Research

What We Do

We investigate human-centric AI systems with common sense that can team with people intuitively and reliably.

We perform fundamental research on commonsense AI and investigate its application to challenging domains, informed by empirical insights and cognitive theories.

Research Topics:

1. Intelligence studies: We evaluate the ability of AI models to exhibit cognitive phenomena. Our recent work focuses on analogical reasoning, visual abstraction and generalization, lateral thinking, and perceiving objects of various sizes and with various properties.
2. Neuro-symbolic methods for commonsense reasoning: We develop methods for neuro-symbolic commonsense reasoning. Recently, these include methods using prototype-based networks, combining LLMs with deterministic engines, relational analogical engines, inducing rules from structural resources, and reasoning with scene knowledge graphs.
3. AI for good: We investigate how AI can serve the social good. Current domains of interest include 1) content safety online, where we address misinformation and hate speech in internet memes and logical fallacies; 2) what-if reasoning in traffic; 3) devising analogical and rule-based explanations in an education setting; and 4) informing sustainability policies through knowledge-based AI.

For further details, see my recent publications.

Core Principles

• Diversity: As someone who has lived in three countries and has an appreciation for cultural differences, I embrace diversity in terms of demography, background knowledge, interest, and opinion.
• Cross-discplinarity: Most standing AI challenges cannot be solved purely by engineering. I am working hard to build bridges with other disciplines, including software engineering, cognitive psychology, social science, and linguistics.
• Motivation and guidance: I enjoy working with students and PostDocs. My firm belief is that with the right guidance and dedication, people can stay motivated and make substantial contribution while enjoying their work. I meet with each of my team members every week, and we communicate on slack in the meantime as needed.
• Growth: Growing at our job is important. I do my best to help each of my team members grow into a proud researcher with strong technical skills, robust conceptual understanding, and clarity toward applying the technology for good.
• Mistakes are ok, not learning less so: We are all humans and have knowledge gaps, make mistakes, etc. But not learning from one’s mistakes I understand less - giving the same advice over and over again is unproductive and disrespectful.
• Teamwork and collaboration: Seeing science as a team sport, I have been actively seeking collaborations, resulting in effective co-supervision and co-organization with academics (USC, CMU, RPI, University of Lyon, VUA, UvA, University of Bielefeld) and industry partners (Bosch Research, NEC Labs, Merit Technologies, Tencent). I am currently expanding my collaboration network on other continents beyond Europe and North America.
• Deliberate work and non-work: To have productive and meaningful working days, we need to be fully focused when we work. When we are not working, we should be equally determined to preserve the time and space for leisurely activities.
• Minimize disruptions: The biggest enemy of productive and meaningful work are disruptions. I do my best to minimize disruptions both from and to me. I advise my team to do the same.