Research map / Architectures and efficiency / RNNs and normalization
Hierarchical nucleation in deep neural networks
Diego Doimo, Aldo Glielmo, Alessio Ansuini, Alessandro Laio
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 |
|---|---|---|---|
| compare ImageNet with MNIST, modMNIST and CIFAR-10 for the sharpness of the chi(l,gt) transition | ResNet152 · ILSVRC2012 vs MNIST/CIFAR-10 | chi(l,gt) transition sharpness | Mixed |
| increase neighborhood size k to 300 (ablation) | chi(l,gt) and chi(l,out) profiles · ResNet152, ILSVRC2012 subset | transition detectability: k=300 | Hurt |
| vary dataset size N with constant class ratio (ablation) | chi(l,gt) and chi(l,out) profiles · ResNet152, ILSVRC2012 subset | chi profile | No clear effect |
| decrease Gaussian CKA kernel width sigma (ablation) | Gaussian CKA with output layer · ResNet152, 5000 ILSVRC2012 images | transition sharpness | Improved |
| use linear CKA instead of neighborhood overlap (ablation) | similarity measure · ResNet152, 5000 ILSVRC2012 images | detection of abrupt transition | Hurt |
| change architecture across ResNet and VGG depths | chi(l,gt) profile · ILSVRC2012 subset | chi(l,gt) trend | No clear effect |
| fewer classes (10) in modMNIST and CIFAR-10 versus 300 in ImageNet | ResNet152 · modMNIST, CIFAR-10 | lag phase of chi(l,gt) | Improved |
Closest papers by meaning
- U-Nets as Belief Propagation: Efficient Classification, Denoising, and Diffusion in Generative Hierarchical Models (ICLR 2025 · similarity 0.84)
- How Deep Networks Learn Sparse and Hierarchical Data: the Sparse Random Hierarchy Model (ICML 2024 · similarity 0.84)
- How Compositional Generalization and Creativity Improve as Diffusion Models are Trained (ICML 2025 · similarity 0.83)
- Progressively Knowledge Distillation via Re-parameterizing Diffusion Reverse Process (AAAI 2024 · similarity 0.83)
- How Diffusion Models Learn to Factorize and Compose (NeurIPS 2024 · similarity 0.83)
- Generalization in diffusion models arises from geometry-adaptive harmonic representations (ICLR 2024 · similarity 0.83)
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.
