Deep Generative Models and Downstream Applications Workshop- 14 December 2021

Sponsors - Workshop - Schedule- Speakers - Accepted Papers - Poster Session - About

Workshop

Click here to join the workshop talks

Click here to join Gathertown

Schedule

2:00 p.m. - 2:10 p.m. (GMT) 9:00 a.m. - 9:10 a.m. (EST)	Opening remarks
2:10 p.m. - 2:30 p.m. (GMT) 9:10 a.m. - 9:30 a.m. (EST)	Invited talk #1: Aapo Hyvärinen
2:30 p.m. - 2:50 p.m. (GMT) 9:30 a.m. - 9:50 a.m. (EST)	Invited talk #2: Finale Doshi-Velez
2:50 p.m. - 3:10 p.m. (GMT) 9:50 a.m. - 10:10 a.m. (EST)	Invited Talk #3: Rianne van den Berg
3:10 p.m. - 3:20 p.m. (GMT) 10:10 a.m. - 10:20 a.m. (EST)	Particle Dynamics for Learning EBMs
3:20 p.m. - 3:30 p.m. (GMT) 10:20 a.m. - 10:30 a.m. (EST)	VAEs meet Diffusion Models: Efficient and High-Fidelity Generation
3:30 p.m. - 3:35 p.m. (GMT) 10:30 a.m. - 10:35 a.m. (EST)	Contributed poster talk #1-2 Q&A
3:35 p.m. - 4:00 p.m. (GMT) 10:35 a.m. - 11:00 a.m. (EST)	Break #1
4:00 p.m. - 4:20 p.m. (GMT) 11:00 a.m. - 11:20 a.m. (EST)	Invited talk #4: Chris Williams
4:20 p.m. - 4:40 p.m. (GMT) 11:20 a.m. - 11:40 a.m. (EST)	Invited talk #5: Mihaela van der Shaar
4:40 p.m. - 5:00 p.m. (GMT) 11:40 a.m. - 12:00 p.m (EST)	Invited Talk #6: Luisa Zintgraf
5:00 p.m. - 5:10 p.m. (GMT) 12:00 p.m. - 12:10 p.m. (EST)	Your Dataset is a Multiset and You Should Compress it Like One
5:10 p.m. - 5:20 p.m. (GMT) 12:10 p.m. - 12:20 p.m. (EST)	Contributed poster talk #3 Q&A + Best paper awards
5:20 p.m. - 6:00 p.m. (GMT) 12:20 p.m. - 1:00 p.m. (EST)	Break #2
6:00 p.m. - 7:00 p.m. (GMT) 1:00 p.m. - 2:00 p.m. (EST)	Poster session #1- Gathertown
7:00 p.m. - 7:30 p.m. (GMT) 2:00 p.m. - 2:30 p.m. (EST)	Panel Discussion
7:30 p.m. - 7:50 p.m. (GMT) 2:30 p.m. - 2:50 p.m. (EST)	Invited Talk #7: Romain Lopez
7:50 p.m. - 8:10 p.m. (GMT) 2:50 p.m. - 3:10 p.m. (EST)	Break #3
8:10 p.m. - 8:30 p.m. (GMT) 3:10 p.m. - 3:30 p.m. (EST)	Invited talk #8: Alex Anderson
8:30 p.m. - 8:40 p.m. (GMT) 3:30 p.m. - 3:40 p.m. (EST)	AGE: Enhancing the Convergence on GANs using Alternating extra-gradient with Gradient Extrapolation
8:40 p.m. - 8:50 p.m. (GMT) 3:40 p.m. - 3:50 p.m. (EST)	Sample-Efficient Generation of Novel Photo-acid Generator Molecules using a Deep Generative Model
8:50 p.m. - 8:55 p.m. (GMT) 3:50 p.m. - 3:55 p.m. (EST)	Contributed poster talk #5-6 Q&A
8:55 p.m. - 9:15 p.m. (GMT) 3:55 p.m. - 4:15 p.m. (EST)	Invited talk #9: Zhifeng Kong
9:15 p.m. - 9:35 p.m. (GMT) 4:15 p.m. - 4:35 p.m. (EST)	Invited talk #10: Johannes Ballé
9:35 p.m. - 9:45 p.m. (GMT) 4:35 p.m. - 4:45 p.m. (EST)	Bayesian Image Reconstruction using Deep Generative Models
9:45 p.m. - 9:55 p.m. (GMT) 4:45 p.m. - 4:55 p.m. (EST)	Grapher: Multi-Stage Knowledge Graph Construction using Pretrained Language Models
9:55 p.m. - 10:00 p.m. (GMT) 4:55 p.m. - 5:00 p.m. (EST)	Contributed poster talk #7-8 Q&A
10:00 p.m. - 11:00 p.m. (GMT) 5:00 p.m. - 6:00 p.m. (EST)	Poster session #2 -Gathertown

