Dates : 18th - 20th December 2013
The Mysore Park Workshop series is a forum to promote research discussion in Computer Science and Engineering in India, through workshops on contemporary topics modeled after the Dagstuhl Workshop series. These workshops are envisioned to invite leading researchers from various related disciplines to promote cross-fertilization of ideas, and inter-disciplinary research. There is particular emphasis on the overall quality of discussion. To keep quality high, participation is limited to experts from academia and industry. More information about previous workshops is available here.
This workshop will focus specifically on the area of distributed machine learning and optimization. Click here for the theme of workshop.
The workshop is on invitation basis only.
The workshop will be held at the Infosys Mysore campus, located approximately 150 kms from Bangalore. Accommodation will be provided on the campus.
The closest international airport to Mysore is Bangalore/Bengaluru (BLR), which has direct connections to several European ports (Paris, Frankfurt, London). Travel by road between Bangalore and Mysore takes approximately 3 hours. We will arrange transportation, via a bus service, from Bangalore to Mysore and vice versa.
Accomodation and Food
During the workshop, accommodation and food will be provided at no cost to the invited participants at the beautiful campus of Infosys at Mysore. Note: Infosys is an alcohol-free campus.
Infosys Campus, Mysore
The Rs. 260 crore corporate training centre has been set up by Infosys, which adds 12,000 employees every year. The US $ 60 million training centre is housed in a 270 acre campus at Mysore, the city known for its academic and research institutions and heritage buildings.
The facilities created at this centre include 24 hours food court, sports facilities (including a swimming pool), multiplex theatre, and education research block, beside the trainee hostel. It is spread over 16,00,000 sq ft of area.
Infosys Mysore Development Center reflects its commitment to be a responsible consumer of energy, water and other resources. Construction and renovation projects are driven by the long-term goals of carbon neutrality and water sustainability.
New buildings at the center have been designed to disperse maximum daylight. It minimizes energy consumption and reduces heat load from artificial lighting. Insulation techniques used in the buildings and green procurement practices reduce the environmental impact of operations.
The ambiance of this place makes it a perfect for learning and sharing of knowledge.
Most international visitors require a visa to enter India. You should contact the Indian embassy in your country to obtain more details. Please note that "Business Visa" is the most appropriate visa category for your visit. The "Conference Visa" category is not applicable as it is meant only for conferences organized by a set of Government institutes.
The past decade has witnessed an explosion of information and data. This has happened due to a convergence of trends -- growth of the web, expansion of connectivity via mobile devices and social networks, and availability of new big data sources in scientific domains such as Computational Biology, Neuroscience and High-energy physics. These trends have two important aspects. Firstly, the data can no longer be stored on a single disk or machine but instead resides on a shared memory/disk distributed over an entire cluster of machines. Secondly, the scale of this data is leading to completely new and challenging machine learning problems across a variety of domains. New classes of distributed machine learning algorithms have started emerging that not only need to take care of computational efficiency but also communication cost and correctness of parallel execution over the cluster/network. This field is attracting both academia and industry due to novel modeling and algorithm design aspects as immediate relevance to practical big data problems.
In this workshop, we aim to focus specifically on the area of distributed machine learning and optimization. We plan to have a 3 day workshop with each day focusing on the below topics:
a) Programming Frameworks: Distribution computation has several critical system aspects like communication: when (synchronous vs. asynchronous), how (master-slave vs. de-centralized), bandwidth, fault tolerance etc. as well as tricks like hashing, caching for speed-up. As a result, different frameworks like Map-reduce, MPI, Dryad, Graph lab etc. have been proposed that try to optimize a subset of above aspects while potentially compromising on others. The key challenges include understanding any underlying unified framework, evolution of programming frameworks, optimally choosing the framework for various machine learning algorithms and big data applications.
b) Machine Learning: There has been lot of work going on in parallelizing common supervised (SVMs, Neural Networks, Decision Trees, Deep Belief Nets, Graphical models etc.) as well as un-supervised algorithms (K-means clustering, Latent Dirichlet Allocation, Graphical models etc.) . As a result several novel generic templates for distributed machine learning have come up, that trade-off different aspects like time, accuracy, generality, algorithmic complexity and so on.
c) Optimization: There are several challenges from optimization point of view once data is distributed across nodes. For example: communication bandwidth becomes bottleneck for algorithms that communicate heavily, convergence guarantee to global single node solution becomes non-trivial, and so on. As a result there has been lot of focus on optimization techniques like ADMM, Dual decomposition etc. that are naturally suited for distributed optimization as well as asynchronous and de-centralized optimizations that remove communication bottlenecks while still ensuring good convergence.
d) Big Data Applications: Various web based companies are collecting huge data every day in the form of user/search logs for various applications like content recommendation, click prediction etc. Moreover, even non IT companies are hugely getting benefitted from big data analysis: measurements from various sensors can be used to build anomaly detection models saving them from huge losses due to technical shutdowns.
Accepted Speakers so far (in no particular order):
Angelia Nedich (UIUC)
Chih-Jen Lin (NTU)
SVN Vishwanathan (Purdue)
Vijay Saraswat (IBM Watson)
Francesca Rossi (Univ. of Padova, Italy))
Ganesh Ramakrishnan (IIT Bombay)
Tara Sainath (IBM Watson)
Inderjit Dhillon (UT Austin)
Peter Richtarik (Univ. of Edinburgh)
Milind Bhandarkar (Greenplum, EMC)
Sundararajan Sellamanickam (MSR India)
Saketh Nath (IIT Bombay)
Srinivasan Parthasarathy (Ohio State Univ.)
Prabhanjan Kambadur (IBM Watson)
Manik Varma (MSR India)
Shirish Shevade (IISC)
Chih-Jen Lin (National Taiwan University)
Dhruv Mahajan (Microsoft Research, India)
Vikas Sindhwani (IBM Watson) are the program chairs.