# metric learning

A computing technique for learning perceptual embeddings in which similar instances are placed close together and dissimilar instances farther apart.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## SHAMISA: SHAped Modeling of Implicit Structural Associations for Self-supervised No-Reference Image Quality Assessment

DevFeed: [SHAMISA: SHAped Modeling of Implicit Structural Associations for Self-supervised No-Reference Image Quality Assessment](<https://devfeed.tech/articles/shamisa-shaped-modeling-of-implicit-structural-associations-for-self-supervised-no-reference-image-quality-assessment-142511.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.13669>)

Author: Mahdi Naseri, Zhou Wang

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Image](<https://devfeed.tech/topics/image.md>), [Self-supervised learning](<https://devfeed.tech/topics/self-supervised-learning.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [annotations](<https://devfeed.tech/tags/annotations.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [assessment](<https://devfeed.tech/tags/assessment.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [bottleneck](<https://devfeed.tech/tags/bottleneck.md>), [comparisons](<https://devfeed.tech/tags/comparisons.md>), [construction](<https://devfeed.tech/tags/construction.md>), [content](<https://devfeed.tech/tags/content.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [control](<https://devfeed.tech/tags/control.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [diagnostics](<https://devfeed.tech/tags/diagnostics.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [encoder](<https://devfeed.tech/tags/encoder.md>), [image](<https://devfeed.tech/tags/image.md>), [modeling](<https://devfeed.tech/tags/modeling.md>), [quality](<https://devfeed.tech/tags/quality.md>), [quality-assessment](<https://devfeed.tech/tags/quality-assessment.md>), [reference](<https://devfeed.tech/tags/reference.md>)

### AI overview

SHAMISA trains a no-reference image quality encoder from unlabeled distorted images, avoiding IQA opinion-score labels and contrastive loss during pre-training. Its compositional distortion engine varies one distortion factor at a time across continuous degradation parameters, allowing soft, content-sensitive relations: images with shared distortion patterns are brought together, while severity changes create structured shifts. Dual-source relation graphs combine known degradation profiles with learned structural affinities. The encoder is then frozen and paired with a linear regressor. The abstract reports strong performance on synthetic, authentic, and cross-dataset benchmarks, improved transfer in the reported cross-dataset comparisons, and favorable qualitative robustness diagnostics; it gives no specific scores or benchmark names.

### Source excerpt

arXiv:2603.13669v3 Announce Type: replace-cross Abstract: No-Reference Image Quality Assessment (NR-IQA) aims to estimate perceptual quality without access to a pristine reference. Learning an NR-IQA model faces a fundamental bottleneck because it requires many costly human perceptual labels. We propose SHAMISA, a non-contrastive self-supervised framework that learns from unlabeled distorted images through explicitly structured relational supervision. Unlike prior methods that impose rigid binary similarity constraints, SHAMISA shapes soft, controllable implicit structural associations that are both distortion-aware and content-sensitive. A key innovation is its compositional distortion engine, which generates an uncountable family of degradations from continuous parameter spaces and groups the resulting images so that exactly one distortion factor varies at a time. This construction enables fine-grained control over representational similarity during training. Images with shared distortion patterns are pulled together in the embedding space, while severity variations produce structured shifts. Dual-source relation graphs integrate known degradation profiles with emergent structural affinities throughout pre-training. The learned encoder is then frozen and paired with a linear regressor. Extensive experiments on synthetic, authentic, and cross-dataset NR-IQA benchmarks demonstrate strong overall performance, improved transfer in the reported cross-dataset comparisons, and favorable qualitative robustness diagnostics. Self-supervised pre-training requires no IQA opinion-score annotations or contrastive loss.

## TRACE: Topological Representations for Anomaly detection through Contrastive Embeddings

DevFeed: [TRACE: Topological Representations for Anomaly detection through Contrastive Embeddings](<https://devfeed.tech/articles/trace-topological-representations-for-anomaly-detection-through-contrastive-embeddings-137716.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21548>)

Author: Sydney Johns

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [threat detection](<https://devfeed.tech/topics/threat-detection.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [graph theory](<https://devfeed.tech/topics/graph-theory.md>)

