Tuesday, December 13, 2022, 12 – 3:20pm

► 12:00 pm - 12:40 pm → Gates Hillman 4405 and Zoom
     Understanding of Molecular and Evolutionary Mechanisms for Protein Temperature Adaptation
     — Yuting Deng

Billions of years of evolution have produced millions of diverse species. Adaptation is driven by changes in molecules whose functions give rise to life. Temperature acting as the driving force for evolutionary processes decides the adaptation of protein stability and activities. To demystify the molecular mechanisms underlying protein and temperature adaptation, this project aims to investigate proteins' thermostability through computational methods. Since a well-annotated functional database of protein is lacking, we curated a large-scale well-annotated functional database across archaea and bacterias, including ~9947 ortholog groups annotated by 40 critical function labels under rigorous quality control. In the future, this dataset would be of great value in dissecting biophysical forces that drive protein evolution through identifying temperature-associated residues and trends in temperature-associated residue properties and interactions.

► 12:40 pm - 1:20 pm → Gates Hillman 4405 and Zoom
     Information Retrieval: NERQ (Named Entity Recognition for Queries)
     — Dhruv Arya, Nidhi Dhar, Sarthak Tandon

Named Entity Recognition on web search queries (NERQ) has been noted as a challenging problem in the literature. The reduction in context, irregular grammar, and lack of proper casing in queries makes it difficult for off-the-shelf NER models to perform well on queries. This paper introduces a new and challenging supervised dataset for NERQ created from the MS MARCO dataset and performs experiments to improve the performance of the current state-of-the-art models on this dataset. These models show significant performance gains in experiments that perform finetuning with domain-transferred data.
   
► 1:20 pm - 2:00 pm → Gates Hillman 8102 and Zoom
     Exploring Trust and Strategies in Agent-Human Teaming
     — Annie Johnson

This work delves into evaluating trust in agent-human teaming in the gaming domain. This capstone project combines a Human-Computer Interaction (HCI) component as well as an analytic component. Here we provide a definition of “trust” relevant to the Overcooked game and perform a study where volunteers play this game and answer survey questions based on their experience. The game-play videos recorded during the study were used to train an Object Recognition model to test the robustness of the video. Self-Organizing Maps were used to cluster the trajectories obtained from the game to identify the factors that influence trust.

► 2:00 pm - 2:40 pm → Gates Hillman 4405 and Zoom
     AI2F: Artificial Intelligence Integrated Fires
     — Simon Knapp, Eric Youn, Rebecca Wilson

The current military planning process is slow and deliberate and requires days or even weeks of dedicated staff personnel. Our solution attempts to solve this problem for artillery planning by applying reinforcement learning to develop agents that can rapidly develop courses of action and wargame various scenarios. On a broader scale, the simulation environment we are currently developing alongside the US Army Engineer Research and Design Center (ERDC) provides researchers with a valuable tool for rapidly developing agents for other aspects of military planning. Our experiments explore the training methods for agents using this framework and how agents learn to apply indirect fires and ammunition selection in it.

► 2:40 pm - 3:20 pm → Gates Hillman 4405 and Zoom
     Nudge
     — Hermes Suen

Nudge is a project aimed at creating a platform for individuals, institutions, and organizations to invest in real-life behavior change. It is built on the assumption of memetic desire and social contagion and provides monetary rewards to creators of viral TikToks that are related to the desired behavior change.

Event Type: Project Presentations
Room Number: In Person and Virtual - ET
Building: Reddy Conference Room, Gates Hillman 4405 and Zoom
For More Information: ahan2@andrew.cmu.edu
Affiliations: Language Technologies Institute (LTI)
Organization(s): MCDS
Event Website Title: Full Schedule of Daily Presentations
Event Website URL: mcds-cmu.github.io…