AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Unsupervised Co-Learning on $G$-Manifolds Across Irreducible Representations

Yifeng Fan, Tingran Gao, Zhizhen Jane Zhao

NeurIPS 2019 · Label-efficient and robust learning › Recognition and long-tail · approach: Contrastive representation theory · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use multi-irrep power-spectrum affinity instead of scalar edge weightsSpectral clustering with SO(2) · K=2, p=0.16Rand index: 0.569 ± 0.069 -> 0.670 ± 0.065Improved
Use multi-irrep optimal-alignment affinity instead of VDMSpectral clustering with SO(2) · K=2, p=0.16Rand index: 0.526 ± 0.036 -> 0.687 ± 0.011Improved
Use multi-irrep bispectrum affinity instead of scalar edge weightsSpectral clustering with SO(2) · K=10, p=0.20Rand index: 0.948 ± 0.015 -> 0.997 ± 0.003Improved
Use multi-irrep affinity rather than single-representation VDMSpectral clustering with SO(3) · K=2, p=0.16Rand index: 0.600 ± 0.048 -> 0.921 ± 0.038Improved
Use multi-irrep bispectrum affinity rather than VDMSpectral clustering with SO(3) · K=10, p=0.20Rand index: 0.919 ± 0.013 -> 0.943 ± 0.009Improved
Use multi-irrep optimal-alignment affinity rather than VDMS² nearest-neighbor identification · p=0.10percentage of identified neighbor pairs with inner product > 0.95: 27.56 -> 87.72Improved
Increase cutoff m_k from 2 to 20 (ablation)S² nearest-neighbor identification with Opt affinity · p=0.10percentage of identified neighbor pairs with inner product > 0.95: 15.03 -> 87.72Improved
Increase cutoff m_k from 20 to 100 (ablation)S² nearest-neighbor identification with Opt affinity · p=0.10percentage of identified neighbor pairs with inner product > 0.95: 87.72 -> 82.66Hurt
Increase maximum frequency k_max from 2 to 50 (ablation)S² nearest-neighbor identification with Opt affinity · p=0.10percentage of identified neighbor pairs with inner product > 0.95: 43.42 -> 92.07Improved
Increase maximum frequency k_max from 2 to 100 (ablation)Spectral clustering with SO(2), K=10 · p=0.16Rand index: 0.920 ± 0.019 -> 0.985± 0.003Improved
Increase maximum frequency k_max from 2 to 100 (ablation)Spectral clustering with SO(2), K=10 · p=0.25Rand index: 0.997 ± 0.001 -> 1 ± 0No clear effect
Increase cutoff m_k beyond the cluster count (ablation)Spectral clustering with SO(2), K=10 · p=0.16, 0.20, 0.25Rand indexMixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy