2020 POEM Lab Accomplishments
Hello from 2021! It's now been almost three years(!) since I last posted here, as I've gotten caught up in adjusting to faculty life. Without making any promises for the future, I have some ideas for revitalizing this blog, or otherwise regularly updating the public on my & my lab's professional activities. I'd like to start with a report on what my students accomplished in 2020.
Despite 2020 being a challenging year for everyone, my POEM lab crew -- a combination of PhD, masters, and undergraduate research students -- accomplished some truly amazing things. In the interest of celebrating accomplishments, I asked my students to each share something they were proud of from the preceding year. Their responses are below, supplemented with a bit of my own bragging.
Claire Aguiar, a sophomore undergraduate student, joined a cross-disciplinary team with Political Science faculty and students to help us study the potential for board games to operationalize, explore, and generate data for illicit networks (e.g. drug trafficking and antiquities smuggling). She modified an existing game to work over Zoom and collected playtest data, then iterated on a new design based on the bluffing game Sheriff of Nottingham.
Jake Matteson, a junior undergraduate, conducted a study on the role of feedback for improving player skill in fighting games. He analyzed the tutorials and runtime feedback mechanisms for three fighting games, then modified an open source fighting game engine to track special cases of player behavior, then summarize aggregate data for players. Apart from research, Jake is proud of improving at drawing and writing a lot.
Aaron Williams, a senior undergraduate, wrote an action model learning algorithm called Blackout that uses failed actions and some semblance of invariant detection to produce better models than at least one published alternative (FAMA). Aaron also wrote an interface that allows for a human to play a version of Sokoban that’s backed by a planning domain, and generates play traces that can be used in action model learning. Aaron was second author on our AIIDE paper “Towards Action Model Learning for Player Modeling,” along with Abhijeet and me, and he developed the 2-minute video pitch for our poster presentation.
Thomas Bacher, 2nd year masters student, conducted a lab study of target user interactions with our interactive fiction AI authoring tool, Villanelle. Thomas is very proud of having run the study this year despite the setbacks of having to move to a digital format -- we had originally planned to conduct in-person task observations and interviews. Thomas is now in the process of writing up his results and applying to Ph.D. programs.
Jennifer Wellnitz, a 2nd year PhD student, made significant progress on two research projects: a write-up of an effort to model character knowledge exchange in the game Elsinore in our Dynamic Epistemic Logic framework Ostari, and an operationalization of 5-factor personality in a character authoring framework based on behavior trees. Jennifer published a workshop paper at INT on the Elsinore project and recorded a 15-minute video presentation for the workshop.
Abhijeet Krishnan, 3rd year PhD student, wrapped up his study of the potential of Action Model Learning for player modeling, and he published his first paper at a proper academic conference this year (AIIDE) on this topic. More recently, Abhijeet has been interested in the problem of offering strategy-level feedback to game players, and he is currently investigating explainability mechanisms for reinforcement learning for this purpose. Abhijeet mentored Jake on his fighting game project, as well.
Sasha Azad, 4th year PhD student, continued her groundbreaking work on social simulation by conducting a survey of social interaction mechanics in prior simulation projects. She crawled through the codebase for 8 research social simulations and the wikis of two commercial games, curated a huge list down to about 700 interactions, did open-coding and thematic analysis on this data, evaluated the codes, and is currently in the process of analysing the taxonomy rules we’ve designed. Sasha is also particularly happy about the two new methods of data analysis she found after a lot of difficult searching this year, one statistical and one rule-based process for visualizing co-occurrence relations.
Alexander Card, 5th year PhD student, is happy to have gotten classroom interventions in CSC 281 and concrete goals for the CSC 281 Planner. He’s also pleased that he has found a direction for his oral exam and dissertation, and he’s making progress towards papers supporting the exams.
Chinmaya Dabral, 5th year PhD student, completed development of a new technique for generating conflict- and intention-based narrative plans using Answer Set Programming. He passed his written preliminary exam by presenting this work, and he coauthored a paper about it for AIIDE 2020, which was nominated for best paper.