PhD position in Logic and Machine Learning, IRIT Toulouse, France

PhD position in Logic and Machine Learning
Artificial and Natural Intelligence Toulouse Institute (ANITI)
Institut de Recherche en Informatique de Toulouse (IRIT)
Toulouse University
France

The interdisciplinary institute in artificial intelligence of Toulouse, named the Artificial and Natural Intelligence Toulouse Institute (ANITI), is one of four institutes spearheading research on AI in France. A program of 24 chairs is funded by ANITI. This includes the chair “Empowering Data-driven AI by Argumentation and Persuasion”. Emiliano Lorini, one of the members of the chair, is seeking a PhD student to work on the research project “Explaining Learning Agents”. The PhD thesis will start in September 2020 and will be funded on a three-year contract with net salary of 2600€ per month with some teaching (64 hours per year on average).

Description of the Research Project

Research on machine learning (ML) is nowadays dominant in artificial intelligence (AI). This includes research on artificial neural networks, of which deep learning is the most representative example, and reinforcement learning (RL). The success of ML lies in two interrelated aspects, namely, the availability of voluminous data sets (big data) that can be used for training neural networks and reinforcement learning algorithms as well as its enormous impact on a large spectrum of domains and applications ranging from vision and pattern recognition, through natural language processing (NLP), to robotics and game-playing. Research on ML is often seen in opposition to research in the area of knowledge representation and reasoning (KR). The latter includes planning, argumentation, belief revision and update as well as graphical and compact representation of uncertainty and preferences including Bayesian networks and CP-nets. Logic is certainly the core of research in KR since all other approaches are expressible in logical terms. For instance, classical planning problems can be expressed in propositional logic, classical belief revision and update are operations on beliefs expressed by propositional formulas, argumentation can be instantiated either in a classical logic setting or in a non-classical one such as defeasible logic, modal logic or dynamic logic. More generally, logic is the main tool for modeling the different aspects of reasoning and rationality that can be integrated in an artificial system such as a robot or an embodied conversational agent (ECA). The need for an integration of ML and KR has been largely emphasized in the artificial intelligence (AI) community. According to (Valiant, 2003), a key challenge for computer science is to come up with an integration of the two most fundamental phenomena of intelligence, namely, the ability to learn from experience and the ability to reason from what has been learned. The PhD thesis will be focused on the integration of logic-based methods and ML methods aimed at endowing agents interacting in a multi-agent system with both predictive and explanatory capabilities, that is to say, with the capacity:
• to form predictions about future event occurrences and future agents’ choices based on their past experiences, and
• to explain past and future event occurrences as well as past and future agents’ choices.
To this aim, we plan to combine concepts and methods from epistemic logic (Fagin et al., 1995; Lorini, 2018), theories of learning in games and multi-agent learning (Fudenberg & Levine, 1998; Tuyls & Weiss, 2012). Moreover, we expect to consider and clarify a number of notions of explanation singled out in the area of explainable AI (Dhurandhar et al., 2018; Ignatiev et al., 2019; Mothilal et al., 2020). We expect the kind of integration proposed in the context of PhD thesis to be relevant for AI applications in social robotics and human-machine interaction, given the importance of combining reasoning and learning as well as prediction and explanation for such applications.

References

• A. Dhurandhar, P.-Y. Chen, R. Luss, C.-C. Tu, P.-S. Ting, K. Shanmugam and P. Das. Explanations based on the Missing: Towards Contrastive Explanations with Pertinent Negatives. In Proceedings of the Annual Conference on Neural Information Processing Systems, NeurIPS, pages 590–601. 2018.
• R. Fagin, J. Y. Halpern, Y. Moses, and M. Vardi. Reasoning about Knowledge. MIT Press, Cambridge, 1995.
• D. Fudenberg, D. K. Levine. The Theory of Learning in Games. MIT Press, Cambridge, 1998.
•  A. Ignatiev, N. Narodytska and J. Marques-Silva. Abduction-Based Explanations for Machine Learning Models. In Proceedings of the The Thirty-Third AAAI Conference on Artificial Intelligence (AAA-19), pages 1511-1519. 2019.
•  Lorini, E. (2018). In Praise of Belief Bases: Doing Epistemic Logic without Possible Worlds. Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18), AAAI press, pp. 1915-1922.
•  R. K. Mothilal, A. Sharma, C. Tan. Explaining Machine Learning Classifiers through Diverse Counterfactual Examples. In Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAT’20). ACM Press, 2020.
•  K. Tuyls and G. Weiss. Multiagent Learning: Basics, Challenges, and Prospects. AI Magazine, 33(3):41, 2012.
•  L. G. Valiant. Three Problems in Computer Science. Journal of the ACM, 50(1):96-99,2003.
•  S.Wachter, B. D. Mittelstadt, C. Russell. Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR. In CoRR, volume abs/1711.00399. 2017. 2018.

Candidate Profile

The PhD is at the intersection of logic, game theory and machine learning. The ideal candidate should have a strong mathematical background and a master’s degree in Computer Science, Logic or Mathematics. Ideally, it should be familiar with propositional logic, modal logic, epistemic and temporal logics, the theory of static and sequential games as well as with basic ML techniques based on artificial neural networks and reinforcement learning.

How to Apply

Please email your detailed CV, a motivation letter, and transcripts of bachelor's degree and master’s degree to Emiliano.Lorini at irit.fr. Samples of published research by the candidate and reference letters will be a plus.

APPLICATION DEADLINE FOR FULL CONSIDERATION: May 30th, 2020.