Tags: [anomaly-detection](<https://devfeed.tech/tags/anomaly-detection.md>), [attack](<https://devfeed.tech/tags/attack.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [contrastive-learning](<https://devfeed.tech/tags/contrastive-learning.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [detection](<https://devfeed.tech/tags/detection.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [framework](<https://devfeed.tech/tags/framework.md>), [graph](<https://devfeed.tech/tags/graph.md>), [graphs](<https://devfeed.tech/tags/graphs.md>), [intrusion-detection](<https://devfeed.tech/tags/intrusion-detection.md>), [malicious](<https://devfeed.tech/tags/malicious.md>), [neural-network](<https://devfeed.tech/tags/neural-network.md>), [representation-learning](<https://devfeed.tech/tags/representation-learning.md>), [topology](<https://devfeed.tech/tags/topology.md>), [trace](<https://devfeed.tech/tags/trace.md>), [validation](<https://devfeed.tech/tags/validation.md>)

### AI overview

TRACE learns representations from benign system-activity graphs by combining message passing with topological features, then scores anomalies in the contrastive embedding space. Attack labels are not used for representation learning; a labeled benign-and-malicious validation set is used to calibrate the anomaly threshold. On StreamSpot, GraSec-IoT, and Unicorn Wget, TRACE reports F1 scores of 0.923, 0.790, and 0.583. It exceeds benign-only GraphCL on StreamSpot and Unicorn Wget but trails it on GraSec-IoT. A supervised graph neural network performs best on the first two datasets, while TRACE leads on Unicorn Wget, so its advantage varies by graph setting.

### Source excerpt

arXiv:2611.21548v1 Announce Type: cross Abstract: Intrusion detectors must recognize evolving malicious behavior even when confirmed attack labels are scarce or incomplete. We present TRACE, a topology-aware graph contrastive learning framework for label-efficient intrusion detection. TRACE learns from benign system-activity graphs by combining message-passing and topological representations, then assigns anomaly scores in the same projected embedding space optimized by the contrastive objective. This design captures both local interactions and graph-scale structural deviations without using attack labels during representation learning. A labeled validation set containing benign and malicious graphs is used only to calibrate the anomaly threshold. We evaluate TRACE on StreamSpot, GraSec-IoT, and Unicorn Wget against a supervised graph neural network and the benign-only GraphCL baseline. TRACE obtains F1 scores of 0.923, 0.790, and 0.583, respectively. Compared with GraphCL, this is higher on StreamSpot (0.450) and Unicorn Wget (0.203), but lower on GraSec-IoT (0.975). The supervised model leads on StreamSpot and GraSec-IoT, whereas TRACE has the highest F1 on Unicorn Wget. The mixed results show that topology-aware benign-only learning can improve detection in some graph settings but does not provide a uniform advantage.

## Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings

DevFeed: [Provable Cross-Modal Transfer of Hierarchical Concept Structure in CLIP Embeddings](<https://devfeed.tech/articles/provable-cross-modal-transfer-of-hierarchical-concept-structure-in-clip-embeddings-135817.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17671>)

Author: Nghia Nguyen, Tianjiao Ding, Lachlan Ewen MacDonald, Ren\\'e Vidal

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [clip](<https://devfeed.tech/tags/clip.md>), [contrastive-learning](<https://devfeed.tech/tags/contrastive-learning.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [hierarchy](<https://devfeed.tech/tags/hierarchy.md>), [image](<https://devfeed.tech/tags/image.md>), [language](<https://devfeed.tech/tags/language.md>), [learning](<https://devfeed.tech/tags/learning.md>), [linear](<https://devfeed.tech/tags/linear.md>), [mechanism](<https://devfeed.tech/tags/mechanism.md>), [modal](<https://devfeed.tech/tags/modal.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [node](<https://devfeed.tech/tags/node.md>), [pairs](<https://devfeed.tech/tags/pairs.md>), [property](<https://devfeed.tech/tags/property.md>), [prototypes](<https://devfeed.tech/tags/prototypes.md>), [raw](<https://devfeed.tech/tags/raw.md>), [root](<https://devfeed.tech/tags/root.md>), [structure](<https://devfeed.tech/tags/structure.md>)

### AI overview

The paper formalizes hierarchical structure in CLIP embeddings as Hierarchical Angular Separation (HAS): a non-leaf concept should be more similar to its descendants than to concepts on other branches. It reports that HAS appears approximately in raw images and randomly initialized CLIP image embeddings, but in text embeddings only after contrastive learning. In an unconstrained-features model with fixed image embeddings and trained text embeddings, the text-side violation rate at each node is bounded by the image-side violation and near-violation rate. The proposed mechanism is hierarchical co-occurrence in image-text pairs, which draws text embeddings toward root-centered image prototypes. The authors give a closed-form account for linear contrastive loss and analogous guarantees for InfoNCE under sufficient regularization; synthetic experiments also show transfer beyond the analyzed regimes.

