[cs-talks] Upcoming CS Seminars: NRG (Mon) + IVC (Tues) + BUSec (Wed) + Student Sem (Thurs) + Data Sem (Fri)

Greenwald, Faith fgreen1 at bu.edu
Mon Nov 30 10:39:35 EST 2015

NRG Seminar
Reducing Virtual Machine networking complexity
Sander Vrijders, University College Ghent, Belgium
Monday, November 30, 2015 at 11am in MCS 148

Abstract: Virtualization is an enabling technology that improves scalability,reliability and flexibility. Virtualized networking is tackled by emulating or paravirtualizing Network Interface Cards (NICs). This approach, however, leads to complexities (implementation and management) and has to conform to some limitations imposed by the Ethernet standard. The Recursive InterNetwork Architecture (RINA) is a recently proposed network architecture based on first principles. RINA turns the current approach to virtualized networking on its head: instead of emulating networks to perform inter process communication on a single processing system, it sees networking as an extension to local inter-process communication. We show how RINA can leverage a paravirtualization approach to achieve a more manageable solution for virtualized networking. We also present experimental results performed on IRATI, the reference open source implementation of RINA, which shows the potential performance that can be achieved by deploying our solution.

Bio: Sander Vrijders received his M.Sc. Degree in applied engineering: computer science in 2012 from University College Ghent, Belgium. Since then he has been working at the Internet Based Communications Networks and Services group at Ghent University, where he is a PhD candidate. His current interests are in future network architectures such as the Recursive InterNetwork Architecture, funded through FP7 IRATI and PRISTINE.

IVC Seminar
Using scene context to track the (almost) invisible
Vitaly Ablavsky
Tuesday, December 1, 2015 at 2pm in MCS 148

Abstract: Automatic video-based analysis of dynamic scenes has applications in many domains including content-based retrieval, surveillance, and autonomous navigation. In practice, automatic scene analysis often relies on identifying and tracking salient objects in the scene. However, in many real-world scenarios salient objects follow complex motion patterns, their apparent size is small, and they undergo occlusions. Object tracking under such conditions remains a challenge.

Our approach to tracking (almost) invisible objects is to design scene models that include information about the global context. This talk focuses on two such scene models that we developed to solve challenging real-world problems.

The first scene model addresses the problem of tracking people in non-overlapping camera views in the presence of "relocatableoccluders" (e.g., vehicles in parking lots, shopping carts in supermarkets). Handling occlusions is essential for reliable tracking, but learning an occlusion map is impossible when the occluding objects are relocatable.  In our approach, we formulate an occluder-centric representation, called a graphical model layer, where a person’s motion in the ground plane is defined as a first-order Markov process on activity zones, while image evidence is aggregated in 2D observation regions that are depth-ordered with respect to the occlusion mask of the relocatable object. We represent real-world scenes as a composition of depth-ordered, interacting graphical model layers, and account for image evidence in a way that handles mutual overlap of the observation regions and their occlusions by the relocatable objects.

The second scene model addresses the problem of ball-tracking in team sports, such as basketball and soccer. Although multiple overlapping views are available, the players often shield the ball from the opponents, the ball's motion patterns tend to be complex, its small apparent size is small, and the images are corrupted by motion blur. These conditions make per-view ball detection unreliable. In our approach, we formulate a scene representation that models correlations between the ball and the players. Given this representation, the problem of tracking the ball takes the form of first tracking the players, and then inferring which player, if any, possesses the ball.

We validate our scene representations (for pedestrian tracking and ball tracking) by implementing end-to-end systems that track objects of interest over extended periods of time.  The tracking accuracy of our systems compares favorably with the state of the art.

Bio: Vitaly Ablavsky is a member of the Video and Image Understanding group at Systems & Technology Research in the Boston area. Previously he was a post-doctoral researcher in the Computer Vision Laboratory (CVLab) at EPFL, working with Prof. Pascal Fua and Prof. Vincent Lepetit. Prior to his appointment at EPFL, he conducted research toward his Ph.D. in the Image and Video Computing group (IVC) at Boston University, advised by Prof. Stan Sclaroff. His interests are object tracking and machine learning with applications to video-based surveillance, action recognition, and segmentation.

BUSec Seminar
Optimal-Rate Non-Committing Encryption in a CRS Model
Oxana Poburinnaya, BU
Wednesday, December 2, 2015 at 10am in MCS 180- Hariri Seminar Room

Abstract: Non-committing encryption (NCE) was introduced in order to implement secure channels under adaptive corruptions in situations when data erasures are not trustworthy. In this work we are trying to optimize the rate of NCE, i.e. the number of bits one needs to send in order to transmit a single bit of a plaintext.

In initial constructions (e.g. Canetti, Feige, Goldreich and Naor, STOC 96) the length of both the receiver message, namely the public key, and the sender message, namely the ciphertext, is m*poly(k) for an $m$-bit message, where k is the security parameter. Subsequent works improve efficiency significantly. Specifically, the work of Hemenway, Ostrovsky and Rosen (TCC 15) achieves O(m log m)+poly(k) ciphertext size, under the Phi-hiding assumption. Still, the public key (which can be used for only a single message) has size m * poly(k), and thus the protocol requires sending poly(k) bits for each bit of a plaintext.

We show the first construction of a constant-rate NCE. In fact, our public key has size only poly(k), and the ciphertext size is m+poly(k), which is comparable to efficiency of a plain semantically secure encryption. We also need a common reference string (CRS) of size poly(m*k), but the CRS is reusable for an arbitrary polynomial number of m-bit messages. We assume one way functions and indistinguishability obfuscation for circuits.

In addition, our NCE protocol is the first NCE protocol with perfect correctness.

Joint work with Ran Canetti and Mariana Raykova

Student Seminar
Applying for Grants and Funding
Chris DeVits
Thursday, December 3, 2015 at 12pm in MCS 148

Data Seminar
Process Trace Clustering: A Heterogeneous Information Network Approach
Vatche Ishakian
Friday, December 4, 2015 at 11am in MCS 148

Abstract: Process mining is the task of extracting information from event logs, such as ones generated from workflow management or enterprise resource planning systems, in order to discover models of the underlying processes, organizations, and products. As the event logs often contain a variety of process executions, the discovered models can be complex and difficult to comprehend. Trace clustering helps solve this problem by splitting the event logs into smaller subsets and applying process discovery algorithms on each subset, resulting in per-subset discovered processes that are less complex and more accurate. However, the state-of-the-art clustering techniques are limited: the similarity measures are not process-aware and they do not scale well to high-dimensional event logs. In this paper, we propose a conceptualization of process's event logs as a heterogeneous information network, in order to capture the rich semantic meaning, and thereby derive better process-specific features. In addition, we propose SeqPathSim, a meta path-based similarity measure that considers node sequences in the heterogeneous graph and results in better clustering. We also introduce a new dimension reduction method that combines event similarity with regularization by process model structure to deal with event logs of high dimensionality. The experimental results show that our proposed approach outperforms state-of-the-art trace clustering approaches in both accuracy and structural complexity metrics.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs-mailman.bu.edu/pipermail/cs-talks/attachments/20151130/30e105de/attachment.html>

More information about the cs-talks mailing list