Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves
Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves
Authors: Kartik Tandon, Julian Gould, Tanishq Bhatia, Francesca Dominici, Alejandro Ribeiro, Claudio Battiloro Date: 2026-05-07 Paper ID: openalex:2605.06395
Summary
This paper introduces HilbNets, a geometric deep learning framework designed to process infinite-dimensional signals supported on manifolds. By leveraging Hilbert bundles and connection Laplacians, the authors extend convolutional operators to signals where each point resides in a unique Hilbert space. They provide theoretical guarantees for discretization, showing that their sheaf-based approach converges to continuous architectures and offers cross-sampling transferability.
Key Contributions
- Introduces HilbNets, a convolutional learning framework for infinite-dimensional signals on manifolds using connection Laplacians.
- Proves that sampling a manifold with a Hilbert bundle induces a Hilbert Cellular Sheaf whose sheaf Laplacian converges to the connection Laplacian.
- Demonstrates that discretized HilbNets are consistent and transferable across different samplings of the same bundle.
Open Questions & Future Work
Key Concepts
- hilbnets: A convolutional neural network framework designed for infinite-dimensional signals supported on manifolds using Hilbert bundle connection Laplacians.
- hilbert-cellular-sheaf: A graph-based discretization framework for Hilbert bundles that preserves geometric consistency via sheaf Laplacians.
Archivist Review
I approved the two core architectural concepts (HilbNets and Hilbert Cellular Sheaf) as they establish a foundational mechanism for geometric deep learning on infinite-dimensional fiber spaces. The open question was approved for articulating the specific theoretical bottleneck of extending graph-to-manifold convergence results to infinite-dimensional Hilbert bundles. No datasets were approved as none were central to the methodology or sufficiently distinct.
Approved Concepts
- HilbNets: Represents the primary architectural innovation for processing infinite-dimensional signals via connection Laplacians.
- Hilbert Cellular Sheaf: Formally bridges continuous manifold-based signal processing with discrete graph-based implementations.
Approved Open Questions
- Geometric Learning for Infinite-Dimensional Signals: Addressing this gap is necessary to provide theoretical guarantees, such as consistency and transferability, for deep learning models operating on complex data structures like time series and measure-valued representations.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.06395
- paper_id
- 2605.06395
- paper_source
- openalex
- domain
- geometric-learning
- tags
-
- architectures
-
- datasets
-
- concept_slugs
- hilbnetshilbert-cellular-sheaf
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-10T06:05:59Z