• Upcoming
• Past

• Upcoming
• Past
• Add New Seminar

Series: Phd Colloquium
Title: Designing energy-aware optimization techniques through program behaviour analysis

• Speaker: Ananda Vardhan K
• Faculty Advisor: Y.N. Srikant
• Date and Time: Wednesday, February 10, 2010, 4:00 PM
• Venue: CSA Seminar Hall (Room No. 254, First Floor)

Abstract
Modern programs are compute and memory intensive. Designing computing systems that cater to these programs pose major challenges with respect to the power as well as performance. Designing efficient solutions for these challenges require understanding different aspects of program behaviour. The main objective of the thesis is to design techniques that reduce energy consumption while incurring minimal performance penalty by understanding the program behaviour. Specifically, this work focuses on three hardware assisted compiler-directed techniques that reduce leakage and dynamic energy consumption in functional units and data caches.

In the first part, we study the leakage energy consumption in data caches. We use a novel (context sensitive) profile based technique to capture the heap based data structures in the form of disjoint data regions. We then use critical path analysis to identify non-critical data regions. We propose an optimization that reduces leakage in the data caches by fetching non-critical data regions into the drowsy caches. Our approach achieves 30-40% energy savings with minimal performance degradation.

In the second part, we study the dynamic energy consumption in data caches. We analyze the locality characteristics of the identitled disjoint data regions. We propose a novel reuse distance measurement algorithm, that uses partitioned stacks; traditional techniques use a monolithic stack. We apply this algorithm to an apriori-based selection scheme to identify subsets of data regions that have low locality. A new heterogeneously partitioned cache architecture is proposed, where low locality regions are allocated to larger partitions, while the high locality regions are alloted to smaller partitions. This approach reduces energy consumption in the data caches by 25% with minimal performance degradation.

In the third part, we study the leakage energy consumption in functional units. We propose a new instruction scheduling technique called Transition Aware Scheduling,that increases the length of the continuous idle periods, where the functional units are turned off. This approach reduces energy consumption in functional units by 25% without hampering the performance.

Speaker Bio:
Ananda Vardhan is a Ph.D student in the CSA department. His research interests are in the area of energy-aware optimizations via both architecture innovations and program analysis.

Series: Department Seminar
Title: Discovery of Application Workloads from Network File trace

• Speaker: Ms. Neeraja Yadwadkar
• Date and Time: Tuesday, January 19, 2010, 3:30 PM
• Venue: CSA Seminar Hall (Room No. 254, First Floor)

Abstract
An understanding of Input/Output data access patterns of applications is useful in several situations. First, gaining an insight into what applications are doing with their data at a semantic level helps in designing efficient storage systems. Second, it helps to create benchmarks that mimic realistic application behavior closely. Third, it enables autonomic systems as the information obtained can be used to adapt the system in a closed loop. All these use cases require the ability to extract the application-level seman- tics of I/O operations. Methods such as modifying application code to associate I/O operations with semantic tags are intrusive. It is well known that network file system traces are an important source of information that can be obtained non-intrusively and analyzed either online or offline. These traces are a sequence of primitive file system operations and their parameters. Simple counting, sta- tistical analysis or deterministic search techniques are inadequate for discovering application-level semantics in the general case, because of the inherent variation and noise in realistic traces. In this paper, we describe a trace analysis methodology based on Profile Hid- den Markov Models. We show that the methodology has powerful discriminatory capabilities that enables it to recognize applications based on the patterns in the traces, and to mark out regions in a long trace that encapsulate sets of primitive operations that represent higher-level application actions. It is robust enough that it can work around discrepancies between training and target traces such as in length and interleaving with other operations. We demonstrate the feasibility of recognizing patterns based on a small sampling of the trace, enabling faster trace analysis. Preliminary experiments show that the method is capable of learning accurate profile models on live traces in an online setting. We present a detailed evaluation of this methodology in a UNIX environment using NFS traces of selected commonly used applications such as compilations as well as on industrial strength benchmarks such as TPC-C and Postmark, and discuss its capabilities and limitations in the context of the use cases mentioned above.