See NeurIPS website

Invited Speakers

List of Keynote Abstracts

Johannes Ballé (Google Research)
Mihaela van der Shaar (University of Cambridge)
Alex Anderson (WaveOne)
Aapo Hyvärinen (University of Helsinki)
Romain Lopez (UC Berkeley)
Luisa Zintgraf (University of Oxford)
Chris Williams (Edinburgh University)
Rianne van den Berg (Microsoft)
Zhifeng Kong (University of California)
Finale Doshi (Harvard University)

Accepted Papers

Authors	Title	Acceptance
Igor Melnyk, Pierre Dognin, Payel Das	Grapher: Multi-Stage Knowledge Graph Construction using Pretrained Language Models	Oral
Jan Zuiderveld, Marco Federici, Erik J Bekkers	Towards Lightweight Controllable Audio Synthesis with Conditional Implicit Neural Representations	Oral
Daniel Severo, James Townsend, Ashish J Khisti, Alireza Makhzani, Karen Ullrich	Your Dataset is a Multiset and You Should Compress it Like One	Oral
Samuel C Hoffman, Vijil Chenthamarakshan, Dmitry Zubarev, Daniel P Sanders, Payel Das	Sample-Efficient Generation of Novel Photo-acid Generator Molecules using a Deep Generative Model	Oral
Kushagra Pandey, Avideep Mukherjee, Piyush Rai, Abhishek Kumar	VAEs meet Diffusion Models: Efficient and High-Fidelity Generation	Oral
Huan He, Shifan Zhao, Yuanzhe Xi, Joyce Ho	AGE: Enhancing the Convergence on GANs using Alternating extra-gradient with Gradient Extrapolation	Oral
Razvan Marinescu, Daniel Moyer, Polina Golland	Bayesian Image Reconstruction using Deep Generative Models	Oral
Kirill Neklyudov, Priyank Jaini, Max Welling	Particle Dynamics for Learning EBMs	Oral
Junwen Bai, Shufeng Kong, Carla P Gomes	Gaussian Mixture Variational Autoencoder with Contrastive Learning for Multi-Label Classification	Poster
Donghun LEE, Ingook Jang, SEONGHYUN KIM, Chanwon Park, Junhee Park	Stochastic Video Prediction with Perceptual Loss	Poster
Luis Armando Pérez Rey, Dmitri Jarnikov, Mike Holenderski	Content-Based Image Retrieval from Weakly-Supervised Disentangled Representations	Poster
Jose Ignacio Delgado-Centeno, Paula Harder, Ben Moseley, Valentin Bickel, Siddha Ganju, Miguel Olivares-Mendez, Alfredo Kalaitzis	Single Image Super-Resolution with Uncertainty Estimation for Lunar Satellite Images	Poster
V Manushree, Sameer Saxena, Parna Chowdhury, Manisimha Varma Manthena, Harsh Rathod, Ankita Ghosh, Sahil Khose	XCI-Sketch: Extraction of Color Information from Images for Generation of Colored Outlines and Sketches	Poster
Ekansh Verma, Souradip Chakraborty	Uncertainty-aware Labelled Augmentations for High Dimensional Latent Space Bayesian Optimization	Poster
Ramon Winterhalder, Marco Bellagente, Benjamin Nachman	Latent Space Refinement for Deep Generative Models	Poster
Sheikh Shams Azam, Taejin Kim, Seyyedali Hosseinalipour, Carlee Joe-Wong, Saurabh Bagchi, Christopher Brinton	A Generalized and Distributable Generative Model for Private Representation Learning	Poster
Quang H. Le, KAMAL YOUCEF-TOUMI, Dzmitry Tsetserukou, Ali Jahanian	Instance Semantic Segmentation Benefits from Generative Adversarial Networks	Poster
Fouad OUBARI, Antoine De mathelin, Rodrigue Décatoire, Mathilde MOUGEOT	A Binded VAE for Inorganic Material Generation	Poster
Ho-Sang Chan, Siu-Hei Cheung, Victoria Ashley Villar, Shirley Ho	Searching for the Weirdest Stars: A Convolutional Autoencoder-Based Pipeline For Detecting Anomalous Periodic Variable Stars	Poster
