1 of 13

CMA Core Meet & Greet�14 December 2023

Trustworthy AI/ML in Multi-agent Systems

​

Md Tamjid Hossain

PhD Candidate

Advanced Robotics and Automation (ARA) Lab

University of Nevada, Reno (UNR)

PI: Dr. Hung La, Associate Professor, UNR

2 of 13

Outline

  • Cooperative Multi-agent Systems (MASs)
  • Cooperative MASs + Trustworthy AI/ML
  • Adversarial Reinforcement Learning
  • Publications and Future Works

​

​

​

3 of 13

Cooperative Multi-Agent Systems (MASs)

Oral Mucosal Microbes1

Road Traffic: Merging2

Battlefield3

1000-Robot Swarm4

Starship Delivery Robots5,6

4 of 13

Cooperative Multi-Agent Systems (MASs)

Threats

MAS Network

Agent

5 of 13

Threat Outcomes

Traffic congestion

Incorrect Path Selection

Slow Action Learning

Missile Misfiring

6 of 13

Trustworthy AI/ML

Confidentiality

Integrity

Availability

Data SLAs

Denial-of-service (DoS)

Distributed Denial-of-service (DDoS)

Data Poisoning

False data injection

Data Privacy

Data Quality

Membership Inference

Eavesdropping

Info Assurance

private & accurate data

fast, accurate data

fast, private data

Scalable Extraction of Training Data from (Production) Language Models

Milad Nasr et al., Google deepmind

28 Nov 2023

7 of 13

Training

Decision

Execution

Malicious

Benign

Advisee

Advisors

Advice

Querying

Inferring

Reconstructing

Malicious

Benign

Advisee

Advisors

Query

’s state

’s next action

’s reward

’s policy

’s trajectory

 

Weighted Experience Aggregation

Adaptive Neighbor Selection

Best (Neighbor) Exp.

Self Exp.

Final Exp.

w

 

Noisy Exp.

 

 

Noisy Exp.

Noisy Exp.

Secure MAS

Framework

Threat

Results

Manipulation

Inference

Cooperative MASs + Trustworthy AI/ML

BRNES Framework

​

✅ Data Poisoning

✅ Dynamic Neighbor Selection

✅ Weighted Experience Agg.

​

✅ Membership Inference

✅ Differential Privacy

Results published in

IROS 2023

Venue: Detroit, MI, USA

8 of 13

Privacy technique itself could be exploited!

Attackers tuning noise variation for achieving stealthiness!

Degraded Performance

Results published in ICMLC 2023

Venue: U of Adelaide, AU

PeLPA Attack

​

✅ False Data Injection

✅ Attack Stealthiness

Cooperative MASs + Trustworthy AI/ML

9 of 13

Private Payload Delivery Network (PPDN), ACM 2023*

Adversarial Reinforcement Learning

UAV Network for Monitoring Wildfire Front, ICRA 2024*

10 of 13

Undergoing minor revision in

ACM Journal of Autonomous Transportation, 2023

RAMPART: Reinforcing Autonomous Multi-agent Protection through Adversarial Resistance

in Transportation

​

✅ Intelligent Path Planning

✅ Differential Privacy-exploited Attack

✅ Generative Adversarial Network (GAN)-based defense

​

Adversarial Reinforcement Learning

11 of 13

Multi-agent Systems

Search

Machine Learning

Knowledge Representation

Sociology

Cognitive Science

Economics

Management Science

Cybersecurity

Our research

Ongoing/Near-future Research

  • Generative Adversarial Network (GAN)
    • Adaptability under uncertainties
    • Anomaly Detection
  • Cognitive Science
    • Dual Action Processing (Thinking Fast and Slow)

12 of 13

Recent Publications (2023)

  • M. T. Hossain, H. La, and S. Badsha, “BRNES: Enabling Security and Privacy-aware Experience Sharing in Multiagent Robotic and Autonomous Systems,” In Proceedings of the 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2023), October 1–5, 2023, Detroit, MI, USA.
  • M. T. Hossain, and H. M. La, “Hiding in Plain Sight: Differential Privacy Noise Exploitation for Evasion-resilient Localized Poisoning Attacks in Multi-agent Reinforcement Learning,” Proceedings of The 22th International Conference on Machine Learning and Cybernetics (ICMLC 2023), July 9-11, 2023, The University of Adelaide, Adelaide, Australia.
  • M. T. Hossain, H. La, and S. Badsha, “RAMPART: Reinforcing Autonomous Multi-agent Protection through Adversarial Resistance in Transportation,” ACM Journal on Autonomous Transportation Systems (ACM JATS), 2023 (undergoing minor revision)
  • M. T. Hossain, Shahriar Badsha, Hung La, Haoting Shen, Shafkat Islam, Ibrahim Khalil,and Xun Yi, “Adversarial Analysis of the Differentially-Private Federated Learning in Cyber-Physical Critical Infrastructures,” IEEE Transaction on Dependable and Secure Computing (IEEE TDSC) (under review)
  • G. Srikar, M. T. Hossain, and H. La, "CRADLE: Cooperative Adaptive Decentralized Learning and Execution for UAV network to monitor Wildfire Front," 2024 IEEE International Conference on Robotics and Automation (ICRA 2024) (under review).

​

13 of 13

Thanks

If you want to know more about our recent research, please visit our Lab website (https://ara.cse.unr.edu/)