This workshop focuses on the application of differential geometry to deep learning following the recent surge in interest of research in related topics such as learning on spaces of constant curvature, generative modelling and optimization methods designed for manifolds. As this is the first workshop in a nascent but burgeoning area we welcome both original research, datasets, novel visualization methods and software packages that enable deep learning practitioners to train models on manifolds.
Potential topics include but are not limited to:
Learning on non-Euclidean spaces, e.g. hyperbolic and spherical, and mixed curvature spaces
Modelling Symmetries, Invariances and Equivariances
Generative modeling on Riemannian manifolds
Stochastic optimization techniques for manifolds
Visualization methods for manifolds
Benchmark datasets and evaluation methods
Applications of Differential Geometry to Computer Vision, Natural Language Processing and Graphs
We invite extended abstracts of 4 pages (excluding references) in NeurIPS format. Submissions are to be anonymized and reviewing will follow the typical NeurIPS double blind format but should additionally change the footer to distinguish the submission from a typical NeurIPS submission ---i.e. "NeurIPS 2020 workshop on Differential Geometry meets Deep Learning " is required. Additionally, submissions may use unlimited additional pages for references or appendices, but reviewers will not be required to assess the appendices. Supplemental material, such as code or data, will not be reviewed, however we encourage authors to link to any such artifacts in the camera ready version of the submission. Please submit 1 PDF file including appendices and all other supplementary material as a zip file titled "supp.zip". Submissions that are currently under review or that have been recently accepted for publication to another conference (other than NeurIPS 2020) are permitted. All accepted papers will be presented as posters, with three contributed works, as well as three most promising novel applications and original research, being selected for an oral presentation. Accepted papers at the NeurIPS 2020 main conference or other archival venues are not allowed to be resubmitted to this workshop. Submissions on papers that are currently under review in parallel venues are permitted.
Camera Ready Instructions: Accepted papers must be 4 pages long. Authors of accepted papers are encouraged to incorporate feedback received in their reviews when updating their submission. Authors will not receive an additional page to further address reviewer concerns, however unlimited references and appendices are allowed. Please note that accepted papers must change the footer in the NeurIPS style guide to read "NeurIPS 2020 workshop on Differential Geometry meets Deep Learning". Finally, Accepted papers will be required to record a 3 minute video which will be available through the workshop website but not livestreamed. Papers accepted as Contributed talks are invited to record a 6 minute video which will be livestreamed during the workshop. Finally, all accepted papers will be further presented in a virtual poster session held on gather.town.
Please submit here by clicking the button below. The submission deadline will be October 16th, at 23h59 anywhere on earth.
Submission Date: Oct 16, 2020 (23:59 AOE)
Acceptance Notification: Nov 2, 2020
Camera Ready Deadline: Nov 14, 2020
Recording Deadline for Accepted Papers: Nov 14, 2020
Poster Deadline Dec 1, 2020
Virtual Workshop: Dec 11, 2020