AutoDL workshop @NeurIPS 2019

Vancouver, Canada - 14 Dec 2019


The year of 2019 has seen the success of several competitions organized in the AutoDL challenge series, which provides a reusable benchmark in the domain of AutoML and Deep Learning. These competitions pose challenging questions of AutoML on many (if not all) domains that Deep Learning has been known to be successful in: Computer Vision, Natural Language Processing, Speech Recognition, etc.

Surprisingly, the AutoML problem being challenging, the winner solutions of each competition generally made great progress compared to baseline methods, given the specific domain (e.g. image, video, speech or text). In this workshop, we'll let the winners of each challenge share their winning approach. The challenges and the winners are:

  • AutoCV - 1st: kakaobrain 2nd: DKKimHCLee 3rd: base_1
  • AutoCV2 - 1st: kakaobrain 2nd: tanglang 3rd: kvr
  • AutoNLP - 1st: DeepBlueAI 2nd: upwind_flys 3rd: txta
  • AutoSpeech - 1st: PASA_NJU 2nd: DeepWisdom 3rd: Kon
  • AutoWSL - 1st: DeepWisdom 2nd: Meta_Learners 3rd: lhg1992

(the teams in bold will be present in this workshop)

One highlight of this workshop is that we will kick off the most challenging and the last competition in this series:

AutoDL challenge

(to be launched on 14 Dec 2019)

In this final challenge of the series, all domains will be combined! And participants will still be asked to provide one single algorithm that works well on all these domains.


AutoML, Deep Learning

Computer Vision, Natural Language Processing, Speech Recognition, Meta-learning, Any-time Learning

Introduction video


Date: Sat, 14 Dec 2019

  • 12:00 Welcome and announcement on TPAMI special issue Isabelle Guyon (ChaLearn)
  • 12:05 General presentation of AutoDL challenge series. Launch of final AutoDL challenge Zhengying Liu (Inria / U. Paris-Saclay) [slides]
  • 12:20 Winner approach of AutoCV/AutoCV2 Ildoo Kim, from team kakaobrain, (Kakao Brain) [slides]
  • 12:30 Winner approach of AutoNLP (video projection) Zhipeng Luo, from team DeepBlueAI, (DeepBlue Technology) [slides]
  • 12:40 Winner approach of AutoSpeech Noriaki Ota, from team Kon, (NS Solutions Corporation) [slides]
  • 12:50 Winner approach of AutoWSL Jie Hu, from team DeepWisdom, ( [slides]
  • 13:00 Closing remarks, organization of next challenges (Lifelong meta-learning, AutoGraph, etc) Wei-Wei Tu (4paradigm) [slides]


Room 116 + 117, West Building, Vancouver Convention Center


Workshop chairs:

  • Zhengying Liu (U. Paris-Saclay; UPSud, France)
  • Wei-Wei Tu (4paradigm, China)

Main organizers of AutoDL challenges:

  • Olivier Bousquet (Google, Switzerland)
  • AndrĂ© Elisseef (Google, Switzerland)
  • Isabelle Guyon (U. Paris-Saclay; UPSud/INRIA, France and ChaLearn, USA)
  • Zhengying Liu (U. Paris-Saclay; UPSud, France)
  • Wei-Wei Tu (4paradigm, China)

Other contributors to the organization, starting kit, and datasets, can be found on the Home page.