### Source excerpt

arXiv:2611.17671v1 Announce Type: cross Abstract: Contrastive vision-language models such as CLIP are trained to align images and text, not to represent taxonomies. Yet empirical studies show that CLIP embeddings exhibit hierarchical structure aligned with the hierarchy of semantic concepts. A theoretical characterization of why such hierarchy emerges from a non-hierarchical objective remains open. In this paper, we formalize this phenomenon as Hierarchical Angular Separation (HAS), a geometric property whereby each non-leaf concept embedding is more similar to its descendants than to off-branch concepts. Empirically, we show that HAS holds approximately in raw images and randomly initialized CLIP image embeddings, while in text embeddings it holds approximately only after contrastive learning. To theoretically explain this asymmetry, we analyze an unconstrained features model in which the image embeddings are fixed and only the text embeddings are trained. At the global minimizer, the text-side HAS violation percentage at each node is at most the image-side HAS violation and near-violation percentage. The key mechanism is the hierarchical co-occurrence structure of image-text pairs, which biases text embeddings toward root-centered image prototypes. For a linear contrastive loss, we derive a closed-form solution that makes this mechanism explicit. We extend the analysis to the InfoNCE loss and establish analogous guarantees under sufficient regularization. Synthetic experiments validate the theory and show that HAS is transferred even beyond the regimes covered by our analysis.

## WETW: Learning Pair-Specific Weighted Euclidean Distances for Efficient DTW Search

DevFeed: [WETW: Learning Pair-Specific Weighted Euclidean Distances for Efficient DTW Search](<https://devfeed.tech/articles/wetw-learning-pair-specific-weighted-euclidean-distances-for-efficient-dtw-search-135064.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17412>)

Author: Christos Panourgias, Qitong Wang, George Tzagkarakis, Themis Palpanas

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [ann](<https://devfeed.tech/topics/ann.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [also](<https://devfeed.tech/tags/also.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [best](<https://devfeed.tech/tags/best.md>), [build](<https://devfeed.tech/tags/build.md>), [building](<https://devfeed.tech/tags/building.md>), [but](<https://devfeed.tech/tags/but.md>), [collections](<https://devfeed.tech/tags/collections.md>), [compare](<https://devfeed.tech/tags/compare.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [comparisons](<https://devfeed.tech/tags/comparisons.md>), [computing](<https://devfeed.tech/tags/computing.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [learning](<https://devfeed.tech/tags/learning.md>), [pair](<https://devfeed.tech/tags/pair.md>), [search](<https://devfeed.tech/tags/search.md>)

### AI overview

WETW speeds up DTW time-series search by predicting pair-specific weights for fixed-length embeddings, without first computing an alignment. In a filter-and-verify pipeline, it uses those estimates to shortlist candidates and exact DTW to verify them, so returned distances are exact even though search may miss neighbors. Across eight datasets, WETW-FV was 11.4x faster than the fastest tested exact method at mean recall@10 of 0.93; the best learned baseline reached 0.59 recall at the same speed. For 48K-point series, it was 40x faster than the UCR Suite at recall@10 of 0.91. The proposed WESEE encoder's multi-scale features retain timing information and raised recall@1 from 0.90 to 0.95. The abstract also reports that ranking regularization raised recall@1 from 0.79 to 0.95, as fitting distances by regression alone misordered nearest neighbors.

