Title: Learning to Represent and Reason through Visual Puzzles – the Role of Biases
Abstract: Computational learning approaches to solving visual reasoning tests, such as Raven’s Progressive Matrices (RPM), critically depend on the ability to identify the visual concepts used in the test (i.e., the representation) as well as the latent rules based on those concepts (i.e., the reasoning). However, learning of representation and reasoning is a challenging and ill-posed task, often approached in a forward manner (first representation, then reasoning). In this talk I will consider two factors that play key roles in learning to represent and reason; one is the inductive reasoning bias and the other the representation bias. For the former, I will discuss a general generative graphical model for RPMs, embedded in the novel learning framework called DAREN (Disentangling-based Abstract Reasoning Network), and how we applied it to solve the reasoning tests. For the latter, our model named SAViR-T (Spatially Attentive Visual Reasoning with Transformers) injects representation bias in the form of spatio-visual tokens to learn the intra-image as well as the inter-image token dependencies.
You can learn more about DAREN, the ICPR’22 best student paper, in https://arxiv.org/abs/2109.13156 and SAViR-T, soon-to-be-presented at ECML-PKDD, in https://arxiv.org/abs/2206.09265.
Bio: Vladimir Pavlovic is a Professor of Computer Science at Rutgers University in New Jersey, USA. From 2018 until 2021 he was also a Principal Scientist and Director/Head of Future Interactions at the Samsung AI Center in Cambridge, UK. Vladimir’s research interests include probabilistic machine learning, mulimodal representation learning, and next generation human sensing. Over the past twenty years Vladimir has published extensively in the domains of computer vision and human-computer interaction, including his seminal works on human gesture modeling and recognition, human motion analysis, and non-verbal human affect understanding. At Rutgers, he co-leads the Center for Accelerated Real Time Analytics (CARTA), https://carta.rutgers.edu, is a member of the Executive Committee of Rutgers Center for Cognitive Science (RUCCS), https://ruccs.rutgers.edu, and an associate member of the Center for Quantitative Biology (CQB), https://cqb.rutgers.edu, and Management Science and Information Systems at Rutgers Business School. Vladimir received his Ph.D. in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign.
Please click this URL to start or join. https://iastate.zoom.us/j/96810972944?pwd=SVVLWlY2cVdZYXhxWWg4ZHF1cVdSZz09
Or, go to https://iastate.zoom.us/join and enter meeting ID: 968 1097 2944 and password: 334840
Join from dial-in phone line:
Dial: +1 309 205 3325 or +1 312 626 6799
Meeting ID: 968 1097 2944
Participant ID: Shown after joining the meeting
International numbers available: https://iastate.zoom.us/u/aqUgrVklM