Research map / Probabilistic and scientific modeling / Dynamical systems
Sparsity in Continuous-Depth Neural Networks
Hananeh Aliee, Till Richter, Mikhail Solonin, Ignacio Ibarra, Fabian J Theis, Niki Kilbertus
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add PathReg path-level sparsity instead of unregularized NODE | NODE · Synthetic second-order ODE | extrapolation MSE: 3.2\times 10^{-3} -> 3.9\times10^{-4} | Improved |
| Use L0 regularization instead of no sparsity | NODE · Synthetic second-order ODE | extrapolation MSE: 3.2\times 10^{-3} -> 3.3\times10^{-2} | Hurt |
| Use LassoNet instead of no sparsity | NODE · Synthetic second-order ODE | extrapolation MSE: 3.2\times 10^{-3} -> 4.0\times 10^{-2} | Hurt |
| Use GroupLasso (augmented with initial velocities) | NODE · Synthetic second-order ODE | extrapolation MSE: 3.2\times 10^{-3} -> 1.5\times 10^{-3} | Improved |
| Use PathReg instead of C-NODE | NODE · Synthetic second-order ODE | weight sparsity (%): 16.7 -> 35.2 | Improved |
| Add L0 regularization | NODE · MNIST in-distribution | accuracy (%): 97.34 -> 98.36 | Improved |
| Add PathReg | NODE · MNIST-C mean over corruptions | accuracy (%): 63.49 -> 69.32 | Improved |
| Use L0 instead of PathReg | NODE · MNIST-C mean over corruptions | accuracy (%): 69.32 -> 74.83 | Improved |
| Add LassoNet | NODE · MNIST-C mean over corruptions | accuracy (%): 63.49 -> 59.36 | Hurt |
| Add PathReg | NODE · Human motion capture | feature sparsity (%): 0.0 -> 71.33 | Improved |
| Use outer LassoNet instead of inner LassoNet | LassoNet · Human motion capture | feature sparsity (%): 7.01 -> 57.48 | Improved |
| Use outer LassoNet instead of inner LassoNet | LassoNet · Human motion capture walk test | MSE: 1.2 \times 10^{-3} -> 8.5 \times 10^{-3} | Hurt |
Closest papers by meaning
- Out-of-Domain Generalization in Dynamical Systems Reconstruction (ICML 2024 · similarity 0.87)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.87)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.86)
- Scaling Laws for Sparsely-Connected Foundation Models (ICLR 2024 · similarity 0.86)
- Sparse maximal update parameterization: A holistic approach to sparse training dynamics (NeurIPS 2024 · similarity 0.86)
- Asymmetric Scaling Laws from Sparse Features (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
