Jeffrey M. Girard

University of Kansas

Talk title: TBA

Talk abstract: TBA

Bio: Dr. Girard studies how emotions are expressed through verbal and nonverbal behavior, as well as how interpersonal communication is influenced by individual differences (e.g., personality and mental health) and social factors (e.g., culture and context). This work is deeply interdisciplinary and draws insights and tools from various areas of social science, computer science, statistics, and medicine.

Georgios N. Yannakakis

University of Malta

Talk title: Embracing Human Feedback via Games

Talk abstract: Why is subjective and ambiguous human feedback such a critical element of every aspect of AI nowadays? What are the opportunities video games offer for the reliable collection of subjective human feedback and eventually the modeling of ambiguous phenomena such as affect? How can human feedback help us test games, represent the games we play, design creative AI algorithms, offer agency to AI, and ultimately understand player experience? In this talk, I will attempt to address these questions through a series of research studies that have led to a number of methods, tools and protocols for the reliable annotation and modeling of human feedback, in games and beyond. 

Bio: Georgios N. Yannakakis is a Professor at the Institute of Digital Games, University of Malta, and a co-founder and research director of modl.ai (Malta). He is a leading expert of the game artificial intelligence research field with core contributions in machine learning, evolutionary computation, affective computing and player modelling, computational creativity and procedural content generation. He has published over 350 journal and conference papers in the aforementioned fields and his work has received several awards from top-tier academic conferences and journals. His research has been supported by numerous European grants and has appeared in Science Magazine and the New Scientist among other venues. Georgios has been involved in a number of journal editorial boards and he is currently the Editor in Chief of the IEEE Transactions on Games and an Associate Editor of the IEEE Transactions on Evolutionary Computation. He is the co-author of the Artificial Intelligence and Games textbook and the co-organiser of the Artificial Intelligence and Games summer school series. Georgios is an IEEE Fellow.

Phil Woodland

University of Cambridge

Talk title: Modelling Annotation Ambiguity in Spoken Emotion Data

Talk abstract: Different speakers and contexts (linguistic, social, dialogue) can cause large differences in emotion expression. Furthermore, different listeners perceive certain emotions again based on differing contexts, and hence, there is inherent variability in the annotation labels of spoken emotion data. Often, data that doesn’t have a majority-agreed label is ignored in class-based spoken emotion recognition, even though this data can be valuable. This talk first describes some techniques to model and evaluate the handling of nonmajority agreed data via extra classes, the use of evidential deep learning (EDL) to quantify uncertainty, and also extending EDL to model emotion distributions. An EDL-based approach is also developed and applied for attribute-based emotion annotations. An alternative approach of variability aware human annotator simulation is then discussed based on conditional softmax flow (for classes) and conditional integer flow (for attributes). It is demonstrated that the approaches discussed can both yield excellent classification accuracy/mean prediction as well as matching the distribution of human annotators. The talk concludes with conclusions and outlook.

Bio: Prof. Woodland’s research is in the area on speech and language technology with a major focus on developing all aspects of speech recognition systems. After working with British Telecom Research Labs for three years, he returned to a Lectureship at the University of Cambridge in 1989 and became a Reader in 1999 and a (full) Professor in 2002. He has authored or couthored more than 250 papers in the area of speech and language technology with a main focus on speech recognition systems. He was the recipient of number of Best Paper awards including for work on speaker adaptation and discriminative training. He is one of the original coauthors of the HTK toolkit and has continued to play a major role in its development. He was a Member of the Editorial Board of Computer Speech and Language (1994–2009) and is currently a Member of the Editorial Board Member of Speech Communication. He is a Fellow of the International Speech Communication Association, the IEEE and the Royal Academy of Engineering. His group has developed a number of techniques that have been widely used in large vocabulary systems including standard methods for transform-based adaptation and discriminative sequence training. He has worked on the use of deep neural networks for both acoustic models and language models. His current work has a focus on the use and development of end-to-end trainable neural network systems.

Vidhyasaharan Sethu
v.sethu@unsw.edu.au
(University of New South Wales)

Bernd Dudzik
b.j.w.dudzik@tudelft.nl
(Delft University of Technology)

Carlos Busso
busso@utdallas.edu
(University of Texas at Dallas)

Emily Mower Provost
emilykmp@umich.edu
(University of Michigan)