Speaker Bio:
Neeraja Yadwadkar is a Masters Student in CSA Dept, IISc. This talk is a practice is based on a recently accepted paper at FAST 2010

Series: Department Seminar
Title: On Cryptographic Protocols Employing Asymmetric Bilinear Pairings

• Speaker: Dr. Sanjit Chatterjee, University of Waterloo, Canada
• Date and Time: Monday, January 18, 2010, 2:30 PM
• Venue: CSA Seminar Hall (Room No. 254, First Floor)

Abstract
Bilinear pairing has turned out to be an extremely useful instrument in the cryptographer's toolbox. Many cryptographic primitives are realized for the first time using such a pairing. Ingenious applications include novel solution to the problem of identity-based encryption, three-party key agreement and short signatures to name a few. A cryptographically suitable bilinear pairing is defined over elliptic curve groups and may take different forms depending upon the underlying mathematical structure. In this talk we take a close look at the functionality, efficiency and security aspects of some well-known cryptographic protocols that are implemented in the setting of asymmetric bilinear pairing.

Speaker Bio:
Dr. Sanjit Chatterjee received his Ph.D from ISI Kolkata in 2006. He is currently a post doctoral fellow at the Centre for Applied Cryptographic Research at the University of Waterloo, Canada. His research interests are in public key encryption and identity-based encryption, key agreement protocols, pairing-based cryptography and provable security.

Series: Msc(Engg) Thesis Defense
Title: Near-Duplicate Detection Using Instance Level Constraints

• Speaker: Mr. Patel Vishal Nandkishor
• Faculty Advisor: Prof. Chiranjib Bhattacharyya
• Date and Time: Sunday, January 17, 2010, 9:00 AM
• Venue: CSA Seminar Hall (Room No. 254, First Floor)

Abstract
For the task of near-duplicate document detection, comparison approaches based on bag-of-words used in information retrieval community are not sufficiently accurate. This work presents novel approach when instance-level constraints are given for documents and it is needed to retrieve them, given new query document for near-duplicate detection. The framework incorporates instance-level constraints and clusters documents into groups using novel clustering approach Grouped Latent Dirichlet Allocation(gLDA). Then distance metric is learned for each cluster using large margin nearest neighbor algorithm and finally ranked documents for given new unknown document using learnt distance metrics. The variety of experimental results on various datasets demonstrate that our clustering method (gLDA with side constraints) performs better than other clustering methods and the overall approach outperforms other near-duplicate detection algorithms.

Series: Department Seminar
Title: All you will never have wanted to know on branch predictors

• Speaker: Andre Seznec
                  Senior Research Director
                  INRIA, Rennes, France
• Date and Time: Friday, January 15, 2010, 4:00 PM
• Venue: CSA Seminar Hall

Abstract
Anticipating the instruction flow was already recognized as an important performance enabler in the late 50's. PC-based conditonnal branch prediction was widely publicized by J. Smith around 1980, then using branch history was proposed in 1991. Then, for about ten years, branch prediction was a hot research topic. Since 2000, interest of the computer architecture research community for branch prediction has faded. However, at the same time, there has been more progress on branch prediction accuracy between 2002 (presentation of the EV8 branch predictor) and 2006 (2nd Championship Branch Prediction) than during the 10 previous years.

In this talk, I will introduce the geometric history length predictors, O-GEHL and TAGE. I respectively presented O-GEHL at the 1st Championship on Branch Prediction (CBP-1) in december 2004 and TAGE at the 2nd Championship on Branch Prediction (CBP-2) in december 2006. Both O-GEHL and TAGE combine several prediction tables. They use a geometric series of history lengths for indexing these prediction tables. O-GEHL computes its final prediction through an adder tree, while TAGE relies on partial tag match. TAGE constitutes currently the state-of-the-art in branch prediction.

Speaker Bio:
André Seznec got a Doctorat ès Sciences in computer sciences from University of Rennes~I in June 1987. He was hired as a researcher at INRIA Rennes in October 1986. He was promoted as Research Director at INRIA in 1994 and as Senior Research Director (DR1) in 2002. He has been leading the CAPS (Compiler Parallel Architecture and Systems) project-team at INRIA Rennes since 1994. From Feb. 1999 to Feb. 2000, he spent a sabbatical year with the VSSAD, Alpha Development Group at Compaq (Shrewsbury, Massachusetts).

André Seznec has focused his research on processor architecture since the beginning of his Ph.D. thesis in 1983. He has made many contributions on vector supercomputers, pipeline architecture and SMT and multicore architecture. His most significant contributions are on cache architecture and branch prediction.

