The workshop will be hosted at The International Conference on Learning Representations (ICLR 2021). ICLR is the premier gathering of professionals dedicated to the advancement of representation learning, which is generally referred to as deep learning. It will be held virtually from May 4th through May 8th, with the workshop taking place on the final day.
We invite submissions of papers on all topics pertinent to deep representation learning in the context of adverse and limited data sets. The workshop will explore challenges and solutions to overcome limited and adverse data. This includes limited and sparse datasets, noisy data, imbalance classes, heterogenous, non-stationary data, and other related topics. Whilst learning from complex and limited datasets is scientifically and technically important, it is clear that models trained and deployed under such settings can have significant unintended implications when deployed into the real-world. The workshop, therefore, also invites participants to consider the social and ethical implications of AI applied in this context. To view the livestream of the workshop and find a more detailed schedule, go to our ICLR Workshop site ICLR S2D-OLAD.
Questions of interest include (but are not limited to):
- What are the challenges and risks associated with deep representation learning from limited and adverse data?
- How do the challenges and required solutions overlap and diverge in deep and shallow representation learning? Can old insights be repurposed for the deep world?
- What are the most pertinent questions related to deep representation learning from data with adverse properties? Questions to consider are: is it possible to generalize few-shot learning across domains? What are the relative advantages of few-shot learning over fine-tuned transfer learning? What are the impacts of, and solutions to, deep representation learning from long-tailed data and data with imbalanced class priors? Moreover, we welcome, and in fact, encourage other questions.
- What are the moral and social issues related to the applications of models trained on limited and adverse data? Can these be mitigated with new technical solutions?
Workshop Logistics
Please checkout our logistics page (here) for deals on how the day will evolve
Important links
- Link to participate in workshop (requires ICLR registration): ICLR S2D-OLAD
- Link to Gather Town poster session (requires ICLR registration): Poster Session
- Link to Gather Town breakout session (requires ICLR registration): Breakout Session
- Link to S2D-OLAD Easy Poll polling site: Have your say
Key dates
- Submission date: 7 March 2021 (AOE) - Deadline Extended!
- Final decisions: 26 March 2021 (AOE)
- Workshop: 7 May 2021 - Revised Workshop Date!
Invited Speakers
Nitesh Chawla
Frank M. Freimann Professor of Computer Science & Engineering and Director of Lucy Family Institute for Data and Society at the University of Notre Dame
SMOTE: From Shallow to Deep
Alex Hanna
Senior Research Scientist on the Ethical AI team at Google
Beyond Bias: Algorithmic Unfairness, Infrastructure, and Genealogies of Data
Bharat Hariharan
Assistant Professor of Computer Science at Cornell University
Learning to see from fewer labels
Hugo Larochelle
Google Brain team in Montreal, Adjunct Professor at Université de Montréal and a Canada CIFAR Chair
Few-Shot Classification by Recycling Deep Learning
This talk will present recent work I’ve been involved in on few-shot learning. For this workshop, I’ll be framing this work around the point of view of “recycling deep learning”. Though not really a new paradigm, in that it overlaps with several others that already exist, I will try to convince you that approaching few-shot classification from that perspective can shift in interesting ways our thinking about what properties we want few-shot learning solutions to satisfy.
Schedule
Time (PST) | Time (EST) | Event |
---|---|---|
05:00 AM | 08:00 AM | Welcome from the Organisers |
05:10 AM | 08:10 AM | Invited Talk — Hugo Larochelle |
06:10 AM | 09:10 AM | Spotlights Session 1 — Voice2Series: Reprogramming Acoustic Models for Time Series Classification |
06:14 AM | 09:14 AM | Spotlights Session 1 — Density Approximation in Deep Generative Models with Kernel Transfer Operators |
06:18 AM | 09:18 AM | Spotlights Session 1 — Adversarial Data Augmentation Improves Unsupervised Machine Learning |
06:22 AM | 09:22 AM | Spotlights Session 1 — On Adversarial Robustness: A Neural Architecture Search perspective |
06:26 AM | 09:26 AM | Spotlights Session 1 — Submodular Mutual Information for Targeted Data Subset Selection |
06:30 AM | 09:30 AM | Spotlights Session 1 — Data-Efficient Training of Autoencoders for Mildly Non-Linear Problems |
06:34 AM | 09:34 AM | Spotlights Session 1 — Min-Entropy Sampling Might Lead to Better Generalization in Deep Text Classification |
06:40 AM | 09:40 AM | Coffee Break + Gather Town Virtual Poster Session 1 |
07:25 AM | 10:25 AM | Invited Talk — Nitesh Chawla |
08:25 AM | 11:25 AM | Breakout Discussion Session |
09:00 AM | 12:00 PM | Lunch Break and Gather Town Discussion Sessions |
10:30 AM | 01:30 PM | Invited Talk — Bharath Hariharan |
11:30 AM | 02:30 PM | Spotlights Session 2 — Leveraging Unlabelled Data through Semi-supervised Learning to Improve the Performance of a Marine Mammal Classification System |
11:36 AM | 02:36 PM | Spotlights Session 2 — Continuous Weight Balancing |
11:40 AM | 02:40 PM | Spotlights Session 2 — Deep Kernels with Probabilistic Embeddings for Small-Data Learning |
11:44 AM | 02:44 PM | Spotlights Session 2 — Boosting Classification Accuracy of Fertile Sperm Cell Images leveraging cDCGAN |
11:48 AM | 02:48 PM | Spotlights Session 2 — Towards Robustness to Label Noise in Text Classification via Noise Modeling |
11:52 AM | 02:52 PM | Spotlights Session 2 — DeepSMOTE: Deep Learning for Imbalanced Data |
12:00 PM | 03:00 PM | Coffee Break + Gather Town Virtual Poster Session 2 |
12:45 PM | 03:45 PM | Invited Talk &mdash Alex Hanna |
13:45 PM | 04:45 PM | Round Table Panel Discussion |
14:45 PM | 05:45 PM | Concluding Remarks by the Organisers |
Accepted papers
- Voice2Series: Reprogramming Acoustic Models for Time Series Classification. [paper, poster] Chao-Han Huck Yang; Yun-Yun Tsai; Pin-Yu Chen.
