Hi there! I’m Haoran, an undergraduate senior in the Department of Information Science at Drexel University majoring in Data Science, minoring in CS and Linguistics. In my first two years of college, I’m glad to get the chance to attend Lanzhou University, Gansu, China.
I’m privileged to be advised by Prof.Jake Williams at Drexel and to have the chance to collaborate with Prof.Noah Goodman and Prof.Tobias Gerstenberg at Stanford CoCoLab & CiCL and Prof.Franz Wotawa from TU Graz.
I’m happy to chat about anything, please feel free to send me an email (hz454 [at] drexel [dot] edu)!
Research Interests
My long-term research goal is to build machines that learn and think like humans - human-level AI. With this goal in mind, one of the biggest challenges I’m thinking about is how to build machines with human cognitive abilities, acknowledging that current AI systems are still far from human capabilities. Considering this, I plan to draw inspiration from Computational Cognitive Science, Neuroscience, and Linguistics to advance AI development. From the AI perspective, I believe that understanding modern LLMs’ true-level language understanding and various reasoning capabilities is a critical first step to knowing how to make them do better. Combining with Computational Cognitive Science, knowing how humans can perform various complicated cognitive tasks, both at the neuro and behavioral levels, is equally important.
Knowing that LLMs are currently the most powerful AI systems, my research will focus on LLMs on the AI side. I see CogSci-inspired AI and NeuroAI as two promising research directions to go.
Two main research toptics I am currently thinking about:
Understanding LLM capabilities with Reasoning: By studying reasoning, I hope to answer: To what extent, LLMs can think like humans and their true level of language understanding
Cognitive Language Modeling: How can we create LMs to mimic human learning efficiency in terms of data and algorithmic innovation?
With the outlined above research interests, I plan to use the following methods to conduct my research:
- designing stimuli to compare LLMs with humans to understand how well LLMs can perform on various human cognitive tasks
- using (neuro)symbolic methods to analyze LLMs’ reasoning abilities
- using meta-learning, lifelong learning, and reinforcement learning methods to improve some of the cognitive abilities in LLMs
- using computational modeling methods to model human cognitive abilities
- using neuroimaging methods to gain a deep and detailed understanding of information processing in the brain
Things I’m eager to learn to improve my skillset:
- probabilistic modeling, bayesian inference
- neuroimaging methods to analyze how the brain processes information
News
- December 2023 — Being selected for Honorable Mention of the 2024 CRA Outstanding Undergraduate Researcher Award!