André Seznec has published more than 20 papers in international journals including IEEE transactions on computers, IEEE transaction on parallel and distributed computing, ACM Transactions on Architecture and Code Optimizations, Journal on Instruction Level Parallelism and ACM Transaction on Modeling and Computer Simulations. He has published over 40 papers in international conferences on computer architecture. André Seznec has directed 14 Ph. D. thesis.

Series: Department Seminar
Title: Intra-Disk Parallelism: A Green Storage Architecture for Data Centers

• Speaker: Dr. Sudhanva Gurumurthi
• Date and Time: Tuesday, January 12, 2010, 4:00 PM
• Venue: CSA Seminar Hall, [Room No. 254, First Floor]

Abstract
Server storage systems are often built using a large number of disks to meet the performance and capacity demands of data-intensive applications. However, such large storage systems can consume a significant amount of power, thereby increasing the power and cooling costs of a data center. In this talk, I will present a novel disk drive design, called "intra-disk parallelism", which can facilitate building high-performance, low-power enterprise storage systems. Intra-disk parallelism extends the conventional hard disk drive architecture by: (i) decoupling the way that the spindle and arm-assembly of a disk drive are used to service I/O requests, so that we can overlap disk seeks with rotational delays, and (ii) decoupling the multiplicity of the components within each of these two electro-mechanical systems to further enhance parallelism.

I will first provide a historical retrospective on intra-disk parallelism, discussing the similarities and key differences between our approach and the multi-actuator drives of the past. I will present an overview of the design space of intra-disk parallelism, identifying the locations within a disk drive where parallelism can be incorporated. Using a set of commercial workloads, I will provide an analysis of the performance and power characteristics of a specific design within this space and show that storage arrays built using such drives consume 40%-60% less power while delivering performance that is comparable to arrays built using conventional disk drives. Finally, I will discuss the key engineering and cost issues involved in building intra-disk parallel drives and show that intra-disk parallelism can be a practical approach to build energy-efficient enterprise storage systems.

Speaker Bio:
Sudhanva Gurumurthi is an Assistant Professor in the Computer Science Department at the University of Virginia. Sudhanva's research area is computer architecture, with specialization in energy-efficient storage systems, silicon reliability, and storage class memory. He received his PhD in Computer Science and Engineering from Penn State in 2005 and his BE from the College of Engineering Guindy, Anna University in 2000. He has held research positions at the IBM Austin Research Lab and Intel Corporation, and is currently a faculty research consultant for Intel. Sudhanva received the NSF CAREER Award in 2007. He is a member of the ACM and the IEEE. More details about his research are available on this homepage: http://www.cs.virginia.edu/~gurumurthi/

Series: Department Seminar
Title: Cooperative Crug Isolation

• Speaker: Mr. Aditya Thakur, Univ. of Wisconsin
• Date and Time: Friday, January 08, 2010, 4:00 PM
• Venue: CSA Seminar Hall (Room No. 254, First Floor)

Abstract
With the widespread deployment of multi-core hardware, writing concurrent programs has become inescapable. This has made fixing concurrency bugs (or crugs) critical in modern software systems. Static analysis techniques to find crugs such as data races and atomicity violations are not scalable, while dynamic approaches incur high run-time overheads. Crugs pose a greater challenge since they manifest only under specific execution interleavings that may not arise during in-house testing. Thus there is a pressing need for a low-overhead program monitoring technique that can be used post-deployment. We present Cooperative Crug Isolation (CCI), a low-overhead instrumentation technique to isolate the root causes of crugs. CCI inserts instrumentation that records occurrences of specific thread interleavings at run-time by tracking whether successive accesses to a memory location were by the same thread or by distinct threads. The overhead of this instrumentation is kept low by using a novel cross-thread random sampling strategy. We have implemented CCI on top of the Cooperative Bug Isolation framework. CCI correctly diagnoses bugs in several nontrivial concurrent applications while incurring only 2-7% run-time overhead. Joint work with Prof. Ben Liblit and Prof. Shan Lu.

Speaker Bio:
Aditya Thakur is a PhD student in the Computer Sciences Department at the University of Wisconsin-Madison. He received his master's degree from the Indian Institute of Science, Bangalore. His research interests include program analysis, verification, and concurrency bug isolation.

