| Speaker: | Arijit Khan |
| Title: | Novel Graph Queries in Large Networks |
| Date: | Wednesday, 21st December |
| Abstract: | In a wide array of disciplines, data can be modeled as an interconnected network of entities, where various attributes could be associated with both the entities and the relations among them. Knowledge is often hidden in the complex structure and attributes inside these networks. While querying and mining these linked datasets are essential for various applications, traditional graph queries may not be able to capture the rich semantics in these networks. With the advent of complex information networks, new graph queries are emerging, including graph pattern matching and mining, similarity search, ranking and expert finding, graph aggregation and OLAP. These queries require both the topology and content information of the network data, and hence, different from classical graph algorithms such as shortest path, reachability and minimum cut, which depend only on the structure of the network. In this tutorial, we shall give an introduction of the emerging graph queries, their indexing and resolution techniques, the current challenges and the future research directions. |
| Speaker | Title | Date | Abstract |
| V. Vijayasaradhi | Machine Learning Paradigms | Wednesday, 24th August | Aim of my talk is to introduce to the audience different machine learning paradigms and their applications in brief. I would be touching upon various learning methods which include supervised learning, unsupervised learning, semi-supervised learning, meta learning, relational learning among many others without going into depths of each of these learning methods. |
| Id-Ul-Fitr | Wednesday, 31st August | Talk postponed to next Wednesday | |
| Krishnamachar Sreenivasan | On A Framework For Analysing Cloud Computing System Performance |
Wednesday, 7th September | Cloud systems provide us a golden opportunity to develop special purpose systems that do one thing well; there is a performance gain in using these special purpose servers to process a request instead of using all things for all people servers. The question then becomes will the extra overhead created by the cloud in searching for the server more than offset the performance gain under all conditions. These issues are examined in this paper using a frame work for analyzing cloud system performance. Bandwidth and response time values can also be calculated using this framework which includes a technique based on using non-dimensional parameters first introduced by Earl Buckingham in 1914. This theory of similitude has been widely used in studying problems in fluid mechanics and heat transfer but this paper applies this technique to a performance model in the context of cloud computing systems. Cloud system workload can be categorized as delay bound, remote server bound, or middleware bound. Basis for this categorization is presented. |
| Amit Awekar | Mining Wikipedia | Wednesday, 14th September | Wikipedia is the most popular online encyclopedia with over 100,000 active users creating content in over 200 languages, and available publicly with detailed edit history. Open issues with Wikipedia range from fundamental problems such as accuracy and bias of crowd sourced content to cultural challenges such as encouraging content in Indian languages. This talk will focus on using data mining techniques to solve these problems with Wikipedia and use same insights in other domains. Current work in progress in IITG will also be discussed briefly. |
| Mid Semester Exams | Wednesday, 21st September | No Colloquium | Ashish Anand | Challenges and promises of biological networks |
Wednesday, 28th September | The advancement in high-throughput technologies has led to unprecedented amount of biological data. This in turn has changed the research paradigm in biology from reductionist to systems approach and at the same time put several challenges for computational biologists. Objective of the talk is to introduce some of the important biological networks being studied towards the system-level understanding of cells. Talk will briefly describe why it is important to study such networks and what are the underlying computational challenges. |
| Puja Holidays | Wednesday, 5th October | No Colloquium | |
| Benny George K | The Probabilistic Method | Wednesday, 12th October | The Probabilistic Method is a technique in combinatorics. It is a method used to prove the existence of structures with some given properties. In this talk I will introduce the technique and explain it via a few examples. We shall also describe how the probabilistic method used in conjunction with Lovasz Local Lemma becomes a powerful tool in combinatorics. |
| Wednesday, 19th October | Talk postponed to 2nd November | ||
| Deepavali | Wednesday, 26th October | No Colloquium | |
| Arabin Kumar Dey | Monte Carlo Markov Chain | Wednesday, 2nd November | Metropolis algorithm is among the ten algorithms that have had the greatest influence on the development and practice of science and engineering in the 20th century. MCMC techniques are often applied to solve integration and optimization problems in large dimensional spaces. These two types of problem play a fundamental role in computing science, physics, statistics, econometrics and decision analysis. The purpose of this introductory talk is twofold. First it reviews the main building blocks of standard Markov chain Monte Carlo simulation. Second, it discusses new interesting research horizons. |
| Hemangee K. Kapoor | Multiprocessor Revolution & Cache Architectures | Wednesday, 9th November | The talk will introduce the multicore revolution and discuss on a specific cache design for chip multi-processors. Moore's law has been redefined by having double the number of cores on chip every generation. Designers are forced to go multicore due to diminishing returns on Instruction Level Parallelism and issues in controlling the Heat and Power consumption. The wire delays are also adding new challenges for accessing on-chip data. This leads to having multiple cores on a chip and manage cache accesses effectively. We will discuss a particular cache management scheme, the reactive NUCA, which aims to meet the advantages of both the private and shared cache designs. R-NUCA cooperates with the operating system to support intelligent placement, migration and controlled replication of blocks. The design avoids overhead of coherence mechanism and at the same time gives large effective cache capacity. |
| Aryabartta Sahu | Shared Resource Management for Multi-Core Systems in presence of Aggressive Pre-fetchers | Wednesday, 16th November | In this talk I will discuss basic of perfecting algorithms and mechanisms to both manage the shared resources of a multi-core chip to obtain high performance and fairness, and also exploit pre-fetching. |