[cs-talks] Upcoming Seminars: CS/Hariri (Thurs) + IVC (Thurs) + Ph.D Defense (Mon)
cs at bu.edu
Wed Apr 22 12:35:15 EDT 2015
Tamung Uncertainty, Scale, and Change: A Programming Language Perspective
Suresh Jagannathan, Purdue
Thursday, April 23, 2015 at 3pm in MCS 180 — Hariri Seminar Room
Abstract: The modern-day software ecosystem is a messy and chaotic one. Among other things, it includes an intricate stack of sophisticated services and components, susceptible to frequent (and often incompatible) upgrades and patches; emerging applications that operate over large, unstructured, and noisy data; and, an ever growing code base replete with latent defects and redundancies. Devising novel techniques to tame this complexity, and improve software resilience, trustworthiness, and expressivity in the process, is a common theme actively being explored by several ongoing DARPA programs. This talk gives an overview of three such efforts - PPAML (Probabilistic Programming Advancing Machine Learning), MUSE (Mining and Understanding Software Enclaves), and BRASS (Building Resource Adaptive Software Systems). These programs have seemingly disparate goals - PPAML seeks to democratize machine learning through the use of probabilistic programming abstractions; MUSE aims to exploit predictive analytics over large software corpora to repair and synthesize programs; and, BRASS is concerned with devising self-adaptive software capable of automatically responding to changes in its operating environment. Despite their outward differences, however, all three programs nonetheless critically rely on common foundational advances in programming language design, analysis, and implementation to realize their vision, and share an overarching goal to revolutionize the way we think about software construction and reliability.
Bio: Suresh Jagannathan joined the Information Innovation Office (I2O) at DARPA as a Program Manager in 2013. He is currently on leave from Purdue University where is a Professor of Computer Science. He has been a visiting faculty scholar at Cambridge University, and prior to joining Purdue, was a Senior Research Scientist at the NEC Research Institute. His interests are in programming languages generally, with specific interests in program verification and analysis, concurrent and distributed systems, functional programming, and compiler design. He received his Ph.D from MIT.
The Design and Use of MIT Sloop Retrieval Engine for Animal Biometrics
Sai Ravela, MIT
Thursday, April 23, 2015 at 4pm in MCS B29
Abstract: Identifying individuals in photographs of animals collected over time is a non-invasive approach for ecological monitoring and conservation. This paper describes the design and use of Sloop (sloop.mit.edu), for animal biometrics incorporating crowd-sourced relevance feedback. Sloop's iterative retrieval strategy using hierarchical and aggregated matching and relevance feedback consistently improves deformation and correspondence-based approaches across several species. Its crowdsourcing strategy is successful in utilizing relevance feedback on a large scale. Sloop is in operational use. The user experience and results are presented here to facilitate the creation of a community-based ecological informatics system for conservation planning.
Bio: Sai Ravela directs the Earth Signals and Systems Group (ESSG) in the Earth, Atmospheric and Planetary Sciences at the Massachusetts Institute of Technology. His primary research interests are in stochastic systems science with application to Earth, Atmospheric and Planetary Sciences. He conducts this research through various projects including Autonomous Observation (caos.mit.edu), Animal Biometrics (sloop.mit.edu), Fluid Imaging (flux.mit.edu), Statistical Inference for Coherent Fluids (stics.mit.edu) and Hurricane Risk(hazmet.mit.edu). Dr. Ravela completed a PostDoc in Atmospheric Science and Stochastic Systems from MIT in 2004, and received a PhD in 2003 in Computer Science from the University of Massachusetts at Amherst, specializing in Computer Vision, Multimedia Retrieval and Robotics. He is the co-founder of Windrisktech LLC, a company that uses Learning and Physics to quantify risk from hurricanes, and E5 Aerospace LLC, that builds novel designs of aircraft systems for autonomous observation.
