ARXIV:2605.03410 · AI PRODUCT · SUBMITTED 06 MAY · 20:22 UTC · FRESHNESS STALE

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: unverified proof status

Geometry over Density: Few-Shot Cross-Domain OOD Detection

Shawn Li · You Qin · Jiate Li · Charith Peris · Lisa Bauer · Roger Zimmermann · +1 at arXiv

A universal out-of-distribution detection tool using few-shot learning for diverse domains.

Ship in 2-4 weeks›Score8.0Evidence unverified

Opportunity summary

Pain A universal out-of-distribution detection tool using few-shot learning for diverse domains.

Evidence 0 refs | 4 sources | 67% coverage

Blocker Evidence unverified

Open Build Read PDF Signal Canvas Track

PROBLEM

A universal out-of-distribution detection tool using few-shot learning for diverse domains. Standard OOD detectors are trained on a specific in-distribution (ID) dataset and detect deviations from that single domain.

METHOD

Full abstract

Out-of-distribution (OOD) detection identifies test samples that fall outside a model's training distribution, a capability critical for safe deployment in high-stakes applications. Standard OOD detectors are trained on a specific in-distribution (ID) dataset and detect deviations from that single domain. In contrast, we study few-shot cross-domain OOD detection: given a \emph{single} pre-trained model, can we perform OOD detection on \emph{arbitrary} new ID-OOD task pairs using only a handful of ID samples at inference time, with no additional training? We propose \textbf{UFCOD}, a unified framework that achieves this goal through information-geometric analysis of diffusion trajectories. Our key insight is that diffusion noise predictions are score functions (gradients of log-density), and we extract two energy features: \emph{Path Energy} (integrated score magnitude) and \emph{Dynamics Energy} (score smoothness), that form a discrete Sobolev norm capturing how samples interact with the learned diffusion process. The central contribution is a \textbf{train-once, deploy-anywhere} paradigm: a diffusion model trained on a single dataset (e.g., CelebA) serves as a universal feature extractor for OOD detection across semantically unrelated domains (e.g., CIFAR-10, SVHN, Textures). At deployment, each new task requires only $\sim$100 unlabeled ID samples for inference: no retraining, no fine-tuning, no task-specific adaptation. Using 100 ID samples per task, UFCOD achieves 93.7\% average AUROC across 12 cross-domain benchmarks, competitive with methods trained on 50k--163k samples, demonstrating $\sim$500$\times$ improvement in sample efficiency. See our code in https://github.com/lili0415/UFCOD.

RESULT

ScienceToStartup currently rates this 8.0/10 on the public viability pass. We propose \textbf{UFCOD}, a unified framework that achieves this goal through information-geometric analysis of diffusion trajectories. A public repository is linked, so build verification…

WHY NOW

AI product moved forward this cycle; last verified May 2026. Public score 8.0/10. Implementation evidence is present through a linked repository.

Continue into Read for claims, analysis, references, and neighboring papers.

Opportunity summary

Score8.0

PainA universal out-of-distribution detection tool using few-shot learning for diverse domains.

Evidence0 refs | 4 sources | 67% coverage

Blockerno shell-level blocker reported

Analysis summary

A universal out-of-distribution detection tool using few-shot learning for diverse domains.

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: unverified proof status

{ "contract_version": "paper-r2", "paper_id": "c1816009-89ee-4414-952c-a71c27566a49", "arxiv_id": "2605.03410", "canonical_route": "/paper/geometry-over-density-few-shot-cross-domain-ood-detection", "active_tab": "synced from current hash by the drawer client", "selected_artifact": "geometry-over-density-few-shot-cross-domain-ood-detection", "endpoints": { "paper_pack": "/api/v1/paper/geometry-over-density-few-shot-cross-domain-ood-detection/paper-pack", "build_passport": "/api/v1/paper/geometry-over-density-few-shot-cross-domain-ood-detection/build-passport", "mcp_resource": "sciencetostartup://surfaces/paper-workspace" } }

{ "surface": "paper", "mode": "paper", "query": "Geometry over Density: Few-Shot Cross-Domain OOD Detection", "normalized_query": "2605.03410", "route": "/paper/geometry-over-density-few-shot-cross-domain-ood-detection", "paper_ref": "geometry-over-density-few-shot-cross-domain-ood-detection", "topic_slug": null, "benchmark_ref": null, "dataset_ref": null }

