DeepDelta Lab

We are DeepDelta Lab, a research group at School of Artificial Intelligence in Shanghai Jiao Tong University. The PI of DeepDelta Lab is Dr. Shaobo Cui.

We have four overarching goals:

1. Causal Intelligence: We study how machines can learn and leverage causal relationships, moving beyond correlational prediction toward intervention, counterfactual reasoning, and robust generalization. Our research includes causal representation learning, causal reasoning, causal strength modeling, and the integration of causal mechanisms with foundation models.

2. Causal World Models: We develop world models that can understand environment dynamics, predict the consequences of actions, and reason about interventions and counterfactuals. Key topics include physical and causal mechanism modeling in vision and video, long-horizon temporal reasoning, memory mechanisms, and planning and decision-making in complex environments.

3. Causality/AI for Science (Medicine & Biology): We explore foundation models and causal methods for scientific problems in medicine and biology, including multimodal biomedical data modeling, medical knowledge and literature understanding, molecular and protein modeling, uncertainty quantification, and causal reasoning and decision support for clinical applications and scientific discovery.

4. Alignment of Large Language Models: Build reliable, efficient, and steerable LLMs that can reason, plan, and collaborate. Focus particularly on the alignment stage of LLMs.

我们 (DeepDelta Lab) 是 上海交通大学人工智能学院 的一个研究组。研究组的PI 是 崔少波博士，DeepDelta Lab 的主要研究方向为：

1. 因果智能（Causal Intelligence）：研究机器如何学习和利用因果关系，从相关性预测进一步走向干预、反事实推理与稳健泛化，包括因果表征学习、因果推理、因果强度建模，以及因果机制与基础模型的结合。

2. 因果驱动的世界模型（Causal World Models）：研究能够理解环境动态、预测行动后果并进行干预与反事实推演的世界模型，重点关注视觉与视频中的物理和因果机制建模、长期时序推理、记忆机制，以及面向复杂环境的规划与决策。

3. Causality/AI for Science（医学与生物）：探索基础模型与因果方法在医学、生物等科学问题中的应用，包括多模态生物医学数据建模、医学知识与文献理解、分子与蛋白建模、不确定性量化，以及面向临床和科学发现的因果推理与决策支持。

4. 大语言模型对齐：构建可靠、高效、可引导，且能够推理、规划与协作的 LLM，重点研究对齐阶段的关键问题。