Computing Research Week - Open House 2024

Program Details

Wednesday, 28/2/2024, 10:40 – 11:10 Towards Scalable Game Learning and Solving - Chun Kai LING

Game Theory is a cornerstone of multiagent systems, with many exciting breakthroughs ranging from superhuman performance in video and board games to societal applications such as airport security and wildlife poaching prevention. In this talk, we will first discuss two challenges faced when applying game theoretic approaches to real-world problems, (i) reasoning about games with unknown parameters, and (ii) efficiently solving large general-sum games. We then introduce methods to overcome these challenges and describe some exciting opportunities moving forward.

Bio: Chun Kai LING is a Postdoctoral Research Scientist at Columbia University working with Professors Christian Kroer and Garud Iyengar. He completed his PhD at Carnegie Mellon University under the supervision of Zico Kolter and Fei Fang. He is interested in Computational Game Theory and Multiagent Systems, with a focus on applying machine learning and optimization to learn and solve large games. He is the recipient of the IJCAI 2018 distinguished paper and GameSec 2023 best paper awards. Prior to starting his PhD, he completed his undergraduate studies in the National University of Singapore and worked in DSO National Laboratories.

Wednesday, 28/2/2024, 11:10 – 11:40 AI hardware powering next-generation wearables - PEH Li Shiuan

The talk will give an overview of my group's research into AI hardware and the next-generation wearable systems that such AI hardware enable.

Bio: Peh Li Shiuan joined NUS as Provost’s Chair Professor in the Department of Computer Science, with a courtesy appointment in the Department of Electrical and Computer Engineering in September 2016. Previously, she was Professor of Electrical Engineering and Computer Science at MIT and was on the faculty of MIT since 2009. She was also the Associate Director for Outreach of the Singapore-MIT Alliance of Research & Technology (SMART) from 2015-2016. Prior to MIT, she was on the faculty of Princeton University from 2002. She graduated with a Ph.D. in Computer Science from Stanford University in 2001, and a B.S. in Computer Science from the National University of Singapore in 1995. Her research focuses on networked computing, in many-core chips as well as mobile wearable systems. Her awards include 3 test-of-time paper awards (ACM SIGMOBILE 2017 for ASPLOS 2002 paper, IEEE TCCA 2020 for HPCA 2001 paper, ACM SIGMICRO 2023 for MICRO 2002 paper), IEEE Fellow in 2017, NRF Returning Singaporean Scientist Award in 2016, ACM Distinguished Scientist Award in 2011, MICRO Hall of Fame in 2011, CRA Anita Borg Early Career Award in 2007, Sloan Research Fellowship in 2006, and the NSF CAREER award in 2003.

Wednesday, 28/2/2024, 11:40 – 12:10 How Search Technology Breeds Illegal Transactions: Empirical Evidence from the Darknet - LU Ying

Governments and law enforcement agents have long been troubled by the rapid development of online illegal transaction platforms and influential leaders amidst them. In this paper, we will causally examine the effects of introducing a search-cost-reduction technology on darknet transactions to provide insights for police enforcement. When search cost is reduced on legal markets, either a long tail or an intensified concentration can arise. How search technologies might influence darknet market structure is yet unexplored in previous literature. Our paper is motivated to answer this question. Using a Difference-in-Differences method, we found that leading vendors reap more benefits by experiencing a larger-scale transaction increase relative to small vendors. Such an increased concentration structure can be explained by the trustworthiness prioritized by consumers in the illicit environment, as evidenced in our analyses regarding high-risk and obscure drugs. This study contributes to search literature and sheds light on selective targeting strategies for law enforcement agencies and market regulators.

Bio: I am a Year-5 PhD candidate from Department of Information Systems and Analytics, National University of Singapore. Before joining the PhD program, I obtained my master’s degree from Renmin University of China and worked for one year in Hong Kong University. My research focuses on understanding profit concentration driven by various information technologies in unique and important contexts. I recognize my future role as an advisor to offer suggestions towards better technology deployment with fair profit distribution. My teaching experience includes business analytic modules designed for students with different knowledge backgrounds.

Wednesday, 28/2/2024, 14:00 – 14:30 Data, systems, AI and beyond - Yao LU

AI and large generative models have garnered significant attention in research and production. I will provide a short overview of my previous work at the intersection of AI and systems, and discuss how these topics evolve to intersect with different tiers of the technology landscape, including data, core systems, advanced cloud infrastructures, and a variety of AI applications.