### Source excerpt

arXiv:2611.17412v1 Announce Type: cross Abstract: Dynamic Time Warping (DTW) is the standard distance for similarity search over time series that are misaligned in time, but its cost is quadratic in the series length and lower bounds often prune few candidates, so search over large collections of long series is slow. Learned approximations embed the series to make comparisons cheap, yet compare every pair with the same fixed rule, usually the Euclidean distance, although DTW aligns each pair differently. We observe that, once a pair is aligned, DTW is a Euclidean distance over coordinates selected for that pair. Building on this view, we propose Weighted Euclidean Time Warping (WETW), a learned distance that adapts the comparison of fixed-length embeddings to each pair. A small symmetric network predicts one weight per embedding di- mension and pair without computing an alignment, which halves the error of the best static distance on the same embeddings. The embeddings come from the Weighted Euclidean SEries Encoder (WESEE), whose multi-scale Temporal Feature Extractor keeps the timing information that global pooling removes and raises recall@1 from 0.90 to 0.95, although WETW works with any encoder. We also show that fitting DTW distances is not enough for search, as regression alone misorders the closest neighbours, and that ranking regularization raises recall@1 from 0.79 to 0.95. Our filter-and-verify pipeline, WETW-FV, uses WETW to build a shortlist and exact DTW to check it, so the search is approximate but every returned distance is exact. Over eight datasets, WETW-FV is 11.4x faster than the fastest exact method among the UCR Suite, lower bounds and MESSI-DTW, at a mean recall@10 of 0.93, while the best learned baseline reaches 0.59 at the same speed. On 48K-point series, WETW- FV remains 40x faster than the UCR Suite at a recall@10 of 0.91.

## MALNISI-Infused Graph Construction and Weighted Contrastive Loss based GAT for Unsupervised Representation Learning in HSI

DevFeed: [MALNISI-Infused Graph Construction and Weighted Contrastive Loss based GAT for Unsupervised Representation Learning in HSI](<https://devfeed.tech/articles/malnisi-infused-graph-construction-and-weighted-contrastive-loss-based-gat-for-unsupervised-representation-learning-in-hsi-132320.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17199>)

Author: Soumi Sarkar, Subhadip Boral, Sarbani Palit, Ashish Ghosh

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [clustering](<https://devfeed.tech/topics/clustering.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>)

Tags: [absolute](<https://devfeed.tech/tags/absolute.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [accelerate](<https://devfeed.tech/tags/accelerate.md>), [adjacency-matrix](<https://devfeed.tech/tags/adjacency-matrix.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [article](<https://devfeed.tech/tags/article.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attention](<https://devfeed.tech/tags/attention.md>), [based](<https://devfeed.tech/tags/based.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [capability](<https://devfeed.tech/tags/capability.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [complex](<https://devfeed.tech/tags/complex.md>), [construction](<https://devfeed.tech/tags/construction.md>), [contributions](<https://devfeed.tech/tags/contributions.md>), [convergence](<https://devfeed.tech/tags/convergence.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [data](<https://devfeed.tech/tags/data.md>), [graph](<https://devfeed.tech/tags/graph.md>), [learning](<https://devfeed.tech/tags/learning.md>)

### AI overview

The method builds a weighted graph of hyperspectral image pixels using MALNISI, a neighborhood-based similarity index, then learns compact representations with an upgraded Graph Attention Network. MALNISI weights inform multi-head attention and the contrastive loss, which pulls strongly connected pixels together in latent space and pushes weakly connected ones apart. Residual connections, skip projections, and feature normalization are intended to reduce over-smoothing and preserve spectral detail. The learned embeddings are clustered with spectral clustering. The abstract reports experiments on six benchmark datasets and superior performance to fourteen state-of-the-art approaches, but gives no metrics or dataset-specific results.

### Source excerpt

arXiv:2611.17199v1 Announce Type: cross Abstract: Hyperspectral images (HSI) possess discriminating spectral characteristics with respect to land utilization due to high-dimensional information. Recently, unsupervised representation learning using graph neural networks has unveiled these utilization-specific complex spectral relations. This article has three main contributions. First, it proposes an index, the Mean Absolute Local Neighborhood Inverse Spectral Index (MALNISI), which measures the similarity between two pixels based on their neighborhood information. The index is used to construct a weighted adjacency matrix, giving the HSI a graph structure. Second, it proposes an upgraded Graph Attention Network that learns compact latent representations of hyperspectral data from the constructed graph. In this network, an enhanced multi-head attention mechanism incorporates the MALNISI weights as an auxiliary feature to emphasize learned attention between neighboring pixels. To alleviate over-smoothing and preserve fine-grained spectral details, the architecture integrates residual connections, skip projections, and feature normalization, which together enhance representation capability, stabilize learning, and accelerate convergence. Third, it proposes a weighted contrastive loss function that incorporates neighborhood-aware MALNISI between connected pixels, encouraging strongly connected pixels in the high-dimensional space to remain close in the latent space, and weakly connected pixels to stay apart. The resulting discriminative low-dimensional embeddings are then clustered using spectral clustering. Experimental analysis on six benchmark HSI datasets demonstrates the efficiency and necessity of the three aforementioned contributions, and the comparative study highlights superior performance over fourteen state-of-the-art approaches.

