Investigator: Harold Soh.
The principal aim of this project is to develop core techniques for learning models of other (human) agents: Artificial Theories of Mind and Body (AToM/B).
We intend to build novel hybrid techniques that learn flexible deep models, yet are able to leverage prior knowledge that is expressive and human-interpretable. In essence, we aim to bridge theory-driven and data-driven approaches to model development.
While our goal is to develop general learning methods, we will ground our research in assistive scenarios; we plan to integrate our models into decision-making frameworks and evaluate our approach experimentally using assistive tasks with human subjects.
Funded by: AI Singapore
- Hyperprior Induced Unsupervised Disentanglement of Latent Representations, Abdul Fatir Ansari and Harold Soh, AAAI 2019
- Semantically-Regularized Logic Graph Embeddings, Yaqi Xie, Ziwei Xu, Kuldeep Meel, Mohan Kankanhalli, and Harold Soh, Neural Information Processing Systems (NeurIPS), 2019