Emerging topics in learning from noisy and missing data
On Sunday October 16th 2016 from 14:00 to 17:00.
Room C0.02 of the Roeterseiland complex of the University of Amsterdam.
While vital for handling most multimedia and computer vision problems, collecting large scale fully annotated datasets is a resource-consuming, often unaffordable task. Indeed, on the one hand datasets need to be large and variate enough so that learning strategies can successfully exploit the variability inherently present in real data, but on the other hand they should be small enough so that they can be fully annotated at a reasonable cost. With the overwhelming success of (deep) learning methods, the traditional problem of balancing between dataset dimensions and resources needed for annotations became a full-fledged dilemma. In this context, methodological approaches able to deal with partially described data sets represent a one-of-a-kind opportunity to find the right balance between data variability and resource-consumption in annotation. These include methods able to deal with noisy, weak or partial annotations.
In this tutorial we will present several recent methodologies addressing different visual tasks under the assumption of noisy, weakly annotated data sets. Special emphasis will be given to methods based on deep architectures for unsupervised domain adaptation, low-rank modeling for learning in transductive settings and zero-shot learning. We will show how these approaches exhibit excellent performance in crucial tasks such as pedestrian detection or fine-grained visual recognition. Furthermore, we will discuss emerging application domains which are of great interest to the multimedia community and where handling noisy or missing information is essential. For instance, we will present recent works on multimodal complex scene analysis using wearable sensors, on the estimation of physiological signals from face videos in realistic conditions, and on the recognition of emotions elicited from abstract paintings.
- Challenges in learning from noisy and missing data (N. Sebe - 30 min).
- Robust Low-Rank Modeling & Transductive Learning (X. Alameda-Pineda and E. Ricci - 60 min).
- Zero-shot Learning (T. Hospedales - 45 min).
- Weakly Supervised Domain Adaptation in Deep Neural Architectures (X. Wang - 45 min).
- X. Alameda-Pineda, Y. Yan, E. Ricci, O. Lanz, N. Sebe, "Analyzing free-standing conversational groups: a multimodal approach." ACM Multimedia, 2015.
- S. Tulyakov, X. Alameda-Pineda, E. Ricci, L. Yin, J. F. Cohn, N. Sebe, "Self-Adaptive Matrix Completion for Heart Rate Estimation from Face Videos under Realistic Conditions", IEEE CVPR, 2016.
- X. Alameda-Pineda, E. Ricci, Y. Yan, N. Sebe, "Recognizing Emotions from Abstract Paintings using Non-Linear Matrix Completion", IEEE CVPR, 2016.
- T. Xiao, T. Xia, Y. Yang, C. Huang, and X. Wang, "Learning from Massive Noisy Labeled Data for Image Classification," IEEE CVPR 2015.
- X. Zeng, W. Ouyang, and X. Wang, "Deep Learning of Scene-Specific Classifier for Pedestrian Detection," ECCV 2014.
- Y. Fu, T. M. Hospedales, T. Xiang and S. Gong, Transductive Multi-view Zero-Shot Learning, IEEE PAMI 2015.