Time: 10:00 – 11:00 a.m.
Location: Building 5700, Conference Room E104
Vast Deep learning has been advancing the state-of-the-art in artificial intelligence and achieved great performance in a wide range of application areas. However, the lack of explanation and interpretation of deep learning models from the learned representations to the underlying decision process and absence of control over their internal processes are major drawbacks in the development process of the models. So, deep learning model developers face a lot of trial and error during the development process and spend their efforts in developing their models through analyzing and understanding the results. As such, solutions are needed that help the developers not only understand the outcomes but also suggest the ways to improve their model.
In this talk, I will present visual analytics approaches for understanding and improving deep learning models. First, I will introduce a visual analytics system for exploring classification results, identifying misclassified samples, examining the predicted score distributions of samples, and showing how the outcomes progressively change during the training process. Second, I will present two techniques to visualize the evolutionary processes of deep learning models that based on evolutionary approaches to searching the possible space of network hyperparameters and structure. The final one is for visualizing embeddings of molecular dynamics (MD) simulations to not only evaluate and explain our embedding model but also analyze various characteristics of the simulations.
Junghoon Chae is currently a Postdoctoral Research Associate in the Computational Data Analytics group, Computer Science and Mathematics Division at Oak Ridge National Laboratory. He earned a Ph.D. in the School of Electrical and Computer Engineering at Purdue University, working with Dr. David Ebert. His research expertise and interests are, but not limited to, in the areas of visual analytics for machine learning and deep neural network for explaining and understanding them clearly and visualization of large-scale data for decision making by integrating machine learning, statistical techniques, and human perception.
Host: Chad Steed, Computational Data Analytics Team Lead, 574-7168