scientist, Xiaomi/Founder of Kaldi

 

Title:
The Challenge of Good Documentation for Machine Learning Software

 

Abstract:
A key factor in the success of software is the quality of its documentation. In this talk I will discuss some of the challenges of writing good documentation in the context of software for machine learning. I argue that a documentation-driven approach to the design of software can lead to clearer designs and easier-to-maintain code.

 

Biography:
Daniel Povey completed his PhD at Cambridge University in 2003.  He spent about ten years working for industry research labs (IBM Research and then Microsoft Research), and 7 years as non-tenure-track faculty at Johns Hopkins University; he moved to Beijing, China in November 2019 to join Xiaomi Corporation as Chief Voice Scientist.

 

He is best known as the principal author and maintainer of the open-source software Kaldi, which is the most popular computer speech recognition software toolkit; the technology of many (possibly most) companies that do speech recognition is based on it, including major corporations like Apple and Amazon.

 

He is also known for many different contributions to the technology of speech recognition; his papers have over 20,000 citations.

 

青年报告人 (按姓名首字母排序)


1、杜雅丽  University College London

 

Title:
Agent Learning in the Emergence of Complex World

 

Abstract:
Over the past few years, we have witnessed a great success of AI in many applications, including image classification, recommendation systems, etc. This success has shared a common paradigm in principle, learning from static datasets with inputs and outputs. Nowadays, we are experiencing a paradigm shift from pattern recognition to decision making.  Instead of learning the knowledge from static datasets, we are learning through the feedback of our knowledge. Especially since machine learning models are deployed in the real-world; these systems start having impacts on each other, turning their decision making into a multi-agent problem. Therefore, agent learning in a complex world is a fundamental problem for the next generation of AI to empower various multi-agent environments.

 

As case studies, we present GridNet, which can flexibly control an arbitrary number of agents, and LIIR, which generates diversified behaviors when receiving only a team reward from a cooperative multi-agent system. Our novel methods achieve the new state-of-the-art on the testbed of StarCraft II environments, which has recently emerged as a challenging RL benchmark task with high stochasticity, large state-action space, and delayed rewards. In the end, I will discuss some future directions in multi-agent learning.

 

Biography:

Dr. Yali Du is a postdoctoral research fellow at University College London and a visiting researcher at Huawei London Research Lab. She obtained her Ph.D. in AI in 2019 from University of Technology Sydney, under the supervision of Dacheng Tao. Her research interest lies in how to address multi-agent problems, including controlling with flexibility and diversity, the emergence of interaction, multi-agent credit assignment, and robustness of agent learning. Her research output has been widely published in prestigious venues including ICML, NeurIPS, IJCAI, ACM MM, IEEE TMM, etc.

 

2、黄棱潇  Yale University 

 

Title:
Fairness in Automated Decision-Making Tasks

 

Abstract:
Automated decision-making algorithms are increasingly deployed and affect people's lives significantly. Recently, there has been growing concern about systematically discriminate against minority groups of individuals that may exist in such algorithms. Thus, developing algorithms that are "fair" with respect to sensitive attributes has become an important problem.

 

In this talk, I will first introduce the motivation of "fairness" in real-world applications and how to model "fairness" in theory. Then I will present several recent progress in designing algorithms that maintain fairness requirements for automated decision-making tasks, including multiwinner voting, personalization, classification, and clustering.

 

Biography:

Dr. Lingxiao Huang is a postdoc of theoretical computer science in Yale University, where he is advised by Nisheeth K. Vishnoi and K. Sudhir. Before he was a postdoc in EPFL in 2017-2019, after received his Ph.D. in IIIS, Tsinghua University. His current research interest is algorithm design and computational social choice. He is passionate about creating novel algorithms that are motivated by existing practical challenges.

 

3、王趵翔  the Chinese University of Hong Kong

 

Title:
Improving Policy Optimization: Algorithms and Foundations

 

Abstract:

Reinforcement learning (RL) studies algorithmic approaches to optimize the policy in sequential decision processes. The recent success of RL in a variety of applications has demonstrated its usefulness but also leaves room for improvements. In this talk we discuss methods for variance reduction for high-dimensional action spaces, aiming to prevent the sample complexity from growing exponentially in the number of dimensions. The divide-and-conquer technique we used to achieve this is very general to be applied to other areas. Beyond these algorithmic studies we present our first step toward understanding sequential decisions, through a classic example of the Gambler's problem.

 

Biography:

Baoxiang Wang is a sixth-year PhD student at the Department of Computer Science and Engineering, The Chinese University of Hong Kong. He is advised by Siu On Chan and Andrej Bogdanov. During his PhD, he spent a year in Edmonton visiting a joint lab by University of Alberta and RBC Institute of Research. He obtained his bachelor's degree at the School of Information Security, Shanghai Jiao Tong University.  His research interest lies on reinforcement learning, online learning, and learning theory.

 

4、王鑫  University of California, Santa Barbara

 

Title:
Close the Loop Between Language and Vision for Embodied Agents

 

Abstract:

Humans learn to perceive the world through multiple modalities including visual, auditory, and kinesthetic stimuli. The need for perception is self-evident while humans invented language for communication and documentation. Therefore, language and perception lay foundations for artificial intelligence, and how to ground natural language onto real-world perception is a fundamental challenge to empower various practical applications that require human-machine communication.