Series: Department Seminar
Title: The blessing and the curse of the multiplicative updates

• Speaker: Dr. Manfred K Warmuth
                  UC Santa Cruz
• Date and Time: Friday, January 08, 2010, 3:00 PM
• Venue: CSA Seminar Hall (Room No. 254)

Abstract
Multiplicative updates multiply the parameters by nonnegative factors. These updates are motivated by a Maximum Entropy Principle and they are prevalent in evolutionary processes where the parameters are for example concentrations of species and the factors are survival rates. The simplest such update is Bayes rule and we give an in vitro selection algorithm for RNA strands that implements this rule in the test tube where each RNA strand represents a different model. In one liter of the RNA soup there are approximately 10^20 different strands and therefore this is a rather high-dimensional implementation of Bayes rule. We investigate multiplicative updates for the purpose of learning online while processing a stream of examples. The ``blessing'' of these updates is that they learn very fast because the good parameters grow exponentially. However their ``curse'' is that they learn too fast and wipe out parameters too quickly. We describe a number of methods developed in the realm of online learning that ameliorate the curse of these updates. The methods make the algorithm robust against data that changes over time and prevent the currently good parameters from taking over. We also discuss how the curse is circumvented by nature. Some of nature's methods parallel the ones developed in Machine Learning, but nature also has some additional tricks. This will be a high level talk. No background in online learning or Biology will be required.

Series: Phd Colloquium
Title: Algorithms for Profiling and Representing Programs with Applications to Speculative Optimizations

• Speaker: Subhajit Roy
• Faculty Advisor: Prof. Y.N. Srikant
• Date and Time: Thursday, January 07, 2010, 4:00 PM
• Venue: CSA Seminar Hall (Room No. 254)

Abstract
Speculative optimizations have now become prime gear-wheels in modern compilation machinery. Much more aggressive than their traditional ``safe'' counterparts, they are biased towards optimizing frequently executed program paths, even with detrimental effects on the remaining paths; any such penalty is easily counterweighted by the sheer frequency of the optimized paths.

A speculative optimizer needs a good indication of the run-time behaviour of a program, represented in convinient data structures, for use by the analysis and transformation phases. Acyclic path profiles --- execution frequencies of short paths that terminate at backedges --- are the most popular control-flow indicators for today's compilers. These profiles, however, miss control-flow information across loop iterations. Many optimizations have been found to be more effective when this information is available. Also, the static program representation and the dynamic control-flow information are generally maintained in isolated data structures, making it cumbersome for speculative optimizers, which need to work with two watertight data-structures.

We propose a richer control-flow entity, k-iteration paths: longer paths spanning across at most k iterations of a loop, to capture a program's execution pattern more precisely. The k-iteration profiles are overlapping profiles --- strictly more informative than an acyclic path profile on a k-unrolled loop. Essentially via a generalization of the Ball-Larus acyclic path-profiling algorithm, we show that it is possible to number these multi-iteration paths perfectly, allowing for an efficient profiling algorithm for these longer paths as well. Experimental results show that k-iteration profiling is realistic.

We also present an effective way of amalgamating profile information with the static program structure in a novel program representation, the Hot-Path SSA (HPSSA) form, to be savoured as a single unit by speculative optimizers. The Static Single Assignment (SSA) form, which has been acknowledged as a major boon for traditional optimizations, lightens up a deep void in the speculative domain. An SSA form for the speculative domain, the HPSSA form encourages the development of novel, efficient speculative optimizations. We demonstrate via Speculative Sparse Conditional Constant Propagation (SSCP), a novel extension of Wegman & Zadeck's SCP algorithm, that, befriended by the HPSSA form, many existing SSA-based traditional optimizations can encourage a corresponding ally in the speculative orbit.

Speaker Bio:
Subhajit Roy is a Ph.D student in the CSA department. His areas of research interest are program analysis for speculative compiler optimizations and verification, dataflow analysis, and profiling algorithms.