Andrea Karlova, Wim Dehaen, Andrei Penciu	How to Reward Your Drug Agent?	Poster
Yordan Yordanov, Vid Kocijan, Thomas Lukasiewicz, Oana-Maria Camburu	Few-Shot Out-of-Domain Transfer of Natural Language Explanations	Poster
Alban Petit, Caio Filippo Corro	Preventing posterior collapse in variational autoencoders for text generation via decoder regularization	Poster
Jongmin Yu, Hyeontaek Oh, Minkyung Kim, Junsik Kim	Normality-Calibrated Autoencoder for Unsupervised Anomaly Detection on Data Contamination	Poster
Pierre Wüthrich, Jun Jin Choong, Shinya Yuki	An Interpretability-augmented Genetic Expert for Deep Molecular Optimization	Poster
Fares Meghdouri, Thomas Schmied, Thomas Gärtner, Tanja Zseby	Controllable Network Data Balancing With GANs	Poster
Tal Daniel, Thanard Kurutach, Aviv Tamar	Deep Variational Semi-Supervised Novelty Detection	Poster
Anthony L. Caterini, Gabriel Loaiza-Ganem	Entropic Issues in Likelihood-Based OOD Detection	Poster
Cristian Ignacio Challu, Peihong Jiang, Ying Nian Wu, Laurent Callot	Deep Generative model with Hierarchical Latent Factors for Timeseries Anomaly Detection	Poster
Sarah Lewis, Tatiana Matejovicova, Yingzhen Li, Angus Lamb, Yordan Zaykov, Miltiadis Allamanis, Cheng Zhang	Accurate Imputation and Efficient Data Acquisitionwith Transformer-based VAEs	Poster
Jonathan Ho, Tim Salimans	Classifier-Free Diffusion Guidance	Poster
Si-An Chen, Chun-Liang Li, Hsuan-Tien Lin	Improving Model Compatibility of Generative Adversarial Networks by Boundary Calibration	Poster
Ali Nihat Uzunalioglu, Tameem Adel, Jakub Mikolaj Tomczak	Semi-supervised Multiple Instance Learning using Variational Auto-Encoders	Poster
Tobias Weber, Michael Ingrisch, Bernd Bischl, David Rügamer	Towards modelling hazard factors in unstructured data spaces using gradient-based latent interpolation	Poster
Yuanqi Du, Xiaojie Guo, Hengning Cao, Yanfang Ye, Liang Zhao	Learning Disentangled Representation for Spatiotemporal Graph Generation	Poster
Roland S. Zimmermann, Lukas Schott, Yang Song, Benjamin Adric Dunn, David A. Klindt	Score-Based Generative Classifiers	Poster
Kin Gutierrez Olivares, Nganba Meetei, Ruijun Ma, Rohan Reddy, Mengfei Cao	Probabilistic Hierarchical Forecasting with Deep Poisson Mixtures	Poster
Ben Barrett, Alexander Camuto, Matthew Willetts, Tom Rainforth	Certifiably Robust Variational Autoencoders	Poster
Naoya Takeishi, Alexandros Kalousis	Variational Autoencoder with Differentiable Physics Engine for Human Gait Analysis and Synthesis	Poster
Chitwan Saharia, William Chan, Huiwen Chang, Chris A. Lee, Jonathan Ho, Tim Salimans, David J. Fleet, Mohammad Norouzi	Palette: Image-to-Image Diffusion Models	Poster
Jiyoung Lee, Wonjae Kim, Daehoon Gwak, Edward Choi	Conditional Generation of Periodic Signals with Fourier-Based Decoder	Poster
Jiaxin Zhang, Kyle Saleeby, Thomas Feldhausen, Sirui Bi, Alex Plotkowski, David Womble	Self-Supervised Anomaly Detection via Neural Autoregressive Flows with Active Learning	Poster
Howard Zhong, Guha Balakrishnan, Richard Strong Bowen, Ramin Zabih, William T. Freeman	Finding Maximally Informative Patches in Images	Poster
Gautham Narayan Narasimhan, Kai Zhang, Ben Eisner, Xingyu Lin, David Held	Transparent Liquid Segmentation for Robotic Pouring	Poster

