This page includes links to some previous projects. Some links may no longer be active, and the content may not be up-to-date. If you are looking for papers published before 2019, see here.

Quick links to codes and datasets

  1. Switchboard Speech Sentiment (ICASSP'20)
  2. Memex QA (CVPR'18)
  3. Tumblr GIF (CVPR'16)
  4. Video2GIF (CVPR'16)
  5. GPU-FV (ICMR'16)
  6. CUDA-based Matrix Factorization (HDPC'16)
  7. LADF for Person Verification: Exp 1, Exp 2 (CVPR'13)
  8. Delta SimRank (BigDataMining'12, Best Paper Award)
  9. Gender from Body (ACM MM'08)

Paper grouped by projects

Long-Form Speech Recognition and SSL [SLT'21: RNN-T model long form errors ], [ICASSP'21: Non-Streaming Model Distillation On Unsupervised Data], [Interspeech'21: Targeted Universal Adversarial Perturbations], [Interspeech'21: Bridging the gap between streaming and non-streaming ASR], [IEEE JSTSP'22: BigSSL], [Input Length Matters]
Learning from Noisy Labels

[CVPR 2019: Automatic Adaptation of Object Detectors] [ICCV 2017: Learning from Noisy Labels with Distillaton]
Photo Classification, Search and Understanding

[TPAMI 2019: Memex QA] (Junwei's CVPR'20 talk) [TMM 2017: Fashion Images] [TMM 2017: Real Estate Images] [WSDM 2017] [CVPR 2013] [CVPR 2011] [ACMMM 2008] [CVPR 2008] [ICCV 2007]
GIFs and Short Videos

[CVPR 2016: TGIF] [CVPR 2016: Video2GIF]
Video Understanding and Action Detection

[IBMR&D 2015] [ACMMM 2012b] [ACMMM 2012a] [ECCV 2012] [CVPR 2010] [ICCV 2009]
Social Media Mining

[ACMMM 2016] [TMM 2015] [SDM 2011] [WWW 2011] [NeuroComputing 2011] [ACMMM 2010 ]
Understanding Line Drawings

[ICCV 2005] [ECCV 2006] [ACMMM 2006] [TPAMI 2008a] [TPAMI 2008b]


Build products with AI

Product: Google Cloud's Speech APIs (Tech lead of Speech Quality)
Paper: [SLT'21], [ICASSP'21a], [INTERSPEECH'21a]
Product: Speech-to-Text On-Prem (Tech lead and manager)
Press Reports: Forbes, TechTarget, ZDNet.
Product: HelloVera AI for Question Answering (Co-founder and CTO)
Press Reports: [TechCrunch], [PodCast]
Product: Yahoo Mobile Fashion Search at Taiwan (Project lead)
Press Reports: [1], [2] [3] [4] (in Chinese)
Product: making IBM Multimeida Retrieval System (IMARS) 100x times faster (main IC)
Papers: [IBM R&D 2015] [ECCV 2012] [NIPS 2011] [ACMMM 2012] [PIEEE 2012] [CVPR 2011]
Product: Smart Supermarket Surveillance System at Tokyo (IC)

Patents

US Patents 8,639,042    9,014,420    9,165,217    9,251,434    9,471,851    9,659,258    9,659,560    9,672,814    9,734,166    9,911,223    9,947,073    11,265,271