Research Interests
My broad research interests pertain to how we can train computers to learn and perform language in the same manner that humans do, and creating intelligent systems that have application to the real world. My specific research interests include:
- Learning in Interactive Environments: I want to investigate how we can use both (1) natural instructions and (2) extracted textual information from an environment space to teach intelligent agents. I believe that teaching intelligent agents from their experiences in these interactive environments leads to a certain depth of reasoning that cannot be replicated from static data. Additionally, I am really excited about how interactive environments allow agents to learn real world dynamics and thereby lead to better applications in the real world.
- Language Grounding: True meaning of language does not come from the numeric representations of words, but rather the real world entities and concepts that they refer to. Therefore, when expanding the way machines interpret language, we must look beyond the pure linguistic properties of words and be able to relate language to the physical world. My interest in language grounding extends to both embodied textual worlds and multimodal graphical environments.
- Situated Dialogue: The end goal of creating intelligent language systems is to enable users to interact directly with the system through the medium of dialogue. Related to my above interest of teaching agents how to interact with users/other agents in interactive environments, I wish to explore situated dialogue as a means to better improve the human-AI interactions of these systems. Bettering the contextual understanding and reasoning of our agents will lead to more autonomous systems that can generalize to increasingly complex tasks.
- Natural Language Generation: I view natural language generation as an integral component of many other NLP topics, including dialogue systems, and therefore want to build on the state-of-the-art techniques of the field.
Publications
-
1Cademy @ Causal News Corpus 2022: Leveraging Self-Training in Causality Classification of Socio-Political Event Data
Adam Nik, Ge Zhang, Xingran Chen, Mingyu Li, Jie Fu
[Paper] | [Poster]
CASE Workshop @ EMNLP 2022 -
1Cademy @ Causal News Corpus 2022: Enhance Causal Span Detection via Beam-Search-based Position Selector
Xingran Chen, Ge Zhang, Adam Nik, Mingyu Li, Jie Fu
[Paper] | [Poster]
CASE Workshop @ EMNLP 2022, best of shared task report
Projects
-
Snake Game AI with Approximate Q-Learning Agent
This project is an implementation of a Snake Game AI agent using an approximate Q-learning algorithm. This project was my final project for CS321: Making Decisions with AI, which I took in the Fall of 2021. The course was one of my first introductions to the field of artificial intelligence, and the production of this project sparked my interest in Reinforcement Learning.
[Code] | [Demo] -
Scenify, Senior Computer Science Comps Project
In my Senior Comps Project, my group and I created a route generation application that focused on optimizing the scenic value of the route, rather than time or distance. We quantified the scenic value of candidate routes by querying Flickr images along the routes and inferring scenic qualities from the images by classifying them using a convolutional neural network.
[Link] -
Ping Pong LED Snake Game Board
I built an 8x8 wooden LED board with ping pong balls as diffusers as part of my capstone project of CS232: Art, Interactivity, and Robotics. To go along with the board, I coded a Snake game compatible with Arduino board that links to the LED board. The board is thus a user-playable Snake game that uses a joystick as controls to the game. The project provided me with ample experience with working with electronics, as I completed all soldering, wiring, and circuit board work necessary for the board.
[Code] | [Brief Demo]
Essays
-
Survey on Natural Language Generation in Open-Domain Dialogue Systems
I wrote this essay as my final paper for CGSI130: What Minds Are What They Do: An Introduction to Cognitive Science. It is a brief survey on neural approaches to natural language generation and the current challenges of language generation in open-domain dialogue systems. The motivation behind the topic came from studying Turing's ideas of how machines can demonstrate human cognitive abilities and Rumelhart's connectionist architectures of the mind from the 1980s, both of which we studied earlier in the course.
[Paper]
About me
I grew up in Nashville, TN, and moved to Minnesota when I was 13 and have been living here since. Outside of academics, I spend most of my time playing sports and staying active. In college, I was a two-sport varsity athlete (football and swimming), and I stay active nowadays by lifting, hiking, and going on walks. Additionally, I have three cats, named Mocha, Ali, and Rosie.