CS Events

PhD Defense Integrate Learning and Reasoning for Large Language Model and Recommender Systems
Export Event Save iCal Send to Google Calendar Send to Yahoo Calendar Send to Outlook Live Send to Microsoft Outlook Save iCal Preserve formatting in description (only supported in some calendar applications) Close Friday, August 23, 2024, 10:00pm - 11:30pm

Speaker: Jianchao Ji

Location : Virtual

Committee: 

Prof. Yongfeng Zhang (Advisor)

Prof. Dong Deng 

Prof. Hao Wang 

Dr. Yan Liang (external, Senior Applied Scientist)

Event Type: PhD Defense

Abstract: In the current era, recommendation systems has become increasingly integral to our daily lives. Despite the significant advancements in recommendation systems, many models still lack robust reasoning capabilities, which limits their effectiveness and reliability. This research aims to address this gap by enhancing the reasoning abilities of recommendation systems to improve their overall performance.Our initial focus was on integrating counterfactual reasoning and logical reasoning abilities into recommendation models. By applying these reasoning techniques, we observed a marked improvement in the performance of recommendation systems, demonstrating the potential of reasoning-enhanced models in practical applications. Building on these findings, we extended our approach to large language models (LLMs). Our experiments showed that incorporating logical reasoning into LLMs significantly enhanced their performance across various benchmarks. Given their extensive use in natural language processing tasks, we aimed to improve their text reasoning abilities to enhance recommendation performance. This ensures that these models can understand, process, and respond logically to complex textual information.However, the rapid development of large language models also brings forth ethical challenges. It is imperative that we do not solely focus on optimizing model performance without addressing potential moral and ethical concerns. Recognizing this, we developed a benchmark to evaluate the moral reasoning ability of LLMs. This benchmark assesses the ethical implications of the models' outputs, ensuring that they align with societal values and norms.

: 

Contact  Professor Yongfeng Zhang

Zoom Link:
https://rutgers.zoom.us/j/97177193164?pwd=fsodIkbLLM6GEEozoaLdcKIMKB8tu4.1