 

In this talk, I will mainly present two of my research thrusts on developing intelligent embodied agents that connect language, vision, and actions, and that communicate with humans in the real world. First, moving beyond natural language understanding from text-only corpora, I have situated natural language inside interactive environments where communication often takes place (language—>vision). So I will discuss how to effectively ground natural language instructions and visual inputs to actions in real-world navigation tasks with reinforcement learning and imitation learning. Second, in order to enable an agent to describe the visual surroundings for humans (vision—>language), I will explore challenges of language generation conditioned on visual context, and present novel solutions from coarse-grained to fine-grained caption generation, and then to humanlike story generation. In the end, I will conclude with my future research plan.

 

Biography:

Xin Wang is a Ph.D. candidate at the University of California, Santa Barbara. His research interests include natural language processing, computer vision, and machine learning, especially the intersection of them. He works on fundamental research directions that enable intelligent embodied agents to communicate with humans in the real world. He published over 18 papers (including 7 oral presentations) at top-tier CV, NLP, and ML venues such as CVPR, ICCV, ECCV, ACL, NAACL, EMNLP, AAAI, TPAMI. He received the CVPR Best Student Paper Award in 2019. Xin is also professionally active and have organized multiple academic events on the topic of his research, including workshops at ACL 2020, CVPR 2020, and ICCV 2019, and a tutorial at AACL-IJCNLP 2020.  He also served as a session chair for the NLP session at AAAI 2019. He worked at Google AI, Facebook AI Research, Microsoft Research (Redmond), and Adobe Research.

 

5、叶南阳  University of Cambridge

 

Title:
Towards Robust Deep Learning

 

Abstract:

Although deep learning has almost become the default choice in many applications, applications, such as autonomous driving, face recognition. The lack of deep learning theory makes it dangerous to apply it for critical applications. We do not yet have a convincing theory on when will the deep learning works or fails. For example, we can easily fool the traffic sign recognition systems by putting some small white or black strips on the signs. And the face recognition system can easily fail if we wear specially designed glasses. The method to fail deep learning systems is usually referred to as the adversarial attack. This talk is about my research on improving the generalization and robustness of deep learning systems via the lens of Bayesian theory. Besides, I will also demonstrate the application of deep learning for visually lossless image compression.

 

Biography:

Dr. Nanyang Ye is a recently graduated PhD from University of Cambridge. His research interests include but not limited to the safety of machine learning, Bayesian deep learning, and deep learning application in computer graphics.

 

6、于乐全  Stanford University

 

Title:
Towards Intelligent Healthcare: Medical Image Analysis and Reconstruction with Deep Learning

 

Abstract:

Medical imaging is a critical step in modern healthcare procedures. Accurate interpretation of medical images, e.g., CT, MRI, Ultrasound, histology images, fundus images, and endoscopy videos, plays an essential role in computer-aided diagnosis, assessment, and therapy. While deep learning provides an avenue to deliver automated medical image analysis and reconstruction via data-driven representation learning, there remain a series of unique challenges towards intelligent medical imaging, such as high-dimensional data processing, insufficient and heterogeneity training data, and high annotation cost. In this talk, I will discuss our recent efforts on building intelligent systems for medical image analysis and reconstruction, such as anatomical structure segmentation, lesion detection, cancer diagnosis, and CT image reconstruction. The proposed methods cover a wide range of deep learning and machine learning topics, including network architecture design, advanced learning strategies, semi-supervised learning, domain adaptation, multi-modality learning, integrating domain knowledge, etc. The up-to-date progress and promising future directions of AI-powered healthcare will also be covered in this talk.

 

Biography:

Dr. Lequan Yu is a postdoctoral fellow in the Department of Radiation Oncology at Stanford University. He has received his Ph.D. degree in Computer Science and Engineering from The Chinese University of Hong Kong in 2019, and his B.Eng. degree in Computer Science and Technology from Zhejiang University in 2015. His current research lies at the intersection of medical imaging and artificial intelligence. He also has expertise in deep learning for 3D vision. He has published 20+ top-tier papers in this area on topics of medical image segmentation, biomarker detection, computer-aided diagnosis, semi-supervised learning, and domain adaptation. He also won the Best Paper Awards of Medical Image Analysis-MICCAI in 2017 and International Workshop on Machine Learning in Medical Imaging in MICCAI 2017.  He serves as the reviewer for several top-tier journals and conferences, including IEEE-TMI, IEEE-TIP, IEEE-TNNLS, IEEE-TBME, IEEE-TASE, MedIA, SIGGRAPH, CVPR, ICCV, ECCV, MICCAI, etc. His current Google Scholar citation has reached 2600+ with h-index 19.

 

7、袁骁  Stanford University

 

Title:
Error-mitigated Quantum Computing with Near-term Quantum Hardware

 

Abstract:

Realizing a universal quantum computer is challenging with current technology. Before having a fully-fledged quantum computer, a more practical question is what we can do with current and near-term quantum hardware. Focusing on the noisy-intermediate-scaled-quantum regime, we present variational quantum algorithms for solving static and dynamic problems of many-body physics. We show how to suppress device errors due to implementation imperfection on both digital and analog quantum computers. The algorithms are also applicable to other tasks, including quantum machine learning, quantum sensing, and quantum error correction. With the rapid development of quantum hardware, error-mitigated variational quantum algorithms may finally enable genuine quantum advantage in the noisy-intermediate-scaled quantum era.

 

Biography:

Dr. Xiao Yuan received his Bachelor in theoretical physics (major) and computer science (minor) from Peking University in 2012. He got his PhD in physics from Tsinghua University in 2016. Then he worked as a postdoc at Oxford University from 2017 to Sep. 2019. He is now a postdoc at Stanford University. Dr. Xiao Yuan's research interests span the wide spectrum of quantum information science, from fundamental quantum information to algorithms for near-term quantum computers.

相关附件: