Two accomplished speakers will present lectures on their innovative AI research. Their topics include reinforcement learning and AI for biological sciences, with practical uses in both science and engineering. The titles and abstracts of their talks are provided below.
Title: Robust Reinforcement Learning via Adversarial Training
Speaker: Hao-Lun Hsu
Abstract:
Reinforcement Learning (RL) has proven effective in devising optimal strategies across various applications. However, traditional RL methods face challenges in dynamic and unpredictable environments with disturbances. To address this, Robust RL emerges as a solution to bolster the performance and reliability of autonomous agents. In this talk, I will introduce three research threads that align with this overarching theme. These threads will improve the limitations of standard robust RL via adversarial training with a max-min game from the perspectives of (1) efficient exploration, (2) adversarial herding, and (3) adaptive adversary. I will showcase the efficacy of these frameworks through their applications in deep brain stimulation and robotics domains.
Speaker Bio:
Hao-Lun Hsu is a second-year Computer Science Ph.D. student at Duke University. He earned his master’s in Biomedical Engineering from Georgia Tech and bachelor’s degree in Mechanical Engineering from National Taiwan University. His research is centered around the intersection of provable and practical decision-making, specifically in the realms of reinforcement learning (RL) and multi-armed bandits (MAB). His primary focus lies in the domains of robust and sample-efficient RL, with applications spanning across robotics and neuromodulation. He was a NSF TAST-NRT Fellow supported with surgical robotics research.
Title: Machine Learning in Computational Biology: Predicting Protein-DNA Binding
Speaker: Kyle Pinheiro de Oliveira
Abstract:
Research in the field of biology presents myriad opportunities for thoughtful application of machine learning. This talk revolves around work I did early in my Ph.D. on the use of machine learning models to predict the effect of protein-protein interactions on protein binding to DNA, which we call cooperativity. I begin with a brief background on the biological significance of protein binding to DNA and introduce how machine learning has been applied to prediction of such binding. I then describe the problem of predicting cooperativity and discuss the approaches I used to address it. A convolutional neural network outperformed all other models that were evaluated for the cooperativity prediction task, so I will give a brief introduction to the convolutional neural network architecture.
