Timezone: »

Introducing Routing Uncertainty in Capsule Networks
Fabio De Sousa Ribeiro · Georgios Leontidis · Stefanos Kollias

Wed Dec 09 09:00 AM -- 11:00 AM (PST) @ Poster Session 3 #1123

Rather than performing inefficient local iterative routing between adjacent capsule layers, we propose an alternative global view based on representing the inherent uncertainty in part-object assignment. In our formulation, the local routing iterations are replaced with variational inference of part-object connections in a probabilistic capsule network, leading to a significant speedup without sacrificing performance. In this way, global context is also considered when routing capsules by introducing global latent variables that have direct influence on the objective function, and are updated discriminatively in accordance with the minimum description length (MDL) principle. We focus on enhancing capsule network properties, and perform a thorough evaluation on pose-aware tasks, observing improvements in performance over previous approaches whilst being more computationally efficient.

Author Information

Fabio De Sousa Ribeiro (University of Lincoln)
Georgios Leontidis (University of Aberdeen)

George is an Associate Professor in Machine Learning and also the Programme Director of the highly successful and innovative MSc on AI programme. George has a strong interest in both theoretical aspects of Machine/Deep Learning, e.g. variational inference, domain adaptation, self-supervised learning etc., as well as applications, e.g. data imputation with ML in environmental data of COSMOS-UK network (PI in NERC/EPSRC project ENTRAIN), homomorphic encryption with deep learning for enabling data sharing and analytics in food industry (PI - IoFT network plus EPSRC project), anomaly detection in nuclear reactors (Co-PI in H2020 project Cortex - 20 EU partners in total), Optimising retail refrigeration systems with machine learning (Co-PI - IUK project with Tesco), forecasting yield in strawberries and tomatoes (Co-PI - EU Interreg project SmartGreen and PhD studentship), Gas Turbine availability and fault prediction with Siemens Lincoln, etc. I am serving as programme committee in various tier 1 venues, such as NeurIPS, ICML, AAAI, ICLR, IJCAI, etc. and am participating in the UK AI Council’s Data Working Group ecosystem. I am also a member of the Full College of EPSRC and have been a college member of the UKRI FLF scheme, serving as interview and sift panel member.

Stefanos Kollias (University of Lincoln)

Stefanos Kollias has been Founding Professor of Machine Learning in the Computer Science School of the University of Lincoln, UK, since September 2016. He is leading the mlearn Research Group of the University of Lincoln (mlearn.lincoln.ac.uk), composed of six lecturers, postdocs and graduate students. Formerly, he has been Professor in the Computer Science Division of the School of Electrical and Computer Engineering, National Technical University of Athens, since 1997 and Director of the Intelligent Systems, Content & Interaction Laboratory (www.image.ntua.gr). He is an IEEE Fellow (since 2015, suggested by the IEEE Computational Intelligence Society). He has been member of the Executive Committee of the European Neural Network Society, 2007-2016. He has produced a world leading research activity in the fields of machine learning, computational intelligence (with emphasis on artificial neural networks and deep neural networks), multimedia analysis, search retrieval and recognition (with emphasis on visual information), multimodal information analysis, vision, semantic content analysis and metadata interoperability, human computer interaction (affective computing, emotion and sentiment analysis, robot vision, serious games). He has published 110 papers in international journals (30 papers in IEEE Transactions and Journals) and 310 papers in proceedings of International Conferences (110 in IEEE Conferences). He has been Co-Editor of the book ‘Multimedia and the Semantic Web’, Wiley, 2005. He was the General Co-Chair of the 2016 IEEE Symposium Series on Computational Intelligence, 12/2016. He is Technical Program Co-Chair of 2017 EUSIPCO, 8/2017. His Research work is highly referenced: 3500 Citations, h-Index 28 (Scopus), 8400 Citations, h-Index 42 (Google Scholar), 6000 Citations, RGscore 38 (Research Gate). He has supervised 41 Ph.D. students at NTUA; 12 are academic professors in Greece, Europe, USA. He and his students have obtained many best paper awards in international conferences.