## Studies on Sign Language Retrieval in Low-Resource Dictionary Settings

DevFeed: [Studies on Sign Language Retrieval in Low-Resource Dictionary Settings](<https://devfeed.tech/articles/studies-on-sign-language-retrieval-in-low-resource-dictionary-settings-132261.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16621>)

Author: Ushnish Sarkar, Vineet Kumar Rakesh, Bhaswar Chattopadhyay, Tapas Samanta

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [language](<https://devfeed.tech/tags/language.md>), [resource](<https://devfeed.tech/tags/resource.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [sign](<https://devfeed.tech/tags/sign.md>)

### AI overview

A pose encoder trained on American Sign Language can retrieve signs in six other sign languages without target-language training, though performance varies by dataset. With one gallery clip per sign, top-1 accuracy ranges from 36.4% to 80.7%, compared with 16.0% to 69.6% for angle-based dynamic time warping; the six-target mean is 39.3% for an ASL-trained OpenHands graph network. Ablations find better results with per-pair InfoNCE than supervised contrastive or cross-entropy training, and with attention blocked between body and hands, which improves accuracy by up to 5.1 points. The best setup reaches 68.0% one-shot accuracy versus 59.0% for the reference encoder. On CISLR, a 4,765-word Indian Sign Language dictionary with one video per word, results are 17.3-19.0%, slightly above the RGB baseline's 16.8%. Embedding geometry generally changes results by at most 1.6 points, with the tested hyperbolic token-attention/InfoNCE variant 1.4-2.0 points worse.

### Source excerpt

arXiv:2611.16621v1 Announce Type: new Abstract: Most sign languages have no annotated corpus, only small dictionaries with one or a few example videos per sign. This study examines how well a pose encoder trained on a single sign language serves such languages without any target-language training. An encoder trained with metric learning on American Sign Language (ASL-Citizen) is evaluated frozen on six datasets of Turkish, Russian, Indian, Argentinian, Pakistani and Sri Lankan sign languages, with one or four clips per sign enrolled and, where signer identities are available, the query's signer excluded from the gallery. With one enrolled clip, the encoder retrieves the correct sign first in 36.4-80.7 % of queries, against 16.0-69.6 % for the best learning-free method (angle-based dynamic time warping) and a six-target mean of 39.3 % for an ASL-trained OpenHands graph network. In a crossed ablation of the encoder (embedding geometry, aggregation of body and hand embeddings, training objective, and attention between articulators), two choices matter: a per-pair InfoNCE objective outperforms supervised contrastive and cross-entropy training, and blocking attention between body, left hand and right hand inside the encoder raises accuracy by up to 5.1 points. The best configuration reaches 68.0 % one-shot accuracy (59.0 % for the reference encoder). On CISLR, an Indian Sign Language dictionary of 4,765 words with one video per word, the encoders find the correct word first for 17.3-19.0 % of queries, above the 16.8 % of the original RGB baseline. Over eight seeds with the joint encoder, spherical, Euclidean and hyperbolic embedding spaces differ by at most 1.6 points, except the hyperbolic token-attention head trained with InfoNCE, which is 1.4-2.0 points worse, so the geometry is a minor factor.

## XFACTORS: Disentangled Information Bottleneck via Contrastive Supervision

DevFeed: [XFACTORS: Disentangled Information Bottleneck via Contrastive Supervision](<https://devfeed.tech/articles/xfactors-disentangled-information-bottleneck-via-contrastive-supervision-131870.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2601.21688>)

Author: Alexandre Myara, Nicolas Bourriez, Thomas Boyer, Thomas Lemercier, Ihab Bendidi, Auguste Genovesio

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [VAE](<https://devfeed.tech/topics/vae.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [components](<https://devfeed.tech/tags/components.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [factor](<https://devfeed.tech/tags/factor.md>), [kl-regularization](<https://devfeed.tech/tags/kl-regularization.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [replacement](<https://devfeed.tech/tags/replacement.md>), [supervision](<https://devfeed.tech/tags/supervision.md>), [vae](<https://devfeed.tech/tags/vae.md>)

