Projects

• Parallel Reinforcement Learning Framework for Graph Optimization Problems
  • Jan. 2019 - present
  • Design and implement an open-source, parallel AI environment to address large-scale graph problems.
  • Achieve good scalability for both inference and training using up to 192 GPUs.
  • Outperform the state-of-the-art distributed multi-agent reinforcement learning approach by up to 2.6 times.
  • Two paper has been accepted by [ICCS’20] and [TOPC’23].
  • Keywords: distributed reinforcement learning, graph optimization problems, distributed-GPU computing.
• Software Analytics Toolkit
  • Jan. 2018 - Dec. 2018
  • Design and implement a software analytics toolkit for understanding open source community-based scientific code.
  • Applying graph algorithms to the software analysis.
  • Two papers accepted by [ICCS’19] and [SE-HER’19].
  • Keywords: application software analysis, static code analysis, high performance scientific application.
• High Performance Synchronization-reducing Clustering Library
  • Jan. 2016 - Dec. 2018
  • Design and implement a parallel asynchronous clustering algorithm.
  • The new algorithm use the hybrid MPI/Pthreads computing model.
  • Significantly faster than the parallel standard k-means with less communication and similar SSE costs.
  • Two papers accepted by [MLHPC’19] and [IJPHCA’20].
  • Keywords: massively parallel algorithm, synchronization-reducing algorithm, clustering algorithm, multi-thread programming.
• Scalable Universal QR Factorization Solver (suCAQR)
  • Aug. 2014 - Dec. 2015
  • Design and implement a simplified, tuning-less scalable universal QR factorization solver (suCAQR).
  • Achieve 30 times and 30% faster performance than benchmarks ScaLAPACK and DPLASMA on 1024 cores.
  • Two papers accepted by [ICPADS’16] and [PPL’18 ].
  • Keywords: scientific computing, performance analysis and optimization, dataflow runtime system.
• Classifying Images of Bacteria
  • Feb. 2015 - May 2015
  • Given 1800+ images of bacteria, classifying them into four categories.
  • Achieve over 95% classification accuracy and ranked second among 23 teams in “Unsupervised feature learning for classifying images of bacteria at the genus level” Kaggle competition.
  • Keywords: unsupervised feature learning, representation learning, k-means, classification, image processing.
• The Indiana State Parks History Tour App
  • June 2013 - July 2013
  • Developed the mobile App “Centennial State Park Tour Explored Through Visitor Smart-phone” sponsored by Indiana Department of Natural Resources.
  • iOS application named “The Indiana State Parks History Tour” is available in iTunes store.
  • Keywords: mobile App development.

back