Accepted papers: see OpenReview site.

Link to ICLR virtual site: https://iclr.cc/virtual/2021/workshop/2130.

Data compression is a problem of great practical importance and a new frontier for machine learning research that combines empirical findings with fundamental theoretical insights. The goal of this workshop is to bring together researchers from deep learning, information theory, and probabilistic modeling in oder to learn from each other and to encourage exchange on fundamentally novel issues related to neural data compression.

Important update: The entire ICLR conference and all workshops have been pushed up by one day. The workshop is now scheduled for Friday, May 7, 2021.

Schedule

Time (UTC) Session Speaker(s)
10:30 am - 10:35 am Opening Remarks Organizers
10:35 am - 11:00 am Invited talk 1 Fabian Mentzer
11:00 am - 11:10 am Invited talk 1 Q & A Fabian Mentzer
11:10 am - 11:35 am Invited talk 2 Karen Ullrich
11:35 am - 11:45 am Invited talk 2 Q & A Karen Ullrich
11:45 pm - 12:05 pm Oral 1:Lossy Compression for Lossless Prediction Yann Dubois et al.
12:05 pm - 12:10 pm Spotlight 1:A coding theorem for the rate-distortion-perception function Lucas Theis & Aaron Wagner
12:10 pm - 12:15 pm Spotlight 2:Lower Bounding Rate-Distortion From Samples Yibo Yang and Stephan Mandt
12:15 pm - 12:20 pm Spotlight 3:Lossless compression with state space models using bits back coding James Townsend and Iain Murray
12:20 pm - 12:25 pm Spotlight 4:Conditional Coding for Flexible Learned Video Compression Théo Ladune et al.
12:25 pm - 12:30 pm Spotlight 5:Scale Space Flow With Autoregressive Priors Ruihan Yang et al.
12:30 pm - 1:00 pm Q & A + discussion Oral 1 & Spotlight 1-5 Oral 1 and Spotlight 1-5 speakers
1:00 pm - 1:30 pm Break  
1:30 pm - 1:55 pm Invited talk 3 Rianne van den Berg
1:55 pm - 2:05 pm Invited talk 3 Q & A Rianne van den Berg]
2:05 pm - 2:30 pm Invited talk 4 Oren Rippel
2:30 pm - 2:40 pm Invited talk 4 Q & A Oren Rippel
3:00 pm - 4:00 pm Zoom/Gather.town poster session Poster presenters
4:00 pm - 5:00 pm Panel discussion Panel
5:00 pm - 5:25 pm Invited talk 5 Jonathan Ho
5:25 pm - 5.35 pm Invited talk 5 Q & A Jonathan Ho
5:35 pm - 6:00 pm Invited talk 6 Johannes Balle
6:00 pm - 6:10 pm Invited talk 6 Q & A Johannes Balle
6:10 pm - 6:30 pm Oral 2:Improving Lossless Compression Rates via Monte Carlo Bits-Back Coding Yangjun Ruan et al.
6:30 pm - 6:35 pm Spotlight 6:Importance weighted compression Lucas Theis, Jonathan Ho
6:35 pm - 6:40 pm Spotlight 7:COIN: COmpression with Implicit Neural representations Adam Goliński, Emilien Dupont et al.
6:40 pm - 6:45 pm Spotlight 8:Graph Autoencoder for Graph Compression and Representation Learning Yunhao Ge et al.
6:45 pm - 6:50 pm Spotlight 9:Universal Rate-Distortion-Perception Representations for Lossy Compression George Zhang et al.
6:50 pm - 6:55 pm Spotlight 10:Lossy Image Compression with Normalizing Flows Leonhard Helminger et al.
6:55 pm - 7:25 pm Q & A + discussion Oral 2 & Spotlight 6-10 Oral 2 and Spotlight 6-10 speakers
7:25 pm - 7:30 pm Closing Remarks Organizers
7:30 pm - 8:30 pm Zoom/Gather.town poster session Poster presenters

Topics

The workshop solicits work related to neural data compression ranging from information-theoretical aspects to solving issues related to applications.

Submissions are invited for topics on, but not limited to:

  • Image/Video/Audio Compression with AutoEncoders, Flows, AutoRegressive models, Generative Adversarial networks, etc.
  • Neural Network Compression
  • Probabilistic Modeling and Variational Inference for Compression
  • Entropy Coding
  • Minimal Description Length Theory
  • Information Theory and Source Coding

Important Dates

  • Submission deadline: Feb 26, 2021 extended to Feb 28, 2021 (11:59pm anywhere on earth)
  • Notification date: March 26, 2021
  • New Workshop date: May 8, 2021 May 7, 2021 (please note that ICLR has been pushed up by one day)

Submission Instructions

We solicit short workshop paper submissions of up to 4 pages + unlimited references/appendices. Please format submissions in ICLR style. Submissions will be double blind: reviewers cannot see author names when conducting reviews, and authors cannot see reviewer names.

Some accepted papers will be accepted as contributed talks. All accepted papers will be given a slot in the poster presentation session and published via Openreview after the workshop.

Speakers

Fabian Mentzer
ETH Zurich
Karen Ullrich
FAIR NY
Rianne van den Berg
Google Brain
Oren Rippel
WaveOne
Jonathan Ho
Google
Johannes Ballé
Google

Panelists

Alex Alemi
Senior Research Scientist, Google
Irina Higgins
DeepMind
Philipp Krähenbühl
University of Texas at Austin
Jakub Tomczak
Vrije Universiteit Amsterdam
Leonardo Chiariglione
Co-founder and chair of MPEG; MPAI
Scott Labrozzi
VP, Core Media Technology/Engineering Fellow, Disney Streaming Services

Organizers

Stephan Mandt
University of California, Irvine
Robert Bamler
University of Tübingen, Germany
Yingzhen Li
Imperial College London, UK
Christopher Schroers
Disney Research Studios
Max Welling
University of Amsterdam; Qualcomm AI Research
Yang Yang
Qualcomm AI Research
Taco Cohen
Qualcomm AI Research

Organizers affiliations