{ "@context": "https://schema.org", "@graph": [ { "@type": "WebPage", "@id": "https://sciencetostartup.com/paper/geometry-over-density-few-shot-cross-domain-ood-detection#webpage", "url": "https://sciencetostartup.com/paper/geometry-over-density-few-shot-cross-domain-ood-detection", "name": "Geometry over Density: Few-Shot Cross-Domain OOD Detection", "description": "A universal out-of-distribution detection tool using few-shot learning for diverse domains.", "isPartOf": { "@id": "https://sciencetostartup.com/#website" } }, { "@type": "ScholarlyArticle", "@id": "https://sciencetostartup.com/paper/geometry-over-density-few-shot-cross-domain-ood-detection#scholarlyArticle", "headline": "Geometry over Density: Few-Shot Cross-Domain OOD Detection", "description": "A universal out-of-distribution detection tool using few-shot learning for diverse domains.", "url": "https://sciencetostartup.com/paper/geometry-over-density-few-shot-cross-domain-ood-detection", "sameAs": "https://arxiv.org/abs/2605.03410", "identifier": { "@type": "PropertyValue", "propertyID": "arXiv", "value": "2605.03410" }, "isAccessibleForFree": true, "isPartOf": { "@id": "https://sciencetostartup.com/#website" }, "datePublished": "2026-05-05T06:32:16.000Z", "author": [ { "@type": "Person", "name": "Shawn Li", "affiliation": { "@type": "Organization", "name": "University of Southern California" } }, { "@type": "Person", "name": "You Qin", "affiliation": { "@type": "Organization", "name": "National University of Singapore" } }, { "@type": "Person", "name": "Jiate Li", "affiliation": { "@type": "Organization", "name": "University of Southern California" } }, { "@type": "Person", "name": "Charith Peris", "affiliation": { "@type": "Organization", "name": "Amazon" } }, { "@type": "Person", "name": "Lisa Bauer", "affiliation": { "@type": "Organization", "name": "Amazon" } }, { "@type": "Person", "name": "Roger Zimmermann", "affiliation": { "@type": "Organization", "name": "National University of Singapore" } }, { "@type": "Person", "name": "Yue Zhao", "affiliation": { "@type": "Organization", "name": "University of Southern California" } } ], "codeRepository": "https://github.com/lili0415/UFCOD", "additionalProperty": [ { "@type": "PropertyValue", "propertyID": "viabilityScore", "value": 8 }, { "@type": "PropertyValue", "propertyID": "researchDomain", "value": "AI product" }, { "@type": "PropertyValue", "propertyID": "commercialReadiness", "value": "code, repo url" } ] }, { "@type": "SoftwareSourceCode", "@id": "https://sciencetostartup.com/paper/geometry-over-density-few-shot-cross-domain-ood-detection#software", "name": "Geometry over Density: Few-Shot Cross-Domain OOD Detection - Source Code", "description": "A universal out-of-distribution detection tool using few-shot learning for diverse domains.", "codeRepository": "https://github.com/lili0415/UFCOD", "url": "https://github.com/lili0415/UFCOD" }, { "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://sciencetostartup.com" }, { "@type": "ListItem", "position": 2, "name": "AI product", "item": "https://sciencetostartup.com/topics" }, { "@type": "ListItem", "position": 3, "name": "Geometry over Density: Few-Shot Cross-Domain OOD Detection", "item": "https://sciencetostartup.com/paper/geometry-over-density-few-shot-cross-domain-ood-detection" } ] }, { "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "What is the startup potential of \"Geometry over Density: Few-Shot Cross-Domain OOD Detection\"?", "acceptedAnswer": { "@type": "Answer", "text": "A universal out-of-distribution detection tool using few-shot learning for diverse domains." } }, { "@type": "Question", "name": "What products could be built from this research?", "acceptedAnswer": { "@type": "Answer", "text": "Build a standalone OOD detection API using UFCOD as the backend, offering plug-and-play safety enhancements for AI systems." } }, { "@type": "Question", "name": "What are the practical use cases?", "acceptedAnswer": { "@type": "Answer", "text": "A tool for automatically detecting and handling unexpected inputs in autonomous vehicles, ensuring safer operations across varying environments." } }, { "@type": "Question", "name": "What industries could this research disrupt?", "acceptedAnswer": { "@type": "Answer", "text": "This approach replaces extensive domain-specific training requirements for each new task, drastically simplifying OOD setup and broadening flexibility." } } ] } ] }

Geometry over Density: Few-Shot Cross-Domain OOD Detection

Geometry over Density: Few-Shot Cross-Domain OOD Detection

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Related Resources

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Related Resources

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline