About Me
I am broadly interested in Computer Vision, and Embodied AI; however, my specific interests include Multimodal Machine Learning and Multi-agent Reinforcement Learning.
My goal is to utilize multiple modalities in order to create systems that can truly understand the semantics of our world and can effectively use this semantic knowledge in real-world interaction and prediction.
When I’m not at the computer, you can find me playing guitar, skate boarding, rock climbing, or simply running around Hyrule!
About My Mentor
Dr. Peleg currently focuses on how organisms adapt to environmental fluctuations. A major area of her research concerns honeybee clusters as she dissects how bees create a stable structure in the presence of mechanical stress. She also investigates signaling between animals, specifically the methods animals employ to amplify and interpret signals across large space and time scales. To solve this problem, Dr. Peleg uses honeybees, plants, and fireflies as model organisms. In addition, Dr. Peleg models movement ecology to demonstrate how animals translocate food over long distances. This behavior requires interplay between navigation and assessment of the environment to establish the optimal path in terms of accuracy, speed, and effort. Through several diverse projects, Dr. Peleg harnesses the power of computational tools to answer fundamental questions in biology and ecology.
About My Project
Honey bees are social insects that utilize pheromone signals to direct each other, resulting an emergent swarm behavior. Research has shown that bees can solve the shortest path problem and locate their queen by directing these pheromone signals, or “scenting”, in order to collectively create a communication network; however, little work has been done to study the behavior of individual bees in this process. We aim to show the scenting behaviors of individual honey bees directly correlates to the balance of exploration versus exploitation in the pathfinding behavior of the swarm. We utilize XMem, a state-of-the-art long-term video object segmentation tool, for individual bee tracking and segmentation. Combining this with additional computer vision approaches described in the literature, we analyze the behavior patterns of individual honeybees within the swarm. Through this analysis, we anticipate the time-series data of individual honey bees will display the balance between scenting and exploring that allows for this emergent pathfinding capability of the swarm. This improved tracking of individual honey bees will allow us to develop a better multi-agent reinforcement learning model, predicting the behaviors of the swarm.