- Leveraging Unlabelled Data through Semi-supervised Learning to Improve the Performance of a Marine Mammal Classification System. [paper, poster] Mark DJ Thomas; Stan Matwin; Bruce Martin.
- Density Approximation in Deep Generative Models with Kernel Transfer Operators. [paper, poster] Zhichun Huang; Rudrasis Chakraborty; Vikas Singh.
- Continuous Weight Balancing. [paper, poster] Daniel Jeffrey Wu; Avoy Datta.
- Adversarial Data Augmentation Improves Unsupervised Machine Learning. [paper, supplementary, poster] Chia-Yi Hsu; Pin-Yu Chen; Songtao Lu; Sijia Liu; Chia-Mu Yu.
- Deep Kernels with Probabilistic Embeddings for Small-Data Learning. [paper, supplementary, poster] Ankur Mallick; Chaitanya Dwivedi; Bhavya Kailkhura; Gauri Joshi; T. Yong-Jin Han.
- Boosting Classification Accuracy of Fertile Sperm Cell Images leveraging cDCGAN. [paper, poster] Dipam Paul; Alankrita Tewari; Jiwoong Jeong; Imon Banerjee.
- Towards Robustness to Label Noise in Text Classification via Noise Modeling. [paper, poster] Siddhant Garg; Goutham Ramakrishnan; Varun Thumbe.
- On Adversarial Robustness: A Neural Architecture Search perspective. [paper, supplementary, poster] Chaitanya Devaguptapu; Gaurav Mittal; Devansh Agarwal; Vineeth N Balasubramanian.
- Submodular Mutual Information for Targeted Data Subset Selection. [paper, poster] Suraj Kothawade; Vishal Kaushal; Ganesh Ramakrishnan; Jeff Bilmes; Rishabh Iyer.
- Data-Efficient Training of Autoencoders for Mildly Non-Linear Problems. [paper, poster] Muhammad Al-Digeil; Yuri Grinberg; Mohsen Kamandar Dezfouli; Daniele Melati; Jens Schmid; Pavel Cheben; Siegfried Janz; Dan-Xia Xu.
- DeepSMOTE: Deep Learning for Imbalanced Data. [paper, poster] Damien Dablain; Bartosz Krawczyk; Nitesh Chawla.
- Min-Entropy Sampling Might Lead to Better Generalization in Deep Active Text Classification. [paper, poster] Nimrah Shakeel.
Author and style instructions
Please format your papers using the standard ICLR 2021 style files. The page limit is 4 pages (excluding references). Papers should be submitted via https://cmt3.research.microsoft.com/S2DOLAD2021.
In addition to papers describing clear research advances, we encourage the submission of short papers that discuss work in progress, new challenges and limitations, and future directions for representations learning to overcome limited and adverse data, along with socially relevant problems, ethical AI and AI safety.
Selection Criteria
All submissions will undergo peer review by the workshop’s program committee. Accepted papers will be chosen based on technical merit, empirical validation, novelty, and suitability to the workshop’s goals.
The workshop aims to provide an engaging platform for dialog that will push the state-of-the-art in representation learning from limited and adverse data. To this end, selected papers will be works in progress and propose novel topics and future directions. Work that has already appeared or is scheduled to appear in a journal, workshop, or conference (including ICLR 2021) must be significantly extended to be eligible for workshop submission. Work that is currently under review at another venue may be submitted.
Presentation details
All accepted abstracts will be presented in the form of a virtual poster. A small number of submissions will be invited to present 15-minute virtual talks. Accepted papers will be made available on the workshop website. However, this does not constitute an archival publication and no formal workshop proceedings will be made available, meaning contributors are free to publish their work in archival journals or conferences.
Code of conduct
All participants of the workshop must abide by the ICLR code of conduct. We empower and encourage you to report any behavior that makes you or others feel uncomfortable by emailing the ICLR 2021 Program Chairs. You can also contact the organizing committee by email.
Workshop organizers
- Colin Bellinger (National Research Council of Canada; colin.bellinger@nrc-cnrc.gc.ca)
- Roberto Corizzo (American University; rcorizzo@american.edu)
- Vincent Dumoulin (Google Research, Brain Team; vdumoulin@google.com)
- Nathalie Japkowicz (American University; japkowic@american.edu)