Dora Erdos, BU
Monday, April 27, 2015 at 10am in MCS 148
Abstract: In this thesis we investigate two topics in data mining on graphs; in the first part we investigate the notion of centrality in graphs, in the second part we look at reconstructing graphs from aggregate information. In many graph related problems the goal is to rank nodes based on an importance score. This score is in general referred to as node centrality. In Part I. we start by giving a novel and more efficient algorithm for computing betweenness centrality. In many applications not an individual node but rather a set of nodes is chosen to perform some task. We generalize the notion of centrality to groups of nodes. While group centrality was first formally defined by Everett and Borgatti, we are the first to pose it as a combinatorial optimization problem "find a group of k nodes with largest centrality". We give an algorithm for solving this optimization problem for a general notion of centrality that subsumes various instantiations of centrality that find paths in the graph. We prove that this problem is NP-hard for specific centrality definitions and we provide a universal algorithm for this problem that can be modified to optimize the specific measures. We also investigate the problem of increasing node centrality by adding or deleting edges in the graph. We conclude this part by solving the optimization problem for two specific applications; one for minimizing redundancy in information propagation networks and one for optimizing the expected number of interceptions of a group in a random navigational network. In the second part of the thesis we investigate what we can infer about a bipartite graph if only some aggregate information -- the number of common neighbors among each pair of nodes -- is given. First, we observe that the given data is equivalent to the dot-product of the adjacency vectors of each node. Based on this knowledge we develop an algorithm that is based on SVD-decomposition, that is capable of almost perfectly reconstructing graphs from such neighborhood data. We investigate two versions of this problem, in the versions the dot-product of nodes with themselves, a.k.a. the node degrees, are either known or hidden.
Committee: Evimaria Terzi Azer Bestavros Pauli Miettinen Mark Crovella George Kollios (chair)
Data Science Distinguished Lecture
Learning and Inference for Natural Language Understanding
Dan Roth, University of Illinois at Urbana/Champaign
Monday, April 27, 2015 at 3pm in MCS 180 — Hariri Seminar Room
Abstract: Machine Learning and Inference methods have become ubiquitous and have had a broad impact on a range of scientific advances and technologies and on our ability to make sense of large amounts of data. Research in Natural Language Processing has both benefited from and contributed to advancements in these methods and provides an excellent example for some of the challenges we face moving forward.
In his talk, Dan Roth will describe his research in developing learning and inference methods in the pursuit of natural language understanding. Roth will delve deeper into what he believes are key challenges including:
· Learning models from natural interactions without direct supervision
· Knowledge acquisition and the development of inference models capable of incorporating knowledge and reason
· Scalability and adaptation - learning to accelerate inference during the lifetime of a learning system
Within the unified computational framework of Constrained Conditional Models (CCMs) – an Integer Linear Programming formulation that augments statistically learned models with declarative constraints as a way to support learning and reasoning – Roth will discuss old and new results pertaining to learning and inference and how they are used to push forward our ability to understand natural language.
Bio: Dan Roth is a Professor in the Department of Computer Science and the Beckman Institute at the University of Illinois, Urbana-Champaign and a University of Illinois scholar. Roth has published broadly in machine learning, natural language processing, language theory, and knowledge representation and reasoning. He has also developed advanced machine learning based tools for natural language applications that are being used widely by the research community and commercially. Roth is the Associate Editor-in-Chief of the Journal of Artificial Intelligence Research (JAIR) and will serve as Editor-in-Chief for a two-year term beginning in 2015. He was the program chair of AAAI’11, ACL’03 and CoNLL'02. Roth received his BA summa cum laude in Mathematics from Technion in Israel and his PhD in Computer Science from Harvard University in 1995.
Data Management Seminar
Matrix Factorizations Over Non-Conventional Algebras in Data Mining
Pauli Miettinen, Max Planck Institute for Informatics
Tuesday, April 28, 2015 at 11am in MCS 148
Abstract: Matrix factorization methods---and more broadly, linear-algebra-based methods---are well-used and well-loved in data mining. Most popular methods are based on the standard linear algebra, but recent years have seen increased interest in methods based on non-conventional algebras, most notably, the Boolean algebra. Using non-standard algebras can have multiple benefits: the factor matrices can be sparser or easier to interpret, for example, but most importantly, it will allow us to find structures different to those we find using the standard linear algebra. In this talk, I will cover the basic ideas behind the Boolean matrix factorization and its recent developments and applications both in data mining and in related areas. We will also see some other non-standard algebras, such as tropical and subtropical algebras, and their potential applications to data mining problems. In the end, I will briefly discuss ongoing work on generalizing the concept of outer product and approaching the matrix factorizations from this point of view.
