YINGLONG XIA photo

Research Areas

Related

IPDPS 2012, Shanghai, China

IBM

Group Name

ParLearning 2012


Tab navigation

ParLearning 2012

(Click Here for Chinese Version)

Workshop on Parallel and Distributed Computing for Machine Learning and Inference Problems

Shanghai, China, May 25, 2012
In Conjunction with IPDPS 2012

   
  Highlights
 
  • Foster collaboration between HPC community and AI community
  • Applying HPC techniques for learning problems
  • Identifying HPC challenges from learning and inference
  • Explore a critical emerging area with strong academia and industry interest
  • Great opportunity for researchers worldwide for collaborating with Chinese Academia and Industry
   
  Workshop Program
   
 
Session 1 (8:30am~10:00am)
 
(1) [Keynote-1] Accelerating Machine Learning on Heterogeneous Clusters: Lessons Learnt and Future Directions,
Hari Cadambi (60 min, including Q&A)
 
(2) Accelerating the Training of HTK on GPU with CUDA
Zhihui Du; Xiangyu Li; Ji Wu
 
break (10:00am~10:30am)
 
Session 2 (10:30am~12:00pm)
 
(3) Dynamic Linear Solver Selection for Transient Simulations Using Machine Learning on Distributed Systems
Paul Eller; Jing-Ru Cheng; Robert Maier
 
(4) 2D Partitioning Based Graph Search for the Graph500 Benchmark
Koji Ueno; Toyotaro Suzumura
 
(5) OLAP Aggregation Based on Dimension-oriented Storage
Jinghua Zhao
 
Lunch (12:00pm~1:30pm)
 

Session 3 (1:30pm~3:00pm)
 
(6) [Keynote-2] DeepQA technique and its application on healthcare domain,
Feng Cao (60 mins, including Q&A)
 
(7) A GPU-accelerated Approximate Algorithm for Incremental Learning of Gaussian Mixture Model
Chunlei Chen; Dejun Mu; Huixiang Zhang; Bo Hong
 
break (3:00pm~3:30pm)
 

Session 4 (3:30pm~4:30pm)
 
(8) Task Parallel Implementation of Belief Propagation in Factor Graphs
Nam Ma; Yinglong Xia; Viktor K. Prasanna
 
(9) PQH: A Multithreaded Parallel NN Search Index for Content-based Image Retrieval
Hui-zhong Chen
 
   
   
  Keynote Speech
   
  [TALK - 1]

Hari Cadambi
NEC Laboratories America, USA

"Accelerating Machine Learning on Heterogeneous Clusters: Lessons Learnt and Future Directions"

The past decade has witnessed general-purpose hardware platforms evolving from single-core processors to multi- and many-core processors like the GPU and the more recent Intel MIC. The decade has also witnessed the birth (and death) of some special-purpose hardware accelerators. At the same time, applications are evolving to impose more stringent performance constraints and big data requirements. In this talk, we will describe our experiences in parallelizing and accelerating machine learning algorithms. Starting from parallelizing SVM on a cluster of dual-core machines back in 2005, we will walk through our experiments in designing FPGA-based learning engines, accelerating image classification using GPUs, and finally parallelizing digital pathology on current high-end clusters. In each case, we will summarize what worked, what did not, what was hard, what was easy and what we learnt.

Speaker bio:
Hari Cadambi is a senior research staff member at NEC Laboratories America in Princeton, New Jersey, USA. His interests span the breadth of parallel computing systems, including programming frameworks, runtimes and hardware infrastructure. He currently works on software and middleware for heterogeneous clusters, and is particularly interested in how heterogeneity due to the use of many-cores like the GPU or the more recent Intel MIC may be used to improve performance and energy efficiency of big data applications.

 
[TALK - 2]

Feng Cao
IBM Research - China

"DeepQA technique and its application on healthcare domain"

The supercomputer Watson has beaten the human beings in the Jeopardy! game, which demonstrates the ahead of IBM in artificial intelligence. The goal is to have computers start to interact in natural human terms across a range of applications and processes, understanding the questions that humans ask and providing answers that humans can understand and justify. In fact, Watson's breakthrough computation & analytic capability could help to build a smarter planet and help people's work and life. For example, Watson could help doctors to diagnose patients' disease, help to improve the online self-service center, provide information for tourists and citizens, etc. In this talk, we will introduce the challenges we encountered to build watson, and discuss the Watson's DeepQA technique and system architecture to support high performance computing. We will also talk its potential applications in healthcare domain.

Speaker bio:
Dr. Feng Cao, Research Staff Member, is the manager of Business Intelligence & Semantic Search in IBM Research - China. He and his team focus on semantic technology and innovation applications in healthcare. Their research results have been successfully applied in China main land, Taiwan, Korea and US, selected into the final list of Wall Street Asian Best Innovation Award 2010, IBM Century Celebration Global Reference Cases 2011, and demonstrated in CeBit 2011 (the No.1 Business Technology Exhibition, Global Conferences and Networking events) in Germany, Asian HIMSS, MIE in Norway and IOD in US. He got 16 patents filed.

  Call For Papers
   
  (PDF, TXT, Chinese/中文)

Learning and inference using large scale Bayesian Networks

  • Large scale inference algorithms using parallel TPIC models, clustering and SVM etc.
  • Parallel natural language processing (NLP).
  • Semantic inference for disambiguation of content on web or social media
  • Discovering and searching for patterns in audio or video content
  • On-line analytics for streaming text and multimedia content
  • Comparison of various HPC infrastructures for learning
  • Large scale learning applications in search engine and social networks
  • Distributed machine learning tools (e.g., Mahout and IBM parallel tool)
  • Real-time solutions for learning algorithms on parallel platforms
   
  Important Date (US Eastern Time)
   
 
Workshop Paper Due December 19, 2011
January 18, 2012
Author Notification February 6, 2012
Camera-ready Paper Due February 29, 2012
(Hard deadline)
Early Registration March 27, 2012
(Register at HERE)
   
  Paper Guidelines (camera-ready only)
   
  Camera-ready manuscripts are expected to be 10 pages in the IEEE conference style, including figures, tables, and references.