Series: Department Seminar
Title: Simultaneous Representation Problems

• Speaker: Mr. Krishnam Raju Jampani, PhD Student, University of Waterloo, Canada
• Date and Time: Tuesday, January 05, 2010, 4:00 PM
• Venue: CSA Seminar Hall (Room No. 254, First Floor)

Abstract
We introduce the simultaneous representation problem, defined for any graph class C characterized in terms of representations, e.g. any class of intersection graphs. Two graphs G1 and G2, sharing some vertices X (and the corresponding induced edges), are said to have a simultaneous representation with respect to a graph class C, if there exist representations R1 and R2 of G1 and G2 that are â~@~consistentâ~@~] on X. Equivalently (for the classes C that we consider) there exist edges E' between G1 â~H~RX and G2 â~H~RX such that G1 â~Hª G2 â~Hª E' belongs to class C.

Simultaneous representation problems are related to graph sandwich problems, probe graph recognition problems and simultaneous planar embeddings and have applications in any situation where it is desirable to consistently represent two related graphs.

We give efficient algorithms for the simultaneous representation problem on interval, chordal, comparability and permutation graphs. These results complement the recent poly-time algorithms for recognizing probe graphs for the above classes and imply that the graph sandwich problem for these classes is solvable for an interesting special case: when the set of optional edges induce a complete bipartite graph. Moreover for comparability and permutation graphs, our results can be extended to solve a generalized version of the simultaneous representation problem when there are k graphs any two of which share a common vertex set X. This generalized version is equivalent to the graph sandwich problem when the set of optional edges induce a k-partite graph.

This is a joint work with Anna Lubiw.

Series: Department Seminar
Title: The Physics of Theoretical Computation

• Speaker: Prof. Edward Fredkin
                  Carnegie Mellon University
• Date and Time: Monday, January 04, 2010, 3:00 PM
• Venue: CSA Seminar Hall [First Floor, Room No. 254]

Abstract
What does Theoretical Physics have to say about Theoretical Computation? So far, physics has had little to say of any use. Galileo and Newton started a revolution in Physics by brilliantly choosing to study mechanics in the absence of friction and thereby finding mathematical laws that govern the dynamic behavior of moving masses. Ordinary models of computation have obscured the underlying physics of computation because of ever-present friction; at least 0.018 ev (at room temperature) for every bit of information lost in every operation of each Boolean gate. Such gates have more input states than output states. This friction (Landauer Dissipation) had seemed to be a necessary characteristic of all ordinary models of computation from Turing machines to everything that uses digital logic. Luckily, we now know that there are equally efficient theoretical models of computation that dispense with friction and that allow us to understand relationships between theoretical physics and theoretical computation. Physically correct theoretical models of frictionless computation can provide useful and enlightening constraints to microscopic discrete space, time, state models of fundamental processes in theoretical physics. Physics and Computation have a lot to say to each other!

Speaker Bio:
Edward Fredkin is a living legend of computer science. He joined the United States Airforce at the age of 19 and became a jet fighter pilot. Fredkin's computer career started in 1956 when the Air Force assigned him to work at MIT's Lincoln Laboratories. In 1968 Fredkin returned to academia, starting at MIT as a full professor. From 1971 to 1974 he was the Director of CSAIL (formerly "LCS" or "Project MAC"). He spent a year at Caltech as a Fairchild Distinguished Scholar, working with Richard Feynman, and was a Professor of Physics at Boston University for 6 years. More recently he has been a Distinguished Career Professor at Carnegie Mellon University and also a Visiting Professor at MIT.

Prof. Fredkin has been broadly interested in computation: hardware and software. He is the inventor of many things including the Trie data structue, the Fredkin Gate and the Billiard Ball Model. Fredkin and his students did pioneering work on cellular automata and reversible computing. He has also been involved in computer vision, chess and other areas of AI research. Fredkin also works at the intersection of theoretical issues in the physics of computation and computational models of physics. He recently developed Salt, a model of computation based on fundamental conservation laws from physics. (For more details, please look up: http://www.stanford.edu/class/ee380/Abstracts/050126.html)

Series: Department Seminar
Title: Beating the Graph Search Bound for the Simulation of Nondeterministic Turing Machines

• Speaker: Dr. Subrahmanyam Kalyanasundaram, USA
• Date and Time: Wednesday, December 30, 2009, 3:00 PM
• Venue: CSA Seminar Hall, Room No. 254, First Floor,

Abstract
The standard simulation of a nondeterministic Turing Machine (NTM) by a deterministic one essentially searches a large graph, which is of size exponential in the running time of the NTM. The graph is the natural one defined by the configurations of the given nondeterministic Turing Machine. In the worst case all known methods require time linear in the size of this graph, which is exponential of course.

