[cs-talks] CS Upcoming Seminars: PhD Defense (Tues) + BUSec (Wed) + IVC (Thurs)
fgreen1 at bu.edu
Mon Mar 28 14:03:17 EDT 2016
PhD Thesis Defense
Learning Space-Time Structures for Action Recognition and Localization
Shugao Ma, BU
Tuesday, March 29, 2016 at 9:30am in MCS 148
Abstract: Abstract: In this thesis the problem of automatic human action recognition and localization in videos is studied. In this problem, our goal is to recognize the category of the human action that is happening in the video, and also to localize the action in space and/or time. This problem is challenging due to the complexity of the human actions, the large intra-class variations and the distraction of backgrounds. Human actions are inherently structured patterns of body movements. However, past works are inadequate in learning the space-time structures in human actions and exploring them for better recognition and localization. In this thesis new methods are proposed that exploit such space-time structures for effective human action recognition and localization in videos, including sports videos, YouTube videos, TV programs and movies. A new local space-time video representation, the hierarchical Space-Time Segments, is first proposed. Using this new video representation, ensembles of hierarchical spatio-temporal trees, discovered directly from the training videos, are constructed to model the hierarchical, spatial and temporal structures of human actions. This proposed approach achieves promising performances in action recognition and localization on challenging benchmark datasets. Moreover, the discovered trees show good cross-dataset generalizability: trees learned on one dataset can be used to recognize and localize similar actions in another dataset. To handle large scale data, a deep model is explored that learns temporal progression of the actions using Long Short Term Memory (LSTM), which is a type of Recurrent Neural Network (RNN). Two novel ranking losses are proposed to train the model to better capture the temporal structures of actions for accurate action recognition and temporal localization. This model achieves state-of-art performance on a large scale video dataset. A deep model usually employs a Convolutional Neural Network (CNN) to learn visual features from video frames. The problem of utilizing web action images for training a Convolutional Neural Network (CNN) is also studied: training CNN typically requires a large number of training videos, but the findings of this study show that web action images can be utilized as additional training data to significantly reduce the burden of video training data collection.
Stan Sclaroff (Advisor) - Computer Science, Boston University
Margrit Betke - Computer Science, Boston University
Leonid Sigal - Disney Research Pittsburgh
George Kollios - Computer Science, Boston University
Mark Crovella (Chair) - Computer Science, Boston University
Uncovering Cryptographic Failures with Internet-Wide Measurement
Zakir Durumeric, University of Michigan
, March 30, 2016 at 9:45am in MCS 180- Hariri Seminar Room
Abstract: Despite advances in cryptography, there remains a significant gap between developed algorithms and how systems are protected in the real world. In this talk, I will discuss two studies in which Internet-wide measurement has uncovered catastrophic cryptographic failures in practice. In the first, we investigate the Diffie-Hellman key exchange, finding it far less secure than widely believed. I'll present Logjam, a novel flaw in TLS that lets a man-in-the-middle downgrade connections to “export-grade” Diffie-Hellman, and then go on to consider how a small number of fixed or standardized groups may allow for passive eavesdropping by nation-state attackers.
Next, I'll discuss our recent analysis of mail delivery security. We find that the top mail providers all proactively encrypt and authenticate messages. However, these best practices have yet to reach widespread adoption with only one third of top domains successfully configuring encryption and 1% supporting mail authentication. Unfortunately, this patchwork has led to an ecosystem where servers favor failing open to allow gradual deployment. We find that downgrade attacks are commonplace in the real world and highlight seven countries where more than 20% of inbound Gmail messages arrive in cleartext due to network attackers
Efficient Diffeomorphisms: From Representation to Inference
Thursday, March 31, 2016 at 2pm in MCS 148
Abstract: We propose novel well-behaved transformations, obtained by (fast and highly-accurate integration) of continuous piecewise-affine velocity fields. The proposed method is simple yet highly expressive, effortlessly handles optional constraints (e.g., volume preservation), and supports convenient modeling choices (e.g., smoothing priors, coarse-to-fine analysis). Importantly, the proposed approach, partly due to its rapid likelihood evaluations and partly due to its other properties, facilitates tractable inference over rich transformation spaces, including using methods based on Markov-Chain Monte Carlo. I will discuss several computer-vision and machine-learning applications, including better data augmentation for training image classifiers.
Publications (ICCV '15; AISTATS '16; PAMI submission under review) and other links related to this talk can be found in
Oren Freifeld is a postdoc at John Fisher's Sensing, Learning and Inference group at MIT CSAIL. His main research areas are computer vision, probabilistic modeling, statistical inference, and machine learning. He earned his PhD (advisor: Michael Black) and ScM in Applied Mathematics from Brown University. During his PhD he was also a Visiting Scholar at Stanford University Electrical Engineering department (host: Krishna Shenoy). Earlier, he graduated from Tel-Aviv University with MSc (advisors: Hayit Greenspan and Jacob Goldberger) and BSc in Biomedical Engineering.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cs-talks