Authors can purchase up to 2 additional pages for camera-ready papers after acceptance. Please find details on www.ipdps.org.

Here are the style templates:
LaTex Package (ZIP)
Word Template (ZIP)

Check the IPDPS web page for additional information (www.ipdps.org) shortly after the author notification

All papers must be submitted through the EDAS portal. Click HERE or visit http://edas.info/N11575 to submit your paper.

Please follow the instructions on IPDPS web (http://www.ipdps.org/ipdps2012/2012_author_resources.html) (see section IPDPSW AUTHORS (Workshops & PhD Forum)) for preparing and submitting your camera-ready paper. Please pay attention to the hard deadline. Authors may want to upload a version of your paper a few days before the deadline so as to check the format of your camera-ready version.

Students with accepted papers may have a chance to apply for a travel award. Please find details at www.ipdps.org.

Please find our Paper ID as follows:
01 Accelerating the Training of HTK on GPU with CUDA Zhihui Du; Xiangyu Li; Ji Wu
02 A GPU-accelerated Approximate Algorithm for Incremental Learning of Gaussian Mixture Model Chunlei Chen; Dejun Mu; Huixiang Zhang; Bo Hong
03 PQH: A Multithreaded Parallel NN Search Index for Content-based Image Retrieval Hui-zhong Chen
04 Task Parallel Implementation of Belief Propagation in Factor Graphs Nam Ma; Yinglong Xia; Viktor K. Prasanna
05 Dynamic Linear Solver Selection for Transient Simulations Using Machine Learning on Distributed Systems Paul Eller; Jing-Ru Cheng; Robert Maier
06 2D Partitioning Based Graph Search for the Graph500 Benchmark Koji Ueno; Toyotaro Suzumura
07 Clustering-Based Distributed Linear Support Vector Machine Algorithm Ye Li
08 Feature Selection Method Based on Parallel Binary Immune Quantum-Behaved Particle Swarm Optimization Chao Chen
09 OLAP Aggregation Based on Dimension-oriented Storage Jinghua Zhao
   
  Proceedings
   
  All papers accepted by the workshop will be included in the proceedings of the IEEE International Symposium on Parallel & Distributed Processing, Workshops and PhD Forum (IPDPSW), indexed in EI and possibly in SCI.
   
  Registration and Hotel Info
   
  At least one author must register for the workshop via IPDPS registration page (http://www.ipdps.org/) and present the paper at the workshop. The registration fees are available HERE.

Attendees should book hotel rooms themselves. The conference hotel is the Regal Shanghai East Asia Hotel, Set in the Xu Jia Hui commercial district, the Regal Shanghai East Asia Hotel is 15 minutes from the Heng Shan Road entertainment area, 25 minutes from downtown, and convenient to Shanghai's industrial zones. The hotel offers a full service business center, sophisticated meeting and conference facilities, Premier and Regal Club Floors, five restaurants and bars, and a state-of-the-art health club and indoor heated swimming pool.

The Special Run-of-the-house Rates for IPDPS 2012 are available HERE. These special guaranteed rates are available from May 16th through May 30th. This rate cannot be guaranteed for reservations made after April 20, 2012.

Please find the 3 ways to make reservations at http://www.ipdps.org/ipdps2012/2012_hotel.html

Please feel free to book other hotels up on your convenience.
   
  Organization
   
  General Co-chairs:
Sutanay Choudhury, Pacific Northwest National Laboratory, USA
George Chin, Pacific Northwest National Laboratory, USA
Yinglong Xia, IBM T.J. Watson Research Center, USA

Local Chair:
Yihua Huang, Nanjing University, China

Program Co-chairs:
John Feo, Pacific Northwest National Laboratory, USA
Chandrika Kamath, Lawrence Livermore National Laboratory, USA
Anshul Gupta, IBM T.J. Watson Research Center, USA

Program Committee:
Arindam Banerjee, University of Minnesota, USA
Weizhu Chen, Microsoft Research, China
Jatin Chhugani, Intel Corp., USA
Edmond Chow, Georgia Tech, USA
Tina Eliassi-Rad, Rutgers University, USA
Mahantesh Halappanavar, Pacific Northwest National Lab, USA
Lawrence B. Holder, Washington State U., USA
Yihua Huang, Nanjing University, China
Yan Liu, University of Southern California, USA
Arindam Pal, Indian Institute of Technology, India
Yangqiu Song, Microsoft Research, China
Oreste Villa, Pacific Northwest National Lab, USA
Jun Wang, IBM T.J. Watson Research Center, USA
Yi Wang, Tencent Holdings Lt., China
Haixun Wang, Microsoft Research, China
Lexing Xie, Australian National University, Australia
   
   
  Local Information
   
  Shanghai, with a population of more than 23 million (with over 9 million migrants), is one of the largest and most developed city in China. Shanghai was the largest and most prosperous city in the Far East already during the 1930's, and has remained the most developed city in mainland China. In the past 20 years Shanghai has again become an attractive city for tourists from all over the world. The world once again had its eyes on the city when it hosted the 2010 World Expo, recording the greatest number of visitors in the event's history...(click here)

Travel information of Shanghai can be find in here.
   
  Contact
   
  Should you have any questions regarding the workshop or this webpage, please contact yxia ~AT~ us DOT ibm DOT com, or sutanay DOT choudhury ~AT~ pnnl DOT gov.