Revealing the Hidden Lives of Cryptic Carnivores with Machine Learning and AI
Briana Abrahms (PI), Department of Biology, Center for Ecosystem Sentinels
Zaid Harchaoui (PI), Department of Statistics & Paul G. Allen School in Computer Science & Engineering
Kasim Rafiq (Postdoc), Department of Biology
Medha Agarwal (PhD student), Department of Statistics
Ronak Mehta (PhD student), Department of Statistics
Project Summary:
Large carnivores are major contributors both to maintaining healthy ecosystems and to a range of services that benefit people. Despite their socioecological importance, there is very limited understanding of how and why climate changes affect carnivore behavior and ecology. The aim of this project is to use modern machine learning and AI to connect 8 years of high-resolution GPS and accelerometry data from free-ranging African wild dogs (Lycaon pictus), an IUCN red-listed Endangered species, to specific behaviors like hunting and eating, creating a “Rosetta stone” that translates animal-borne sensor data into discrete behaviors. More precisely, we shall use multiple time series from the animal-borne sensor data to classify discrete behaviors. Ultimately, these classified behaviors will be used to assess how changes in climate affect the behaviors of free-ranging large carnivores.
Foundations of Online Human Preference Adaptation for Trustworthy Human-AI Shared Autonomy
Karen Leung (PI), Department of Aeronautics & Astronautics
Abhishek Gupta (Co-PI), Paul G. Allen School in Computer Science & Engineering
Maryam Fazel (Co-PI), Department of Electrical and Computer Engineering
Isaac Remy (Research scientist), Control and Trustworthy Robotics Lab
Daphne Chen (PhD student)
Kazuki Mizuta (PhD student), Department of Aeronautics & Astronautics
Project Summary:
AI-enabled assistive shared autonomy aims to enhance the capabilities and experiences of human users which otherwise would not be possible alone. The goal of this project is to improve human trust in shared autonomy settings by developing theoretically sound adaptive robot decision-making algorithms that seamlessly account for users’ dynamically varying preferences.
Deep Spatiotemporal Embedding for Multiscale Quantum Matter Analysis
Ting Cao (PI), Materials Science & Engineering
Sheng Wang (Co-PI), Paul G. Allen School of Computer Science and Engineering
Xiaowei Zhang (Postdoc), Materials Science & Engineering
Yueyao Fan (PhD student)
Zixuan Liu (PhD student), Paul G. Allen School of Computer Science and Engineering
Project Summary:
Contemporary quantum materials face challenges due to their intricate structures and interfaces functioning across multiple scales. Current high-performance computer (HPC) simulation tools like density functional theory (DFT) and ab initio molecular dynamics (AIMD) struggle with these multiscale systems. Leveraging data science and artificial intelligence (AI) with HPC simulations offers new potential. This project will combine physics-based data structures, machine learning, and AI with HPC simulations to enhance the understanding of multiscale materials. By HPC materials simulations, the project seeks to explore the spatial and temporal correlations of the obtained datasets. Using unique embedding techniques, the research will address challenges faced by current models in atomic simulations and capture distinct physical laws at different time and length scales. This study will advance the understanding of quantum materials. Moreover, it will develop novel machine learning methods to handle long time-series and capture critical outbreak events.
Data-Driven Optimization of Stochastic Computational Experiments
Youngjun Choe (PI), Industrial & Systems Engineering
Kevin Jamieson (Co-PI), Paul G. Allen School in Computer Science & Engineering
Ribhu Sengupta (PhD Student), Industrial & Systems Engineering
Noah Darwin Feinberg (Undergraduate Research Assistant), Disaster Data Science Lab
Nathan Dennis (Undergraduate Research Assistant), Disaster Data Science Lab
Iris Zhou (Undergraduate Research Assistant), Disaster Data Science Lab
Project Summary:
The goal of this project is to innovate domain-agnostic data-driven algorithms to optimize stochastic computational experiments under computational resource constraints. Stochastic computational experiments are ubiquitous in science, engineering, and medicine. This project focuses on motivating applications in disaster management, where the proposed optimization can potentially save billions of dollars and many lives from disasters. The objectives for this project include: 1) collaboratively formulate novel optimization problems motivated by disaster management; 2) solve the problems by advancing the theory and method in closed-loop learning; and 3) validate the resulting closed-loop learning algorithms through high-impact case studies. These case studies will demonstrate 1) how cooperative multi-agent reinforcement learning can improve future coordinated recovery of interdependent infrastructure systems disrupted by a major earthquake, thereby reducing economic loss and saving lives, and 2) how the optimal adaptive experiment design can accelerate the analysis of rare extreme events (e.g., cascading blackout).