### AI overview

XFactors is a weakly supervised VAE that separates representations into a residual subspace and factor-specific subspaces for selected target factors. For each target, an InfoNCE loss brings together latents with the same factor value and separates mismatches; KL regularization organizes the residual and aggregated factor subspaces with Gaussian structure, without additional supervision for untargeted factors. The approach avoids adversarial objectives and auxiliary classifiers. The abstract reports state-of-the-art disentanglement scores across multiple datasets using constant hyperparameters, consistent qualitative factor alignment, controlled factor swapping by latent replacement, and scaling with increased latent capacity; it also reports evaluation on CelebA. These are claims from the abstract, which provides no numerical results or detailed evaluation setup.

### Source excerpt

arXiv:2601.21688v3 Announce Type: replace-cross Abstract: Disentangled representation learning aims to map independent factors of variation to independent representation components. On one hand, purely unsupervised approaches have proven successful on fully disentangled synthetic data, but fail to recover semantic factors from real data without strong inductive biases. On the other hand, supervised approaches are unstable and hard to scale to large attribute sets because they rely on adversarial objectives or auxiliary classifiers. We introduce \textsc{XFactors}, a weakly-supervised VAE framework that disentangles and provides explicit control over a chosen set of factors. Building on the Disentangled Information Bottleneck perspective, we decompose the representation into a residual subspace $\mathcal{S}$ and factor-specific subspaces $\mathcal{T}_1,\ldots,\mathcal{T}_K$ and a residual subspace $\mathcal{S}$. Each target factor is encoded in its assigned $\mathcal{T}_i$ through contrastive supervision: an InfoNCE loss pulls together latents sharing the same factor value and pushes apart mismatched pairs. In parallel, KL regularization imposes a Gaussian structure on both $\mathcal{S}$ and the aggregated factor subspaces, organizing the geometry without additional supervision for non-targeted factors and avoiding adversarial training and classifiers. Across multiple datasets, with constant hyperparameters, \textsc{XFactors} achieves state-of-the-art disentanglement scores and yields consistent qualitative factor alignment in the corresponding subspaces, enabling controlled factor swapping via latent replacement. We further demonstrate that our method scales correctly with increasing latent capacity and evaluate it on the real-world dataset CelebA. Our code is available at \href{https://github.com/ICML26-anon/XFactors}{github.com/ICML26-anon/XFactors}.

## Dual-Granularity Learning for Regression with Continuous Noisy Labels

DevFeed: [Dual-Granularity Learning for Regression with Continuous Noisy Labels](<https://devfeed.tech/articles/dual-granularity-learning-for-regression-with-continuous-noisy-labels-125215.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10550>)

Author: Hui Guo, Boyu Wang, Grace Y. Yi

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [scikit-learn](<https://devfeed.tech/topics/scikit-learn.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [binning](<https://devfeed.tech/tags/binning.md>), [classification](<https://devfeed.tech/tags/classification.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [framework](<https://devfeed.tech/tags/framework.md>), [learning](<https://devfeed.tech/tags/learning.md>), [noise](<https://devfeed.tech/tags/noise.md>), [order](<https://devfeed.tech/tags/order.md>), [proximity](<https://devfeed.tech/tags/proximity.md>), [regression](<https://devfeed.tech/tags/regression.md>), [robust](<https://devfeed.tech/tags/robust.md>), [robustness](<https://devfeed.tech/tags/robustness.md>)

### AI overview

GRAIN addresses continuous noisy regression by combining two forms of learning: instance-specific residual refinement, which targets small to moderate label deviations, and binning-based contrastive regularization, which uses ordinal structure when large corruptions make exact labels unreliable. A histogram-based bridge lets the structural guidance anchor residual learning while residual refinement corrects deviations that structure alone cannot resolve. The abstract reports improved robustness and generalization across diverse noise conditions, without giving specific datasets or quantitative results.

