报告人 | 第三届青年论坛邀请了谁？
关键词：青年论坛 招聘
北京大学前沿计算研究中心第三届青年论坛将于2019年4月4日在北京大学举办，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丁香柳翠燕园，纵论学术前沿热点，探讨未来科技发展。
特邀报告人在此
AI + Industries: Applying AI to the Real World
张正友 Zhengyou Zhang
Biography
Zhengyou Zhang is the Director of Tencent AI Lab and Tencent Robotics X Lab since March 2018. He is an ACM Fellow and an IEEE Fellow. He was a Partner Research Manager with Microsoft Research, Redmond, WA, USA, for 20 years.. Before joining Microsoft Research in March 1998, he was a Senior Research Scientist with INRIA (French National Institute for Research in Computer Science and Control), France, for 11 years. In 1996-1997, he spent a one-year sabbatical as an Invited Researcher with the Advanced Telecommunications Research Institute International (ATR), Kyoto, Japan. He received the IEEE Helmholtz Test of Time Award in 2013 for his work published in 1999 on camera calibration, now known as Zhang's method.
Learning Neural Character Controllers from Motion Capture Data
Taku Komura
Biography
Taku Komura is a Reader (associate professor) at the Institute of Perception, Action and Behavior, School of Informatics, University of Edinburgh. As the leader of the Computer Graphics and Visualization Unit his research has focused on data-driven character animation, physically-based character animation, crowd simulation, cloth animation, anatomy-based modelling, and robotics. Recently, his main research interests have been the application of machine learning techniques for animation synthesis. He received the Royal Society Industry Fellowship (2014) and the Google AR/VR Research Award (2017).
青年报告人隆重推出
按姓名首字母排序，篇幅原因，部分内容有删减。
Online Timescale Theory for Concurrent Memory Allocation
李鹏程 Pengcheng Li
Abstract
This paper presents a new timescale theory to model the memory demand in real time. Using the new theory, an allocator can adjust its synchronization frequency using a single parameter called allocations per fetch (APF). The paper presents the on-line timescale theory, the design of the tunable APF allocator, and evaluation of its speed and memory efficiency. Compared with four state-of-the-art concurrent allocators, the new allocator improves the throughput of MongoDB by 55%, reduces the tail latency of a Web server by over 60%, and increases the speed of a selection of synthetic benchmarks by up to 24x while using the same amount of memory.
Biography
Dr. Pengcheng Li joined Google in Dec. 2016, mainly working on program analysis and optimization, google-wide profiling, tcmalloc, etc. Prior to this position, he has been working at the University of Rochester since 2012 fall for his Ph.D. studies. Dr. Li's research area focuses on memory management using timescale statistics, cache locality modeling, and program performance optimization. During Ph.D. period, he published 20 papers and patents, including 3 US patents. Before that, he worked at Baidu from 2010 to 2012 on load balancer systems. Dr. Li received his M.S. degree from Institute Computing Technology, Chinese Academy of Science in 2010 and his B.S. degree from University of Science and Technology of China in 2007.
Data-Driven Optimization for User Interaction in the Mobile Era: Information Access, Security and Beyond
刘雪晴 Xueqing Liu
Abstract
As mobile has become a big part of our everyday life, it is important to investigate how to optimize users' mobile interaction experience. In this talk, I will describe my research on automatically optimizing mobile user interactions with data-driven approaches. The talk focuses on my two approaches to address the deficiency in existing interactions: the first approach optimizes user interaction with faceted browsing by leveraging search log data, the second approach optimizes user interaction with security permission requests by leveraging mobile-appstore data. Finally, I will conclude the talk with multiple lines of my future work in the intersections of data mining, information retrieval, security, mobile computing, and software engineering.
Biography
Xueqing Liu is a Ph.D. candidate in Computer Science at the University of Illinois Urbana Champaign, where she works with Professors ChengXiang Zhai and Tao Xie. She received her B.S. degree from the Special Pilot CS Class (directed by Prof. Andrew Yao) at Tsinghua University. Her research interests are interdisciplinary with the general goal of leveraging "big data" to develop new algorithms for the improvement of user experiences. She has published in multiple topic areas including data mining, software engineering, security and information retrieval.
New Nonlinear Machine Learning Algorithms with Applications to Biomedical Data Science
王小倩 Xiaoqian Wang
Abstract
Recent advances in machine learning have spawned innovation and prosperity in various fields. In machine learning models, nonlinearity facilitates more flexibility and ability to better fit the data. However, the improved model flexibility is often accompanied by challenges such as overfitting, higher computational complexity, and less interpretability. Thus, my research has been focusing on designing new feasible nonlinear machine learning models to address the above different challenges posed by various data scales, and bringing new discoveries in both theory and applications. In this talk, I will introduce my newly designed nonlinear machine learning algorithms, such as additive models and deep learning methods, to address these challenges and validate the new models via the emerging biomedical applications.
Biography
Xiaoqian Wang is a Ph.D. candidate in Computer Engineering at the University of Pittsburgh. She received the B.S. degree in Bioinformatics from Zhejiang University in 2013. Her research interests span across multidisciplinary areas of machine learning, data mining, computational neuroscience, cancer genomics, and precision medicine. She has published 21 papers in top-tier conferences such as NIPS, ICML, KDD, IJCAI, AAAI, RECOMB, and ECCB. She received the Best Research Assistant Award at the University of Pittsburgh in 2017.
