Open Problems

Out-of-Distribution & Anomaly Detection

Cross-Modality Robustness and Evaluation of Anomaly Detection Methods Across Tabular and Unstructured Data

Scope to testPartly addressed
Possible candidate · 2/5 runs3 papers report this33% from 2025+

Generated automatically from the limitations stated in 3 papers (ICML, UAI, CVPR), listed under Evidence. It is not a paper, and it does not come from papers submitted to CSPaper.

The problem

Anomaly and out-of-distribution (OOD) detection algorithms are developed in modality-specific silos, tested either exclusively on tabular benchmarks or on high-dimensional unstructured data relying on pre-trained vision/language foundation models. Because no unified empirical evaluation assesses these algorithms across both tabular and unstructured modalities, practitioners cannot determine whether tabular anomaly detectors fail on raw high-dimensional inputs or whether embedding-dependent detectors degrade when pre-trained feature extractors are absent. As a consequence, safety-critical domains with heterogeneous or non-standard data types lack validated guidance on algorithm selection.

Why it matters

Establishes the first systematic boundary map of where modality-specific anomaly detection techniques transfer and where they fail. Enables practitioners to select anomaly detection pipelines based on verified cross-modal robustness rather than unvalidated modality assumptions.

Ways to approach it

Prior-work checks are free with an account. Results someone already ran are shown to everyone.

  1. 1

    Cross-modality benchmark evaluation: Implement a standardized testbed running leading tabular OOD methods (e.g., tree-based, density-based, reconstructive) directly on raw and flattened image/sequence benchmarks, measuring AUROC, FPR95, and computational scaling across input dimensions.

  2. 2

    Embedding-free comparison: Evaluate feature-extractor-dependent OOD algorithms against tabular-native methods on tabular datasets equipped with synthetic or self-supervised embeddings versus raw features, measuring the sensitivity of detection performance to representation quality.

Have a different approach?

Describe how you would tackle this problem and we'll look for papers that already do it. Free; your text stays private.

Free · 3 checks per day

Why it might fail

If tabular-specific methods fail trivially on high-dimensional data purely due to dimensionality scaling issues (e.g., out-of-memory errors) without yielding interesting algorithmic insights, reducing the study to an obvious confirmation of the curse of dimensionality.

Evidence

Each paper's own statement of the limitation, verbatim.

Nearest existing work

Related open problems

Out-of-Distribution & Anomaly Detection

Barrier to removePartly addressed

Representation Adaptation and End-to-End Learning for Anomaly Detection Beyond ImageNet Backbones

Modern visual anomaly and out-of-distribution (OOD) detection methods rely almost entirely on frozen feature representations extracted from ImageNet-pretrained backbones. In specialized domains such as medical imaging, industrial inspection, or scientific imaging, ImageNet features fail to encode domain-specific normal variation, capping downstream anomaly scoring accuracy. Because current pipelines do not adapt or learn backbone representations end-to-end from in-distribution data, anomaly detection is structurally unviable wherever high-quality pre-trained foundation models do not already exist.

Possible candidate · 3/5 runs4 papers report this50% from 2025+
Generated automatically, not curated by hand. Automated prior-work checks catch about a third of existing work, so treat this problem as a lead to investigate.