Our main result is a new simulation method that can simulate an NTM in time roughly the square-root of the size of this graph. This result is not a general method for searching all graphs faster. Instead our theorem relates to the fact that the graph arises from an NTM. We use several properties of Turing tapes and how they interact with guessing to prove our theorem. These ideas and lemmas may be of independent interest.

This is joint work with Richard Lipton, Kenneth Regan and Farbod Shokrieh.

Series: Department Seminar
Title: Concentration of Measure for Dummies

• Speaker: Dr. Devdatt Dubhashi,
                  Dept . of Computer Science and Engg.
                  Chalmers and Gothenburg University,
                  Sweden
• Date and Time: Tuesday, December 29, 2009, 4:00 PM
• Venue: Room No. 254, CSA Seminar Hall, First Floor

Abstract
Over the past decade or so, many powerful techniques have been developed to establish concentration of measure - martingales, Talagrand's inequality, Log Sobolev inequalities etc. These are extremely useful for the analysis of randomized algorithms, but often beginners are daunted by the formidable abstract theory underlying them. However, for applications to the analysis of algorithms, it is often possible to distill these inequalities into very simple forms which are elementary and do not require knowing the underlying theory. We will give a tutorial illustration with some paradigm examples.

Series: Department Seminar
Title: Supply Chain Design for Emerging Markets

• Speaker: Prof. N. Viswanadham
                  Professor and Executive Director
                  Global Logistics and Manufacturing Strategy Center
                  Indian School of Business
                  Hyderabad
• Date and Time: Thursday, December 24, 2009, 4:00 PM
• Venue: CSA Seminar Hall, Room No. 254, First Floor

Speaker Bio:
Professor N. Viswanadham is currently at the Indian School of Business heading the GLAMS Center on Global Logistics and Manufacturing Center. He spent 37 years at the Indian Institute of Science during 1961-1998, as a B.E. student, M.E. Student, Ph D Student, Faculty Member in the Department of CSA (1971-1998), Chairman of the Department of CSA (1990-96), and Chairman of the Division of Electrical Sciences (1996-97). He is a recent recipent of the IISc Distinguished Alumni Award. His areas of current research interest include supply chain design and optimization; business analytics; rural supply chains and emerging markets.

Series: Department Seminar
Title: Where do Rewards Come From?

• Speaker: Prof. Andrew G. Barto
                  Department of Computer Science
                  University of Massachusetts - Amherst
• Date and Time: Friday, December 18, 2009, 11:00 AM
• Venue: Room No. 254, First Floor, Room No. 254

Abstract
In the computational reinforcement learning (RL) framework, the reward function determine the problem the learning agent is trying to solve. Properties of the reward function influence how easy or hard the problem is, and how well an agent may do in trying to solve it, but RL theory and algorithms are insensitive to the source of rewards (except perhaps requiring that reward magnitude be bounded). This is a great strength of the framework because of the generality it confers, but it is also a weakness because it defers key questions about the nature of reward functions. I describe a series of computational experiments recently carried out by Satinder Singh, Rick Lewis, and me that elucidate aspects of the relationship between ultimate goals (cf. reproductive success for an animal) and the primary rewards that drive learning. Among the lessons provided by these experiments are clarification of the traditional notions of extrinsically and intrinsically motivated behavior and that the precise form of an optimal reward function need not bear a transparent relationship to an agent's ultimate goal.