Poster Session

Join the Postersession via [Gathertown (via NeurIPS SSO login)

About

Deep generative models (DGMs) have become an important branch of deep learning which now includes methods such as variational autoencoders, generative adversarial networks, normalizing flows, energy based models and autoregressive models. Many of these methods have been shown to achieve state-of-the-art results in the generation of synthetic data such as text, speech, images, music, molecules, etc. However, besides just generating synthetic data, DGMs are of relevance in practical downstream applications. A few examples are

Imputation and acquisition of missing data
Anomaly detection
Data denoising
Compressed sensing
Image compression
Image super-resolution
Molecule optimization
Interpretation of machine learning methods
One-shot generation of low-energy molecular structures
Computation of free-energy differences between molecular states
Estimating intervention effects from observational data

Despite such important application areas, advances in DGMs are typically quantified using general metrics such as test log-likelihood, conditional data reconstruction error, inception scores or visual inspection of the generated samples. These metrics are useful for justifying gains with respect to baselines when no specific application is in mind, but they may also be poor indicators of performance in practical downstream applications.

At present, there is a gap between researchers working on new DGM-based methods and researchers applying such methods to practical problems (like the ones mentioned above). We aim to fill in this gap by bringing the two aforementioned communities together. We will connect basic researchers working in the area of DGMs with relevant application areas where their methods could have significant practical impact. We will also connect practitioners with the most recent methodological advances in deep generative modeling.

In a highly interactive format, we will outline the current frontiers of practical applications and methodological contributions in DGMs. We aim to use this workshop as an opportunity to establish common language across diverse communities, to actively discuss new research problems, and to collect relevant benchmark tasks by which novel data modeling methods can be benchmarked. The program is a collection of invited talks, alongside contributed posters. A panel discussion will provide different perspectives and experiences of influential researchers and also engage in open participant conversation. An expected outcome of this workshop is the interdisciplinary exchange of ideas and initiation of collaborations.

NeurIPS 2021 and Registration

The Deep Generative Models and Downstream Applications Workshop is part of the 35th Annual Conference on Neural Information Processing Systems. Originally planned to be in Vancouver, NeurIPS 2021 and this workshop will take place entirely virtually (online). Please use the main conference website to register for the workshop.

Call for Papers

The 2021 NeurIPS Workshop on Deep Generative Models and Downstream Applications is calling for contributions in the area of deep generative modeling, with a view on using these methods to solve real-world problems with practical impact. We invite researchers to submit methodological contributions in either variational autoencoders, flows, autoregressive models, energy based models or GANs, as well as applications of these methods to specific real-world problems. Potential applications include but are not limited to the following areas:

imputation and acquisition of missing data
anomaly detection
data denoising
compressed sensing
data compression
image super-resolution
molecule optimization
interpretation of machine learning methods
identifying causal structures in data
generation of molecular structures, etc.

We invite submissions that either address new problems and provide insights or present progress on established problems. The workshop includes a poster session, which will be held online, giving the opportunity to present novel ideas and ongoing projects.

Submission Instructions

We expect most submissions to be around 4 pages in length. If your submission will be longer than 4 pages, there is no need to move material to an appendix as long as the full submission is within 10 pages, not counting references. Submissions will be accepted as contributed talks or poster presentations. Extended abstracts should be submitted by Sep 17, 2021. Papers must be submitted through the Open Review submission system. Final versions will be posted on the workshop website (and are archival but do not constitute a proceedings). Authors should use the following style files for the workshop:

-style files

Work that is presented at the main NeurIPS conference, or accepted for publication somewhere else, will not be accepted for presentation at the workshop.

Please submit at Open Review

Important Dates

Submission Date for Workshop Contributions: Sep 17, 2021 23:59 Anywhere on earth
Submission Date for Workshop Contributions: Oct 3, 2021 23:59 Anywhere on earth
Author notification: Oct 22, 2021
Workshop: Dec 14, 2021

Organizing Committee and Contact

For questions, please contact ellis-admin@eng.cam.ac.uk

Cheng Zhang
Yingzhen Li
José Miguel Hernández Lobato
Weiwei Pan
Yichuan Zhang
Austin Tripp
Oren Rippel

Deep Generative Models and Downstream Applications Workshop- 14 December 2021