NeurIPS paper: Selective Disk Bispectrum keeps all image info minus rotation
ninamiolane · x · 2026-11-21
Nina Miolane's team published a NeurIPS paper tackling how much data, compute and energy vision models waste learning that a rotated cat is still a cat, introducing a new mathematical tool: the Selective Disk Bispectrum (SDB).
- SDB is a complex-valued rotation-invariant vector that preserves all image information except orientation (completeness);
- Theoretical contributions include SDB inversion, reduced spatial/computational complexity vs. the full disk bispectrum, and expectation/variance under noise;
- A numerical approximation comes with guarantees on accuracy and rotation invariance;
- Experiments validate invariance, noise robustness and classification performance, plus reconstruction tests on multi-reference alignment of rotated images.
The work aims for a middle ground between compact hand-crafted invariant representations and learned ones. Paper: arXiv:2511.19706.
More from Research
- Riemann proof released as blog post and 8 zips, no listed author — eigenhector · 2026-11-21
- NeurIPS Paper Derives First Algorithm to Reconstruct Images from Disk Bispectrum Alone — ninamiolane · 2026-11-21
- Impetus Grants Appeals for Philanthropic Funding to Build the Biological Data AI Cures Need — shae_mcl · 2026-11-21
- Snorkel's binary-gated GRPO lifts Qwen3.8-27B from 31.9% to 40.7% on OpenAI's GDPval — ajratner · 2026-11-21
- Artificial Analysis Launches Search API Benchmark Covering 26 Products Across 14 Providers — ArtificialAnlys · 2026-11-21
- The 250,000-page Riemann proof claim: when computational verification outpaces human comprehension — SydSteyerhart · 2026-11-21