| |
 | | Towards Memory-Augmented and Value-Aligned Language Agents |  | Cassandra TAN Hui Ming PhD Candidate School of Computing and Information Systems Singapore Management University FULL PROFILE |
Research Area - Artificial Intelligence & Data Science
- Machine Learning & Intelligence
Dissertation Committee | | Date 27 November 2026 (Friday) Time 1:00pm – 2:00pm Venue Meeting room 5.1, Level 5 School of Computing and Information Systems 1, Singapore Management University, 80 Stamford Road, Singapore 178902 Please register by 25 November 2026. We look forward to seeing you at this research seminar.
|
| ABOUT THE TALK Large language models (LLMs) have enabled the development of intelligent agents that interact with users over extended periods. However, such agents face persistent challenges in retaining past experiences, reasoning about temporal relationships, and ensuring that their behaviour remains aligned with human values. Existing approaches often rely on limited context windows, unstructured retrieval, or implicit alignment mechanisms, making it difficult to support reliable long-term interaction and accountable decision-making.
This dissertation investigates memory-augmented and value-aligned language agents through three complementary research directions. First, it develops ARTEM, a biologically inspired episodic memory framework that enables agents to encode and retrieve events using temporal, spatial, entity, and content information. Second, it introduces VALVE, a runtime governance framework that represents human values through an explicit hierarchy and enforces them over recalled memories and generated responses. Third, it proposes PBTG, a probabilistic temporal reasoning framework that integrates fragmented and uncertain temporal evidence into a coherent belief state.
The research evaluates these approaches on episodic memory, value alignment, and temporal reasoning benchmarks, examining their effectiveness and limitations. Ongoing work further investigates persistent temporal belief representations and adaptive value specifications through i-VALVE. Collectively, the dissertation aims to advance the development of language agents capable of remembering past experiences, reasoning about events over time, and acting in accordance with explicit and accountable human values. | ABOUT THE SPEAKER Cassandra TAN Hui Ming is a PhD candidate in Computer Science at the School of Computing and Information Systems, Singapore Management University, under the supervision of Professor TAN Ah Hwee. Her research focuses on memory-augmented language agents, episodic memory, temporal reasoning, and value alignment. Her work investigates the integration of biologically inspired neural models, probabilistic reasoning, and runtime governance mechanisms to improve the reliability and accountability of intelligent agents. Her research on episodic memory for large language model agents, ARTEM, was published at the AAAI Conference on Artificial Intelligence (AAAI 2026). |
|