Bio: Yao Lu received PhD from the University of Washington. After that, he spent 5 years as a researcher in the Data Systems group at Microsoft Research Redmond lab. He is joining NUS SoC this summer as a tenure-track faculty member.

Wednesday, 28/2/2024, 14:30 – 15:00 Generative Modeling meets Robotics - Harold SOH

This talk will present our lab's advancements in generative modeling for robotics, focusing on two recent works.

First, we'll discuss diffusion models for learning complex, multimodal behaviors from demonstrations. These models traditionally start from Gaussian noise, facing challenges with limited data and a need for rapid inference. We'll discuss a new approach, called BRIDGeR, which addresses these issues by initiating from a more informative source than Gaussian noise, allowing for more effective learning. BRIDGeR offers a versatile solution to imitation learning, outperforming existing state-of-the-art diffusion models in challenging benchmarks.

If time permits, we tackle the problem of physical understanding in robotic manipulation and discuss how to provide LLMs with the "sense of touch". We introduce Octopi, a large-language model capable of recognizing tactile signals from a Gelsight sensor. Octopi combines tactile representation learning with vision-language models to enhance physical reasoning in visually ambiguous scenarios. Our evaluation across various tasks demonstrates Octopi's ability to significantly improve physical understanding of everyday objects.

Bio: Harold Soh is an Assistant Professor of Computer Science at the National University of Singapore, where he leads the Collaborative Learning and Adaptive Robots (CLeAR) lab. He completed his Ph.D. at Imperial College London, focusing on online learning for assistive robots. Harold's research primarily involves machine learning, particularly generative modeling, and decision-making in trustworthy collaborative robots. His contributions have been recognized with a R:SS Early Career Spotlight in 2023, best paper awards at IROS'21 and T-AFFC'21, and several nominations (R:SS'18, HRI'18, RecSys'18, IROS'12). Harold has played significant roles in the HRI community, most recently as co-Program Chair of ACM/IEEE HRI'24. He is an Associate Editor for the ACM Transactions on Human Robot Interaction, Robotics Automation and Letters (RA-L), and the International Journal on Robotics Research (IJRR). He serves as Associate Director of the NUS AI Lab (NUSAIL) and is a Principal Investigator at the Smart Systems Institute. He is also a co-founder of TacnIQ, a startup developing intelligent e-skins.

Thursday, 29/2/2024, 09:00 – 09:30 Tracking microbial function using genome-resolved metagenomics and AI - Niranjan Nagarajan

We live in a microbial world estimated to contain more than a million species, and yet humanity’s adversarial relationship with microbes is shaped by a small fraction of pathogenic species and the pervasive use of antimicrobial agents. Efforts to eradicate microbes often have limited success, with disinfected environments being rapidly recolonized, and antibiotic treatment increasingly selecting for resistant pathogens. The global rise in antimicrobial resistance (AMR) rates for common pathogens (e.g. ESKAPE) is recognized as a pre-eminent threat to healthcare systems. As the range of effective antibiotics shrinks we approach a tipping point where no antibiotic works for a pathogen, putting at risk the lives of millions of vulnerable patients in hospitals worldwide. Already >1 million deaths/year are attributed to AMR, and by 2050 the UN projects that AMR will be responsible for more deaths every year than all cancers (>10 million deaths/year).

We need new approaches to track the transmission of antibiotic resistance across microbes and to understand how we can leverage ecological functions to reduce AMR reservoirs. We propose that the emerging field of genome-resolved metagenomics aided by long-read sequencing [1] can transform our ability to do microbial surveillance, and we showcase its application in tracking pathogens through hospital environments [2] as well as the gut microbiome [3]. In order to decipher how microbial communities assemble and can provide colonization resistance against pathogens, we have developed new AI/modelling approaches that can provide mechanistic insights based on high-throughput metagenomic datasets [4, 5]. Together with other data mining approaches [6], we are now leveraging these to understand how microbiomes recover from the impact of antibiotics and how new classes of biotherapeutics can be developed to prevent the spread of antimicrobial resistant pathogens.

