ICLR 2021

ICLR 2021
Name	ICLR 2021
Genre	Academic conference
Date	May 2021
Location	Virtual
Organizer	International Conference on Learning Representations
Previous	2020
Next	2022

ICLR 2021 ICLR 2021 was the eighth annual meeting of the International Conference on Learning Representations, held as a virtual event in May 2021 that gathered researchers from across the machine learning community. The program combined peer-reviewed papers, keynote presentations, workshops, tutorials, and competitions, attracting participants affiliated with institutions such as Massachusetts Institute of Technology, Stanford University, University of California, Berkeley, Google Research, and DeepMind. The conference reflected ongoing research trajectories influenced by prior venues like NeurIPS and ICML and intersected with applied communities associated with OpenAI and Facebook AI Research.

Overview

The conference program emphasized advances in representation learning, optimization, and large-scale models, extending themes from earlier gatherings such as Yann LeCun's foundational work and developments at labs like Google DeepMind and OpenAI. Participation included authors and attendees from universities including Carnegie Mellon University, University of Toronto, University of Oxford, ETH Zurich, and industry groups including Microsoft Research, Amazon Web Services, and Apple Inc.. ICLR 2021 showcased contributions that linked to foundational methods from researchers associated with Geoffrey Hinton's lineage, contemporary projects at Anthropic and efforts by teams at NVIDIA and IBM Research.

Conference Organization and Format

Organizers adapted the event to an online format similar to other virtual editions like the online NeurIPS 2020 experiment, employing platforms used previously by conferences hosted by Association for Computing Machinery-affiliated meetings. The program committee comprised senior researchers from institutions including Princeton University, Columbia University, University of Washington, Tsinghua University, and Peking University, with submission and review workflows managed using systems comparable to those used by OpenReview and committees that had overseen programs at AAAI and ACL. Sessions were scheduled across time zones to accommodate contributors from regions represented by European Organization for Nuclear Research, Google Europe, and research groups in Japan and Australia.

Keynote Speakers and Invited Talks

Keynote and invited presentations drew from luminaries associated with organizations such as Yoshua Bengio's collaborators, researchers from DeepMind, engineers from OpenAI, and academics from Harvard University and University of Cambridge. Invited talks echoed topics addressed in major symposia like Royal Society lectures and topical panels at SIGKDD and ICASSP, featuring discussions on ethics and robustness from figures linked to Partnership on AI and on scaling laws researched by teams at Google Research and Microsoft Research Cambridge.

Notable Papers and Research Trends

Accepted papers reflected trends in transformer architectures pioneered in works connected to Google Brain and Google Research groups, diffusion models with ties to researchers working at University of Amsterdam and NYU, and representation learning influenced by earlier contributions from Yann LeCun, Geoffrey Hinton, and Yoshua Bengio. Topics included large-scale pretraining building on techniques used by BERT-related teams, self-supervised methods mirroring approaches from SimCLR and groups at Facebook AI Research, and reinforcement learning research with provenance from DeepMind and OpenAI. Other threads related to optimization connected to methods developed by scholars at ETH Zurich and Stanford University, and interpretability work resonated with initiatives associated with Alan Turing Institute and Montreal Institute for Learning Algorithms.

Workshops, Tutorials, and Competitions

The conference hosted workshops and tutorials organized by researchers from institutions such as Massachusetts Institute of Technology, University of California, Berkeley, University College London, Imperial College London, and companies including Google DeepMind and Facebook AI Research. Competitions and benchmarks referenced evaluation practices seen in events organized by ImageNet teams and challenges run by groups at Kaggle and OpenAI Gym contributors, while workshops addressed subfields linked to communities around Computer Vision and Pattern Recognition and Natural Language Processing conferences like ACL.

Proceedings and Publication Statistics

Proceedings were made available through online repositories and submission platforms similar to those used by OpenReview and mirrored in archives frequented by authors from arXiv and institutional repositories at MIT CSAIL and Stanford AI Lab. Acceptance rates and submission counts continued trends observed at peer venues such as ICML and NeurIPS, with authors representing a diverse set of institutions including Seoul National University, Tsinghua University, University of Michigan, Cornell University, and industrial research labs like IBM Research and Amazon Science. The program contributed to citation and dissemination pathways shared with journals and special issues overseen by editorial boards at venues like Journal of Machine Learning Research and conferences coordinated by the Association for Computational Linguistics.

Category:Academic conferences