### Source excerpt

arXiv:2611.10550v1 Announce Type: new Abstract: Learning with noisy labels has been extensively studied for classification, yet its regression counterpart remains relatively underexplored. Unlike classification, regression involves continuous-valued targets where label noise exhibits heterogeneous magnitudes across instances, which poses conflicting requirements on the model: correcting small-to-moderate deviations requires sensitivity to label values for instance-level refinement, whereas large deviations caused by corrupted or outlying annotations render exact label values unreliable, necessitating reduced reliance on exact labels and instead leveraging ordinal structure among samples. These requirements call for methods that capture both global structure and sample-specific deviations in a unified manner. To this end, we propose GRAIN, a dual-granularity framework for robust noisy regression. At the fine-grained level, we derive an explicit residual learning mechanism from a robust loss formulation, introducing instance-specific auxiliary variables to decouple label noise from predictive signals. At the coarse-grained level, we introduce a binning-based contrastive regularization that enforces an order-preserving feature structure aligned with label proximity, addressing noise beyond the capacity of local residual refinement. The two levels are coupled via a histogram-based bridging mechanism that enables bidirectional information flow: coarse-grained structural guidance establishes a well-ordered feature space that anchors residual learning, while fine-grained residual refinement corrects instance-level deviations that structural regularization alone cannot resolve. Experimental results show that GRAIN consistently improves robustness and generalization under diverse noise conditions.

## Evolutionary Adaptive Metric Learning: A Multi-Objective Framework for Learning Local Distance Functions in Nearest Neighbor Classification

DevFeed: [Evolutionary Adaptive Metric Learning: A Multi-Objective Framework for Learning Local Distance Functions in Nearest Neighbor Classification](<https://devfeed.tech/articles/evolutionary-adaptive-metric-learning-a-multi-objective-framework-for-learning-local-distance-functions-in-nearest-neighbor-classification-123276.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08549>)

Author: Isaac Touza

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Hyperparameter optimization](<https://devfeed.tech/topics/hyperparameter-optimization.md>), [Data science and engineering](<https://devfeed.tech/topics/data-science-and-engineering.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [algorithm](<https://devfeed.tech/tags/algorithm.md>), [anchor](<https://devfeed.tech/tags/anchor.md>), [classification](<https://devfeed.tech/tags/classification.md>), [convergence](<https://devfeed.tech/tags/convergence.md>), [correlation](<https://devfeed.tech/tags/correlation.md>), [cross-validation](<https://devfeed.tech/tags/cross-validation.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [error](<https://devfeed.tech/tags/error.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [feature](<https://devfeed.tech/tags/feature.md>), [features](<https://devfeed.tech/tags/features.md>), [field](<https://devfeed.tech/tags/field.md>), [framework](<https://devfeed.tech/tags/framework.md>), [learning](<https://devfeed.tech/tags/learning.md>), [local](<https://devfeed.tech/tags/local.md>), [metric](<https://devfeed.tech/tags/metric.md>), [metric-learning](<https://devfeed.tech/tags/metric-learning.md>), [nearest-neighbor](<https://devfeed.tech/tags/nearest-neighbor.md>)

### AI overview

EAML learns instance-dependent positive-definite distance metrics for nearest-neighbor classification. Each metric combines input-dependent diagonal scaling, interpolated between anchor points with softmax, and a global low-rank correlation component. A real-coded elitist genetic algorithm optimizes a scalar objective incorporating leave-one-out error, robustness, metric smoothness, and regularization. The paper establishes several properties, including positive definiteness and Lipschitz continuity, and studies stability, generalization, and convergence under explicit assumptions. On synthetic data with region-varying feature relevance, EAML recovers local metric changes and outperforms global metrics. Across thirteen public datasets with nested 5-fold cross-validation, it improves results on some datasets but does not significantly beat global metric baselines overall; global diagonal metrics can be preferable with nuisance features or test-time noise.

### Source excerpt

arXiv:2611.08549v1 Announce Type: new Abstract: Nearest neighbour classifiers inherit the geometry of their distance function. Fixed distances and globally learned metrics apply the same geometry throughout the input space, although relevant feature directions may vary across regions. We propose EAML (Evolutionary Adaptive Metric Learning), a framework that learns instance-dependent positive-definite metrics through a low-dimensional parameterization. Each local metric combines an input-dependent diagonal scaling, generated by softmax interpolation between anchor points, with a global low-rank correlation component. A real-coded elitist genetic algorithm optimizes the parameters using a scalar objective combining leave-one-out prediction error, robustness, metric smoothness, and regularization. We establish positive definiteness, metric equivalence, existence of an optimal solution, and Lipschitz continuity of the metric field. We further investigate prediction stability, generalization bounds, and evolutionary convergence under explicit assumptions. Experiments on synthetic data with region-dependent feature relevance show that EAML can recover local metric variations and outperform global metrics. Evaluation on thirteen public datasets using nested 5-fold cross-validation reveals a more nuanced picture: EAML improves performance on selected datasets but does not significantly outperform global metric baselines overall. Additional experiments show that global diagonal metrics can be preferable under nuisance features and test-time noise. These findings identify the conditions under which evolutionary local metric learning is beneficial and highlight its practical limitations.

