Special Seminar: Luis Leiva "Synthetic Generation of Human-like Data"

Synthetic Generation of Human-like Data

Luis Leiva

Abstract:

For a long time, there has been an intense debate in the scientific community of whether it is more important to aim for good data or good algorithms. Given the current democratization of the Machine Learning (ML) landscape --including e.g. open-source frameworks such as TensorFlow or MXNet, affordable and specialized chipsets such as GPUs and TPUs, or highly scalable computing platforms such as Google Cloud or AWS-- it seems clear that today data is the key player. It is particularly crucial to many Human-Computer Interaction (HCI) tasks, where typically a set of representative users is required to annotate as much behavioral data as possible in order to e.g. develop a novel interaction technique or analyze how an application is operated. This is a very time-consuming and costly process, and can be prohibitive to many individuals, research labs, and startups. To bridge this gap, I am working on ML techniques to create synthetic but realistic data. In this talk, I share my vision about artificially generated human-like data and how it can be used to create or enrich existing datasets, unburdening thus both the users and the experimenter. Researchers can also benefit from producing synthetic material in order to conduct experiments at low risk and build data-driven simulations of user behavior. Finally, artificially generated data can be especially useful for a more effective training of data-hungry algorithms such as deep neural networks.

Bio:

Luis A. Leiva is a postdoctoral scholar at Aalto University. He holds two BSc engineering degrees, an MSc in electrical engineering, and a PhD in computer science. He is the former CTO of Sciling, an SME agency specialized in Machine Learning solutions. Previously he was a research fellow of the PRHLT Research Center at the Technical University of Valencia, Spain. His research interests lie at the intersection of Human-Computer Interaction and Machine Learning. He is currently conducting research on computational interaction. He is particularly interested in artificial data generation techniques.