Abstract: Training a neural network involves a number of choices including determining the structure and hyperparameters of the model to selecting and/or constructing a new dataset. Reducing the cost of training a neural network can help reduce the environmental impact of AI models as well as support a wider population to take advantage of higher performing models that also come with increased comptuational cost. In this talk I will discuss recent work in the Image and Video Computing group at Boston University that improves efficiency by re-using parameters across layers in a neural network as well as incorperating new knowledge into an existing model without retraining from scratch, such as adding a new modality to large pretrained model. I will conclude with a discussion on addressing challenges to training a model that is fair across subpopulations when some groups are overrepresented within a dataset.

Bio: Bryan Plummer is an Assistant Professor in the Department of Computer Science at Boston University, and is a core faculty member of the Artificial Intelligence Research (AIR) Initiative in the Rafik B. Hariri Institute for Computing and Computational Science & Engineering. His research interests include multimodal reasoning, detecting manipulated and machine generated media, efficient neural networks, fair and explainable AI, and disentangled and structured representation learning. Bryan obtained his PhD in the computer vision group at the University of Illinois at Urbana-Champaign where he received a 3M Foundation Fellowship and was an NSF GRFP honorable mention.”