Dynamic Neural Networks for Efficient Learning and Inference
王欣 Xin Wang
Abstract
In this talk, I'm going to talk about our two recent works on the dynamic neural networks design. The first one is SkipNet, a modified residual network, that uses a gating network to selectively skip convolutional blocks based on the activations of the previous layer. We evaluate SkipNet on various benchmark datasets to show that it can reduce the runtime computational cost substantially without decreasing the prediction accuracy. The second one is our recent work on task-aware feature embedding networks (TAFE-Net), which learns how to adapt the image representation to a new task in a meta-learning fashion. We find that TAFE-Net outperforms previous approaches on several zero-shot learning benchmarks and the more challenging attribute-object composition task.
Biography
Xin Wang is currently a Ph.D. candidate in the Computer Science Department at UC Berkeley, advised by Professor Trevor Darrell and Joseph E. Gonzalez. She is part of the Berkeley AI Research (BAIR) Lab, Berkeley DeepDrive (BDD) Lab and the RISE Lab. Her research interest lies at the intersection of computer vision and learning systems with an emphasis on dynamic neural network designs for efficient learning and inference. She is also interested in interactive data analysis systems and low latency model serving systems. She is a key member on the web-based annotation platform, Scalabel, developed at BDD and a former member on the Clipper project, a low latency model serving system now maintained by RISE Lab. Prior to Berkeley, she obtained her bachelor's degree from Shanghai Jiao Tong University.
Location-Based Advertising in Vehicle Service Networks
余皓然 Haoran Yu
Abstract
Vehicle service providers can display commercial advertisements in their vehicles based on passengers' origins and destinations to create a new revenue stream. In this talk, we will consider a vehicle service provider who can generate different advertising revenues when displaying advertisements on different arcs. The provider needs to ensure the vehicle flow balance at each location, which makes it challenging to analyze the provider's vehicle assignment and pricing decisions for different arcs. Our key idea is to show that the traffic network corresponds to an electrical network. Through leveraging properties of the electrical network, we can derive the provider's optimal vehicle assignment and pricing decisions in the presence of advertising revenues. Furthermore, we will study the provider's optimal selection of advertisers when it can only display advertisements in the network for a limited number of advertisers.
Biography
Haoran Yu is a Post-Doctoral Fellow in the Department of Electrical and Computer Engineering at Northwestern University. He received the Ph.D. degree from the Chinese University of Hong Kong in 2016. He was a Visiting Student in the Yale Institute for Network Science during 2015-2016. His research interests lie in the field of network economics, with the current emphasis on the mechanism designs for sharing economy, mobile advertising, and location-based services. He has published papers in top-ranked journals and conferences in the area of networking. His paper in IEEE INFOCOM'16 was selected as a Best Paper Award finalist and one of top 5 papers from over 1600 submissions.
Better Algorithms and Generalization for Large-Scale Data
张泓洋 Hongyang Zhang
Abstract
Over the past decade, machine learning models such as deep neural nets have made lots of impact on a variety of tasks involving large-scale data. On the other hand, our understanding of when and why such ML models work are still very limited. Answering these questions often require better understanding of the latent structures of the data, as well as better understanding of the optimization paradigm used in practice. My research aims to provide theoretical foundations and better algorithms to this emerging domain.
This talk will show a few results. First, we study non-convex methods and their generalization performance (or sample efficiency) for common ML tasks. We consider over-parameterized models such as matrix and tensor factorizations. Our result highlights the role of the algorithm in explaining generalization for over-parameterized models, and the benefit to optimization by over-parameterization. Next, we consider the problem of predicting the missing entries of tensor data. We show that understanding generalization can inform the choice of the right model. Lastly, we revisit the shortest path querying problem on large graphs. We provide new algorithms to this classic problem by formalizing the structures of social network data.
Biography
Hongyang Zhang is a Ph.D. candidate in CS at Stanford University, co-advised by Ashish Goel and Greg Valiant. His research interests lie in machine learning and algorithms, including topics such as neural networks, non-convex optimization, social network analysis and game theory. He is a recipient of the best paper award at COLT'18.
Cross Modal Self-supervised Learning: Computer Vision+X
赵行 Hang Zhao
Abstract
Vision has become the major sensory input of many systems and robots in the past few years. Efforts in camera development, dataset annotation, and model design have greatly advanced the frontiers of computer vision. However, many other sensory data are still under-explored, e.g. sounds, thermal, RF, point clouds. In this talk, I will introduce a cross modal self-supervised learning paradigm, to show how we can use our achievements in computer vision to assist the development of other sensory modalities. Such learning paradigm could be solutions for problems suffering from scarcity of annotations. And we envision it to become the major learning scheme of future robots that are equipped with increasing number of sensors.
Biography
Hang Zhao is a Ph.D. candidate at MIT CSAIL, supervised by Professor Antonio Torralba. His research focuses on computer vision and cross modal learning. Before that, he got his Bachelor's degree from Zhejiang University in 2013. He is a 2019 Snap Research Fellow, and has interned at NVIDIA in 2015, MERL in 2016 and Facebook in 2017. His recent work on cross modal self-supervised learning was widely covered by media such as BBC, NBC, MIT News.
论坛最多还能再容纳1-2个报告，3月20日报名截止，抓紧申请！
听众报名
听众参会申请：2019年3月28日前，扫描下方小程序码报名
听众参会确认：2019年3月30日前
论坛时间：2019年4月4日
联系人：杨老师 cfcs@pku.edu.cn
想直接申请中心职位？招聘广告直通车——
（广告长期有效）
往 届 论 坛 回 顾
— 版权声明 —
本微信公众号所有内容，由北京大学前沿计算研究中心微信自身创作、收集的文字、图片和音视频资料，版权属北京大学前沿计算研究中心微信所有；从公开渠道收集、整理及授权转载的文字、图片和音视频资料，版权属原作者。本公众号内容原作者如不愿意在本号刊登内容，请及时通知本号，予以删除。