Speaker Bio:
Andrew Barto is Professor of Computer Science, University of Massachusetts, Amherst. He has been Chair of the UMass Department of Computer Science since 2007. He received his B.S. with distinction in mathematics from the University of Michigan in 1970, and his Ph.D. in Computer Science in 1975, also from the University of Michigan. He joined the Computer Science Department of the University of Massachusetts Amherst in 1977 as a Postdoctoral Research Associate, became an Associate Professor in 1982, and has been a Full Professor since 1991. He is Co-Director of the Autonomous Learning Laboratory and a core faculty member of the Neuroscience and Behavior Program of the University of Massachusetts. His research centers on learning in natural and artificial systems, and he has studied machine learning algorithms since 1977, contributing to the development of the computational theory and practice of reinforcement learning. His current research centers on what psychologists call intrinsically motivated behavior, meaning behavior that is done for its own sake rather than as a step toward solving a specific problem. Recent work is aimed at allowing artificial agents to construct and extend hierarchies of reusable skills that form the building blocks for open-ended learning. He currently serves as an associate editor of Neural Computation, as a member of the editorial boards of the Journal of Machine Learning Research, Adaptive Behavior, and Theoretical Computer Science-C: Natural Computing. Professor Barto is a Fellow of the American Association for the Advancement of Science, a Fellow and Senior Member of the IEEE, and a member of the American Association for Artificial Intelligence and the Society for Neuroscience. He received the 2004 IEEE Neural Network Society Pioneer Award for contributions to the field of reinforcement learning. He has published over one hundred papers or chapters in journals, books, and conference and workshop proceedings. He is co-author with Richard Sutton of the book "Reinforcement Learning: An Introduction," MIT Press 1998, and co-editor with Jennie Si, Warren Powell, and Don Wunch II of the "Handbook of Learning and Approximate Dynamic Programming," Wiley-IEEE Press, 2004.

Series: Department Seminar
Title: Gaussian Process Regression: A Tutorial and Application to Pattern Analysis

• Speaker: Dr. Sargur Srihari
                  Department of Computer Science and Engineering
                  University at Buffalo, The State University of New York, USA
• Date and Time: Friday, December 11, 2009, 10:00 AM
• Venue: Room No. 254, Seminar Hall, First Floor

Abstract
Machine learning is an area with a fifty year history which has enabled most applications of pattern recognition. Most methods of machine learning have been based on models of regression and classification that involve learning of functions defined by parameters. Such models have been extended to incorporate uncertainty in parameters by using the fully Bayesian approach. A startling recent development is the discovery of a non-parametric approach wherein a simple Gaussian distribution defined over functions is able to not only handle most problems of interest but also naturally incorporate the Bayesian approach. We will describe some regression problems such as ranking web pages in information retrieval, finding core points in fingerprints and carbon dioxide emissions (which is at the center of the global warming debate). We will proceed to describe the Gaussian process approach and its implementation, We will conclude with recent results obtained on some regression problems.

(Joint work with Chang Su)

Series: Department Seminar
Title: Approximate Shortest Descent Path on a Terrain

• Speaker: Dr. Sasanka Roy, Post Doctoral Fellow
• Date and Time: Monday, December 07, 2009, 4:00 PM
• Venue: CSA Seminar Hall, Room No. 254, First Floor

Abstract
A path from s to t on a polyhedral terrain is descending if the height of a point p never increases while we move p along the path from s to t. No efficient algorithm is known to find a shortest descending path (SDP) from s to t in a polyhedral terrain. We will discuss an approximation algorithm that solve the SDP problem on general terrains. Our algorithm is simple, robust and easy to implement. We will also discuss the pros and cons for finding the optimum path on general terrains.

Speaker Bio:
Dr. Sasanka Roy received his Ph.D. degree in Computer Science from Indian Statistical Institute, Kolkata. He is currently a Centenary Postdoctoral Fellow at Department of Computer Science and Automation, Indian Institute of Science, Bangalore. Before joining CSA, IISc Bangalore, he had worked at Tata Research Development and Design Centre, Pune, India for about two and half years as a Scientist.

Series: Department Seminar
Title: Fast and Sloppy - scaling up linear models

• Speaker: Prof. Alexander J. Smola, Yahoo
• Date and Time: Wednesday, November 18, 2009, 4:00 PM
• Venue: Room No. 254, Seminar Hall, Room No. 254

Abstract
In this talk I discuss a number of algorithms which, in combination, can be used to scale up linear models to deal with the amounts of data available at Yahoo. In particular, I will discuss issues of collaborative classification with a very large number of classes, hashing to reduce dimensionality, compressed memory representations for collaborative filtering, and algorithms to accelerate online learning on parallel computers.

Speaker Bio:
Dr. A. Smola is currently principal research scientist with Yahoo! Prior to his joining Yahoo! he was a professor at the Australian National University(ANU) and Group leader at NICTA, Australia.

