Human-Agent Learning (HAL) lab

• Introduction
• Our team
• Relevant publications

As artificial intelligence (AI) systems are becoming ubiquitous in our everyday life, it is inevitable that they interact with humans during their operational process. For example, many AI applications serve as an augmentation of human users by putting them at the centre of the system (e.g., digital personal assistants), or humans and AI team up together to jointly make decisions/execute actions (e.g., disaster response teams consisting of human rescuers and autonomous unmanned vehicles). These types of systems are known as human-aware AI in the literature. The main focus of the Human-Agent Learning (HAL) lab is to design adaptive and learning algorithms that work well in such presence of human factors in AI systems.

In particular, we look at the following problems:

• Collaborative human-agent learning: How a mixed team of human and agent members can utilise the complimentary knowledge and expertise of their members to develop efficient solutions for different optimisation problems in AI and machine learning.
• Efficient machine learning against strategic human behaviours: To what extent selfish and strategic human behaviours can manipulate the overall system dynamics to their own benefit, and how we can protect the system from such behaviours.

The main tools we use to tackle these problems include, but not limited to: game theory and mechanism design (Stackelberg games, congestion games, coalitonal formation, etc.), online learning (bandit theory, online optimisation, reinforcement learning etc.), and incentive engineering.

Main investigator(s):

• Long Tran-Thanh

PhD students:

• Nick Bishop (PhD student, Southampton, 2018-) - bilevel optimisation problems in machine learning (co-supervised with Enrico Gerding)
• Tom Davies (PhD student, Southampton, 2019-) - algebraic topology and its applications in machine learning (co-supervised with Corina Cirstea)
• Le Cong Dinh (PhD student, Southampton, 2018-) - learning and last round convergence in asymmetric games (co-supervised with Tri-Dung Nguyen and Alain Zemkoho)
• Taha Gunes (PhD student, Southampton, 2017-) - vulnerability analysis of trust systems in AI (co-supervised with Tim Norman)
• Guillermo Moreno (PhD student, Southampton, 2018-) - influence maximisation games (co-supervised with Markus Brede)
• Tin Leelavimolsilp (PhD student, Southampton, 2016-) - selfish mining in multi-agent blockchain systems (co-supervised with Seb Stein)

Undergrads:

• Balint Gucsi (Southampton) - human-robot collaborative learning

Interns:

• Balint Gucsi (Southampton, 2019) - annoyance aware preference elicitation in human-robot collaboration
• Erich Zimmermann (Southampton, 2019) - AI for asthma attack detection/prediction
• Yoana Paleva (Southampton, 2018) - AI for suicide prevention - Last known position: with Microsoft Cambridge
• Yi Zheng (Southampton, 2018) - security games for smart traffic control - Last known position: PhD student at USC with Sven Koenig
• Will Greedy (Southampton, 2017) - Bayesian adversarial machine learning
• Ionela (Gini) Mocanu (Southampton, 2017) - online learning based data manipulation against ML algorithms - Last known position: PhD student at Univ. of Edinburgh
• Dan Tudosiu (Southampton, 2017) - membership attack approaches against ML algorithms - Last known position: PhD student at King's College London

• Md Mosaddek Khan (PhD student, Southampton, 2015-18) - decentralised constraint optimisation problems - Last known position: Asst. Prof. at Univ. of Dhaka
• Edoardo Manino (PhD student, Southampton, 2015-19) - Bayesian inference techniques in crowdsourcing - Last known position: postdoc at Southampton
• Paolo Serafino (postdoc, Southampton, 2016-18) - strategyproof smart traffic control - Last known position: postdoc at Oxford
• Nhat Truong (PhD student, Southampton, 2015-19) - incentive engineering in crowdsourcing systems

Academic collaborators:

• Bo An (NTU, Singapore) - security games, multi-agent systems
• Tim Baarslag (CWI/Utrecht, Netherlands) - user annoyance
• Hau Chan (Nebraska-Loncoln, US) - discrete optimisation, AI for social good
• Fei Fang (CMU, US) - security games, AI for social good
• Enrico Gerding (Southampton, UK) - mechanism design for truthful machine learning
• Nick Jennings (Imperial College London, UK) - human-agent collectives, multi-agent systems
• The Anh Han (Teesside, UK) - evolutionary game theory
• Md Mosaddek Khan (Univ. of Dhaka, Bangladesh) - decentralised constraint optimisation
• Minming Li (CUHK, Hong Kong) - network games
• Tri-Dung Nguyen (Southampton, UK) - game theory
• Tim Norman (Southampton, UK) - trust modeling in multi-agent systems
• Zinovi Rabinovich (NTU, Singapore) - security games, multi-agent systems
• Gopal Ramchurn (Southampton, UK) - demand-side management
• Alex Rogers (Oxford, UK) - multi-agent systems, sensor networks
• Arunesh Sinha (SMU, Singapore) - empirical games
• Sebastian Stein (Southampton, UK) - crowdsourcing
• Milind Tambe (Harvard, US) - security games, AI for social good
• Danesh Tarapore (Southampton, UK) - human-robot collaborative learning
• Son Tran (NMSU, US) - decentralised constraint optimisation, logic programming + ML
• William Yeoh (WUSTL, US) - decentralised constraint optimisation
• Haifeng Xu (Univ. of Virginia, US) - bandit theory, security games
• Xiaowei Wu (Univ. of Macao) - network games

• Jiarui Gan (Oxford, UK) - deception in Stackelberg games
• Qingyu Guo (NTU, Singapore) - learning in security games
• Peter Key (MSR Cambridge, UK) - online keyword bidding
• Tiep Le (NMSU, US) - home energy management
• Debmalya Mandal (Columbia University)- blocking bandits
• Tao Qin (MSR Asia, China) - budgeted bandits
• Anshuka Rangi (UCSD, US) - budgeted bandits
• Matteo Venanzi (Microsoft, UK)- resource allocation in crowdsourcing systems
• Dong Quan Vu (Grenoble, France) - Colonel Blotto games
• Marcin Waniek (NYU Abu Dhabi, UAE) - Dollar auctions
• Bryan Wilder (Harvard, US) - algebraic topology based optimisation, AI for social good