Timezone: »

 
c-MBA: Adversarial Attack for Cooperative MARL Using Learned Dynamics Model
Nhan H Pham · Lam Nguyen · Jie Chen · Thanh Lam Hoang · Subhro Das · Lily Weng

@
in Workshop on Machine Learning Safety »

In recent years, a proliferation of methods were developed for cooperative multi-agent reinforcement learning (c-MARL). However, the robustness of c-MARL agents against adversarial attacks has been rarely explored. In this paper, we propose to evaluate the robustness of c-MARL agents via a model-based approach, named \textbf{c-MBA}. Our proposed attack can craft much stronger adversarial state perturbations of c-MARL agents to lower total team rewards than existing model-free approaches. Our numerical experiments on two representative MARL benchmarks illustrate the advantage of our approach over other baselines: our model-based attack consistently outperforms other baselines in all tested environments.

[ Poster]  [ Topia] 

Author Information

Nhan H Pham (IBM Research)

I started my PhD in Operations Research in Department of Statistics and Operations Research at University of North Carolina at Chapel Hill in 2017. I am currently working on stochastic methods for non-convex optimization with applications in machine learning, deep learning, and reinforcement learning under supervision by Dr. Quoc Tran-Dinh. In addition, I am also collaborating with Dr. Lam M. Nguyen and Dr. Dzung T. Phan at IBM Thomas J. Watson Research Center. I come from Vietnam where I had my bachelor in Computer Engineering from Deparment of Computer Science and Engineering, Ho Chi Minh City University of Technology (Bach Khoa University). During my undergrad, I was a member of BKIT Hardware Club and participated in the Vietnam Robot Contest under BK4/BKIT Number One team in 2013. My hobbies are travelling with my wife and exploring new places.

Lam Nguyen (IBM Research, Thomas J. Watson Research Center)
Jie Chen (MIT-IBM Watson AI Lab, IBM Research)
Thanh Lam Hoang (IBM Research)
Subhro Das (MIT-IBM Watson AI Lab)

Subhro Das is a Research Staff Member at the MIT-IBM Watson AI Lab in IBM Research Cambridge. He is also a Research Affiliate at MIT, co-leading IBM’s engagement in the Bridge pillar of MIT Quest for Intelligence. He serves as the Co-Chair of the AI Learning Professional Interest Community (PIC) at IBM Research. His research interests are in distributed learning over multi-agent networks, dynamical systems, multi-agent reinforcement learning, accelerated & adaptive optimization methods, and online learning in non-stationary environments – broadly in the areas of AI and machine learning with applications in healthcare and shared prosperity. Before moving to Cambridge, he was a Research Scientist at the IBM T.J. Watson Research Center, New York. Therein, he developed signal processing and machine learning based predictive algorithms for biomedical and healthcare applications. He received PhD and MS degrees in Electrical and Computer Engineering from Carnegie Mellon University in 2016 and 2014, respectively. His dissertation research was in distributed filtering and prediction of time-varying random fields and he was advised by Prof. José M. F. Moura. He completed his Bachelors (B.Tech.) degree in Electronics & Electrical Communication Engineering from Indian Institute of Technology Kharagpur in 2011. During the summers of 2009, 2010 and 2015, he was an intern at Ulm University (Germany), Gwangju Institute of Science & Technology (South Korea), and, Bosch Research (Palo Alto, CA), respectively.

Lily Weng (UCSD)

More from the Same Authors