Bertrand D et al. Hybrid metagenomic assembly enables high-resolution analysis of resistance determinants and mobile elements in human microbiomes. Nature Biotechnology 2019 Aug;37(8):937-944
Chng KR et al. Cartography of opportunistic pathogens and antibiotic resistance genes in a tertiary hospital environment. Nature Medicine 2020 Jun;26(6):941-951
Kang JTL et al. Long-term ecological and evolutionary dynamics in the gut microbiomes of carbapenemase-producing Enterobacteriaceae colonized subjects. Nature Microbiology 2022 Oct;7(10):1516-1524
Li C et al. An expectation-maximization algorithm enables accurate ecological modeling using longitudinal microbiome sequencing data. Microbiome 2019 Aug 22;7(1):118
Li C et al. BEEM-Static: Accurate inference of ecological interactions from cross-sectional microbiome data. PLoS Computational Biology 2021 Sep 8;17(9):e1009343
Chng KR et al. Metagenome-wide association analysis identifies microbial determinants of post-antibiotic ecological recovery in the gut. Nature Ecology & Evolution 2020 Sep;4(9):1256-1267. doi: 10.1038/s41559-020-1236-0

Bio: Dr. Nagarajan is Associate Director and Senior Group Leader in the Genome Institute of Singapore, and Associate Professor in the Department of Medicine and Department of Computer Science at the National University of Singapore. His research focuses on developing cutting edge genome analytic tools and using them to study the role of microbial communities in human health. His team conducts research at the interface of genetics, computer science and microbiology, in particular using a systems biology approach to understand host-microbiome-pathogen interactions in various disease conditions. Dr. Nagarajan received a B.A. in Computer Science and Mathematics from Ohio Wesleyan University in 2000, and a Ph.D. in Computer Science from Cornell University in 2006 (Advisor: Prof. Uri Keich). He did his postdoctoral work in the Center for Bioinformatics and Computational Biology at the University of Maryland working on problems in genome assembly and metagenomics (Advisor: Prof. Mihai Pop).

Thursday, 29/2/2024, 09:30 – 10:00 Ensuring that programs do what we want them to do - Ilya Sergey

In this high-level talk, I will provide the motivation for software verification using formal methods, outlining the key directions in this research area. We will discuss software systems that can be formally specified and verified, state-of-the-art approaches for formal reasoning about programs, and the unsolved research challenges.

Bio: Ilya Sergey is an Associate Professor at the School of Computing of National University of Singapore, where he leads the Verified Systems Engineering lab. Ilya got his PhD in Computer Science at KU Leuven (Belgium). Before joining NUS, he was a postdoctoral researcher at IMDEA Software Institute (Spain) and a faculty at University College London (UK). Prior to becoming an academic, he worked as a software developer at JetBrains. Ilya does research in programming language design and implementation, distributed systems, software verification, and program synthesis.

Thursday, 29/2/2024, 10:00 – 10:30 Information-Theoretic Methods in Data Science - Jonathan Scarlett

Information theory was introduced in 1948 as a mathematical framework for understanding data communication and compression, and has shaped the design of practical systems ever since. Recently, a modern perspective has emerged that information theory is not only a theory of communication, but in fact a far-reaching theory of data that can impact problems throughout the entire data processing pipeline, including data acquisition, inference, learning, and optimization. In this talk, I will give an overview of this viewpoint and outline a number of relevant techniques and results, focusing in particular on my research group's contributions to problems such as group testing, black-box optimization, and signal estimation with deep generative models.

Bio: Jonathan Scarlett is an assistant professor in the Department of Computer Science and Department of Mathematics, National University of Singapore. His research interests are in the areas of information theory, machine learning, signal processing, and high-dimensional statistics. He received the Singapore National Research Foundation (NRF) fellowship, and the NUS Presidential Young Professorship.

Previously, Jonathan received the B.Eng. degree in electrical engineering and the B.Sci. degree in computer science from the University of Melbourne, Australia. From October 2011 to August 2014, he was a Ph.D. student in the Signal Processing and Communications Group at the University of Cambridge. From September 2014 to September 2017, he was post-doctoral researcher at LIONS, EPFL.

