Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs | Pavel's Blog
izmailovpavel.github.io · 2,585 words · saved by 1 readers
Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs Figure 1: visualizat
Loss Surfaces, Mode Connectivity, and Fast Ensembling of DNNs | Pavel's Blog Skip to main content Visualizing Mode Connectivity Blogpost by Pavel Izmailov, Timur Garipov and Andrew Gordon Wilson; visualizations in collaboration Javier Ideami. [ Code , Paper ] Figure 1 : visualization of mode connectivity for ResNet-20 with no skip connections on CIFAR-10 dataset. The visualization is created in collaboration with Javier Ideami ( https://losslandscape.com/ ). Understanding generalization in deep neural networks is a great open question. Neural networks are trained by minimizing loss surfaces th
related reading
- Neural Networks, Manifolds, and Topology -- colah's blogcolah.github.io
- Modular Manifolds - Thinking Machines Labthinkingmachines.ai
- A Gentle Introduction to Graph Neural Networksdistill.pub
- Feature Visualizationdistill.pub
- Thoughts on Loss Landscapes and why Deep Learning works — LessWronglesswrong.com
- Understanding deep learning requires rethinking generalizationarxiv.org
- Thoughts on loss landscapes and why deep learning worksberen.io
- 1412.0233arxiv.org
- arxiv.org/pdf/1805.08522arxiv.org
- Understanding the Neural Tangent Kernel – EigenTaleseigentales.com
- Loss Landscape | A.I deep learning explorations of morphology & dynamicslosslandscape.com
- Understanding Convolutions on Graphsdistill.pub