Deep reasoning
I study how causality and other formal reasoning systems can improve AI reasoning and help solve complex problems that require exploration over thousands of steps. See Plan*RAG for an example on how solving complex tasks can be abstracted out as planning over a directed acyclic graph.
Improving AI reasoning
- A planning algorithm that utilize LLMs for completing complex tasks (Verma et al., 2024)
- Axiomatic training framework for teaching compositional reasoning to AI models (Vashishtha et al., 2024)
Text-based optimization for generative AI
- Optimizing prompts for large language models (Juneja et al., 2024; Srivastava et al., 2024)
References
- ArxivPlan*RAG: Planning-guided Retrieval Augmented GenerationarXiv preprint arXiv:2410.20753, 2024
- ArxivTeaching Transformers Causal Reasoning through Axiomatic TrainingarXiv preprint arXiv:2407.07612, 2024
- ArxivTask Facet Learning: A Structured Approach to Prompt OptimizationarXiv preprint arXiv:2406.10504, 2024
- ACL 2024NICE: To optimize in-context examples or not?In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024