Series: Department Seminar
Title: Migrating into the Cloud

• Speaker: Dr. T.S. Mohan
                  Principal Researcher, ECom Research Lab,
                  Infosys Technologies Ltd.
• Date and Time: Monday, November 16, 2009, 4:00 PM
• Venue: CSA Seminar Hall (Room No. 254)

Abstract
While Cloud Computing is hyped by Gartner to be the top of the top ten Strategic Technology Areas to be watched out for in 2010, there are big challenges for an enterprise in leveraging and using this techno-business disruptive model called cloud. In this talk we focus on key technical issues and research problems as well as solutions in using or adopting or integrating or more specifically migrating into existing Cloud Models and offerings. Cloud offerings are typically modeled at three levels - IAAS, PAAS or SAAS. We will detail what it means to migrate into each of these models as well as the issues and challenges facing the architects developing the migration strategy. We detail a seven step process of Cloud Migration that we had proposed and share the best practices associated with the development of software architecture best fitting for each of these cloud models. We conclude this talk touching upon some of key engineering and research challenges in 'making the cloud happen under the hood'.

Speaker Bio:
Dr. T S Mohan is with E&R's ECom Research Lab working as a Principal Researcher. His areas of research interests include Distributed Systems, High Performance Computing, Cloud and Grid as well as Software Architecture and Engineering. He has a varied experience of 22 years in the academia and industry. He holds a Masters and PhD in Computer Science from IISc and has worked there at SERC for about a decade before moving into the industry, working at HP ISO and interesting IT technology startups. He was an entrepreneur as well having run his own startup for more than six years before joining Infosys. Prof Rajaraman, Emeritus Professor, supervised his PhD Thesis entitled, "Interaction Paradigms in Distributed Object Oriented Programming Languages".

Series: Department Seminar
Title: PerfCenter and AutoPerf: Tools and Techniques for Modeling and Measurement of the Performance of Distributed Applications

• Speaker: Dr. Varsha Apte, Visiting Professor
• Date and Time: Wednesday, November 11, 2009, 11:30 AM
• Venue: Room No. 254, CSA Seminar Hall, First Floor

Abstract
In this talk, we will present the design and methodology underlying two software tools that we have developed in the last few years at IIT Bombay for performance measurement and modeling of distributed applications.

We present a tool, PerfCenter, which can be used for performance oriented deployment and configuration of a multi-tier application in a hosting center, or a data center. While there are a number of tools which aid in the process of performance analysis during the software development cycle, few tools are geared towards aiding a data center architect in making appropriate decisions during the deployment of an application. PerfCenter facilitates this process by allowing specification in terms that are natural to a data center architect. Thus, PerfCenter takes, as input, the number and specs of hosts available in a data center, the network architecture of geographically diverse data centers, the deployment of software on hosts, hosts on data centers, and the usage information of the application (scenarios, resource consumption), and provides various performance measures such as scenario response times, and resource utilizations. We describe the PerfCenter specification, and its performance analysis utilities, and illustrate its use in the deployment and configuration of a Webmail application. PerfCenter works by generating the underlying queueing network model of the distributed system and solving it either by analytical methods or discrete-event simulation. We will provide an insight into the primary challenges of solving this complex model analytically. Finally, we present some validation results, where PerfCenter model predictions were compared against measured data, which confirmed the soundness of the tool.

We also present a load generator and performance measurement tool (AutoPerf ) which requires minimal input and conguration from the user, and produces a comprehensive capacity analysis as well as server-side resource usage prole of a Web-based distributed system, in an automated fashion. The tool requires only the workload and deployment description of the distributed system, and automatically sets typical parameters that load generator programs need, such as maximum number of users to be emulated, number of users for each experiment, warm-up time, etc. The tool also does all the co-ordination required to generate a critical type of measure, namely, resource usage per transaction or per user for each software server. This is a necessary input for creating a performance model of a software system.

Speaker Bio:
Varsha Apte is a faculty member in the Department of Computer Science and Engineering, IIT Bombay, where she has been since 2002. During the year 2009-10 (sabbatical leave from IITB) she is Visiting Faculty at the Computer Science and Automation Dept. at IISC Bangalore and part-time Visiting Researcher in the IBM India Research Lab, Bangalore. Prior to joining IIT Bombay, she was in the Network Design and Performance Analysis Department of AT&T Labs in Middletown, NJ. She received her Ph.D. from Duke University in 1994, and Masters from Pune University in 1989. Her primary research interest is in performance management (modeling, analysis and control) of distributed applications.