## Denghui Zhang - Stevens Institute of Technology - CS/Data Science PhD/Intern ### Research Interests: Advancing Socially-Ethical and Efficient Generative AI ### Research Directions: ### 1. **Understanding LLM Mechanistics for Ensuring Trustworthiness** * Detecting and Preventing Copyright Violations: Develop frameworks to identify and mitigate intellectual property violations, ensuring LLM compliance with copyright laws. * Enhancing LLM Safety: Create predictive models to detect and address risky behaviors before generation at inference time, improving reliability and safety. * Investigating Theory of Mind (ToM): Explore how LLMs simulate reasoning and align with human values, enhancing interpretability and ethical decision-making. ### 2. **Advancing the Fundamental Reasoning Ability of Language Models via Architecture Redesign** * Advancing Reward Models for Reasoning: Develop process-based reward models that emphasize reasoning pathways over outcome-driven objectives, promoting thoughtful and logical decision-making in LLMs. * Innovating Attention Mechanisms: Design structural attention mechanisms to improve LLMs' ability to process hierarchically organized and context-rich data, enhancing accuracy and relevance in reasoning tasks. * Enabling Moral and Legal Reasoning: Equip LLMs to navigate complex social values and legal principles, ensuring their outputs align with societal norms and regulatory standards. ### 3. **Broadening Real-world LLM Applications and Improving Domain Adaptation** * LLM Agents for Financial Decision-Making: Develop models for tasks like investment, risk assessment, and compliance to enhance financial decision-making. * Multi-Modal Interpretation of Accounting and Regulatory Documents: Design models to interpret complex documents using tables, charts, and text for improved problem-solving. * Improving Domain Adaptation: Focus on enhancing LLM capabilities for high-impact applications in finance and regulatory environments. ## Position Detail An interdisciplinary degree program of the Schaefer School of Engineering and Science and the Stevens School of Business, the data science Ph.D. curriculum drives students to master the bedrock principles, methods and systems for extracting insights from rich data sets. Then, you’ll apply those theories, techniques and applications in practical research alongside Stevens faculty who are working at the forefront of the data science field. Our graduates go on to pursue research careers in academia and secure important positions in industries like business, financial services and life sciences. More details and application link: https://www.stevens.edu/program/data-science-phd Application deadline 01/15/2025, please submit ASAP if interested. ## Advisor Profile * Denghui Zhang is an Assistant Professor at Stevens Institute of Technology. He obtained the PhD degree at Rutgers University in 2023, where he was grateful to have Prof. Hui Xiong as his advisor. * He mainly publishes at KDD/NeurIPS/AAAI/SIGNIR/EMNLP/CIKM/IEEE-TKDE… His work has been awarded ICIS 2023 Best Student Paper. * His research has been supported by OpenAI Researcher Access Probram, NSF-NAIRR Program. * He has rich experience in working with students, leading 10 students and interns to publish papers at top conferences. * He deliveres tutorials at top-tier AI/NLP conferences such as AAAI'24, NAACL'25 * He served as reviewers for top AI journals such as TACL, TKDE, TIST, and served as program committee at *DM and *CL conferences (Session Chair at KDD 2022, Demo Track Area Chairs at EMNLP 2024, etc.) * Homepage: https://zhangdenghui.site/ * Email contact: dzhang42@stevens.edu