Thursday, 29/2/2024, 11:00 – 11:30 Towards sticker form-factor computers: Tackling the energy challenge of wireless embedded systems - Ambuj Varshney

Wireless Embedded Systems, aka the Internet of Things, has seen rapid growth with forecasts of a trillion embedded devices to be deployed within our lifetimes. Yet the rapid growth has also posed several challenges, including ensuring their sustainable deployment at a large scale. One of the most prominent challenges with sustainable deployment is the energy challenge. It refers to the fact that today, the power consumption for performing wireless communication remains significantly higher than that of sensing and communication tasks. This talk describes some of our efforts in designing energy-efficient communication systems that balance communication energy costs with other device tasks. We describe how these efforts can eventually lead to computers that can be in a sticker form factor operating for significant periods on energy harvested from the ambient environment.

Bio: Ambuj Varshney is an Assistant Professor in the School of Computing of the National University of Singapore. He was trained as a postdoctoral scholar at the University of California, Berkeley. Even before, he graduated with a doctorate from Uppsala University, Sweden. For over a decade, his research interests have focused on embedded systems, wireless communication, and mobile computing. He has received numerous awards, including the 2019 ABB Research Award in Honour of Hubertus von Gruenberg for his doctoral dissertation with a 300,000 USD endowment hosted at the University of California, Berkeley. His mentored students have also won the prestigious ACM Student Research Competition held at MobiCom twice, specifically in the year 2017 in the graduate category and the year 2023 in the undergraduate category. His research has been funded through grants from the government and industry, including awards from companies like ABB and directed grants from companies like NCS.

Thursday, 29/2/2024, 11:30 – 12:00 Computationally Hard Problems in ML Security - Prateek Saxena

There is an ongoing race between alarming security attacks and corresponding defenses. Will the race ever end? Will the adversary win eventually (in the limit) or will the defender? We posit that we can derive answers to such questions by studying the intrinsic security properties of stochastic gradient descent, the de facto training process of modern ML systems. The essence of the battle boils down to some computational problems which appear to be computationally hard. The talk is set in the context of one application, data poisoning attacks, for which we present the first optimal attacks against statistically optimal defenses. The attacks exploit the inherent difficulty of solving a computational problem in high-dimensional vector spaces. Similar problems lie underneath other attacks and defenses of active interest in ML security research.

Bio: Prateek Saxena is an Associate Professor in the Computer Science Department at NUS. He works in computer security broadly. His ongoing research interests are in ML security, distributed systems security, hardware-assisted security, and secure automatic code translation. He has received several awards including the MIT TR35 Asia Award, NUS Young Investigator Award, Google Security and Privacy Award, and David. J. Sakrison Award for outstanding doctoral work at UC Berkeley. He has co-founded several startups in the past. Please see his recent works at: https://www.comp.nus.edu.sg/~prateeks/

Thursday, 29/2/2024, 12:00 – 12:30 Analyzing Privacy in Machine Learning using Robust Membership Inference Attacks - Reza Shokri

Machine learning algorithms risk leaking substantial information about their training data. This leakage allows users to potentially reconstruct sensitive details from model predictions or parameters. We aim to understand and explain how and why these algorithms leak information. Membership inference is a key method for empirically measuring the information leakage from machine learning algorithms about their training data, a concern addressed by differentially private algorithms. In this talk, I will explain the privacy risks in machine learning. I will also present the state-of-the-art membership inference method RMIA (Robust Membership Inference Attacks).

Bio: Reza Shokri is an Associate Professor of Computer Science at National University of Singapore, and a part-time researcher at Microsoft. His research focuses on data privacy and trustworthy machine learning. He is a recipient of the Asian Young Scientist Fellowship 2023, IEEE S&P Test-of-Time Award 2021 for quantifying location privacy, the Caspar Bowden Award for Outstanding Research in Privacy Enhancing Technologies in 2018 for analyzing privacy risks of machine learning algorithms and the design of membership inference attacks, a Best Paper Award at ACM FAccT 2023 for analyzing fairness in machine learning, and faculty research awards from NUS, VMWare, Meta, Intel, and Google. He obtained his PhD from EPFL.

Computing Research Week - Open House 2024

February 28 and 29

Registration

The event is free of charge. However, please register to help with the organization (to cater for coffee and lunch).

Program

Program Details

Research Group / PI Meeting

Additional Information for Visiting International Students

Travel to Singapore

Accommodation in Singapore

Commute in Singapore

Data roaming / WiFi

Contact

Organizers

Venue

All talks will be held in Multipurpose Hall 1 (COM3-01-26), in the NUS School of Computing, COM3 building. Poster presentation will be held in the atrium outside multipurpose hall 1.