Bio: Pauli did his PhD at University of Helsinki, Finland, in Prof. Heikki Mannila's group, where he graduated at 2009. After a short post-doc period at Helsinki, he moved to Max-Planck Institute for Informatics in Saarbrücken, Germany, where he is currently a senior researcher and the leader of research field Data Mining. His current research focus is on redescription mining and non-conventional matrix and tensor factorizations and their applications to data mining. His work has appeared in numerous publications in top data mining and theoretical computer science venues. He has received two best paper awards and an honorary mention at 2010 ACM SIGKDD Doctoral dissertation awards. While he insists that 1+1=1, he accepts that other people might have different opinions.
Poly. Secure Integrated Circuit (IC) Fabrication Using Obfuscation
Siddharth Garg, NYU
Wednesday, April 29, 2015 at 10am in MCS 180 — Hariri Seminar Room
Abstract: For economic reasons, the fabrication of digital ICs is increasingly outsourced. This comes at the expense of trust - the untrusted fabrication facility ("foundry") could pirate the intellectual property of the IC designer, or worse, maliciously modify the IC to leak secret information from the chip or sabotage its functionality. In this talk, I will present my recent work on two defense mechanisms based on hardware obfuscation to secure computer hardware against such attacks. The first is split manufacturing, which enables a designer to partition a digital circuit across multiple chips, fabricate each separately, and "glue" them together after fabrication. Since each foundry only sees a part of the netlist, its ability to infer the design intent is hindered. I will propose a quantitative notion of security for split manufacturing and explore the resulting cost-security trade-offs. In the second part of the talk, I will discuss another defense mechanism - IC camouflaging. IC camouflaging allows for the Boolean functionality of a gate to be hidden from the attacker. Previous work indicates that if a carefully selected subset of gates in the netlist is camouflaged, an attacker is forced to use a "brute-force search" to decamouflage the circuit. I will present an attack that demonstrates that IC camouflaging is, in fact, less effective than previously thought. I will conclude with some preliminary thoughts on provably secure IC fabrication and how it relates to the foundational work on function obfuscation.
Multi-Modal Visual Attention Capturing, Processing, Exploitation and Dissemination
Dr. Riad Hammoud
Thursday, April 30, 2015 at 5pm in MCS B29
Abstract: Estimation and dissemination of visual attention is useful in many application fields including human-machine interaction, forward collision warning systems/safe driving, inspection/surgery training and scene monitoring. Video/Audio among other sensing modalities are widely employed to capture raw sensing data (e.g., forward scene, eye/head images, verbal annotations). The first part of my talk focuses on low-level methods for video/signal data processing including eye-gaze estimation and moving objects (i.e., pedestrians, vehicles) detection from a moving vehicle. The second part highlights a higher level of data processing (a.k.a., exploitation) to generate semantic representations (i.e., eye dwelling/pattern, graphs of attributes). The third and final part of this talk discusses various multi-modal visual attention dissemination cases and data-association techniques, in practical applications including safe driving, webpage design, training, video indexing and retrieval.
Bio: Dr. Riad I. Hammoud holds a PhD in Computer Vision and Robotics from INRIA since February 2001. He developed several vision and perception systems for automotive, monitoring, geo-localization, robotics, assistive/eye-tracking and biometrics applications. He published numerous conference and journal papers, book chapters, Springer books and patents on object detection, overhead tracking, registration, face recognition, image classification, geo-localization, and data association and fusion. He serves as Principal Investigator, Algorithm/Team lead and Senior Principal Research Engineer on several R&D programs. He also serves as Associate Editor, Area Chair and technical program committee member of top computer vision conferences, workshops and transactions. In 2015 he organizes the 11th IEEE CVPR WS on Perception Beyond the Visible Spectrum and serves as Guest Editor of a special issue of Int'l Journal of Computer Vision (IJCV) on Large-Scale Media Geo-Localization.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cs-talks