## Polyp Matching Using Deep Learning

DevFeed: [Polyp Matching Using Deep Learning](<https://devfeed.tech/articles/polyp-matching-using-deep-learning-117509.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.04624>)

Author: Kristian Dalsb{\\o} Hindberg, Puneet Sharma, Benedicte Schelde-Olesen, Ulrik Deding, Jakob Frederik Fr{\\o}kj{\\ae}r Justsen, Jan-Matthias Braun, Esmaeil S. Nadimi

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Deep learning](<https://devfeed.tech/topics/deep-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [image recognition](<https://devfeed.tech/topics/image-recognition.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Image processing](<https://devfeed.tech/topics/image-processing.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [classification](<https://devfeed.tech/tags/classification.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cosine](<https://devfeed.tech/tags/cosine.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [learning](<https://devfeed.tech/tags/learning.md>), [matching](<https://devfeed.tech/tags/matching.md>)

### AI overview

The study applies contrastive learning with a Siamese neural network to identify whether two colon capsule endoscopy images show the same polyp, aiming to reduce duplicate counts. It trains ResNet18/34/50, DenseNet121, and EfficientNet B0/B3 backbones on 5,103 clinician-annotated same-polyp pairs and 5,103 different-polyp pairs from the Danish CareForColon2015 trial, using cosine vector distance for classification. Reported accuracy reaches 92%, with AUC values close to 0.98. The authors suggest automated matching could help reduce over-counting and unnecessary follow-up colonoscopies, but the supplied abstract does not establish clinical impact.

### Source excerpt

arXiv:2611.04624v1 Announce Type: cross Abstract: A key challenge in colon capsule endoscopy (CCE) analysis is accurately identifying whether multiple images depict the same polyp, which is critical to avoid over-counting and unnecessary clinical interventions. Additionally, this prevents prolonged follow-up colonoscopies where endoscopists search for duplicate polyps without precise localization. This study presents a novel contrastive learning-based approach to polyp matching using a Siamese neural network architecture. The model was trained on a dataset from the Danish CareForColon2015 trial. The dataset comprises 5,103 image pairs of confirmed same polyp and 5,103 pairs of different polyps, annotated by experienced clinicians. The proposed model employs ResNet variants (ResNet18, ResNet34, ResNet50), DenseNet121 and EfficientNet B0/B3 as backbones and utilizes contrastive learning with cosine vector distance metric for classification. Experimental results demonstrate high classification accuracies of up to 92% and AUC values close to 0.98. These results suggest that automated polyp matching has the potential to reduce polyp over-counting, improve patient risk stratification, and reduce unnecessary follow-up colonoscopies.

## Syntactic Segmentation and Collage Generation for Videogame Levels with Perceptual Distance

DevFeed: [Syntactic Segmentation and Collage Generation for Videogame Levels with Perceptual Distance](<https://devfeed.tech/articles/syntactic-segmentation-and-collage-generation-for-videogame-levels-with-perceptual-distance-108138.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.38987>)

Author: Ross Mawhorter, Matthew Guzdial

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [clustering](<https://devfeed.tech/topics/clustering.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [algorithm](<https://devfeed.tech/tags/algorithm.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [clustering](<https://devfeed.tech/tags/clustering.md>), [compare](<https://devfeed.tech/tags/compare.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [design](<https://devfeed.tech/tags/design.md>), [existing](<https://devfeed.tech/tags/existing.md>), [explore](<https://devfeed.tech/tags/explore.md>), [generation](<https://devfeed.tech/tags/generation.md>), [graph](<https://devfeed.tech/tags/graph.md>), [metroid](<https://devfeed.tech/tags/metroid.md>), [procedural](<https://devfeed.tech/tags/procedural.md>)

### AI overview

The paper proposes automatically segmenting existing videogame levels and deriving chunks for procedural generation using perceptual distance, a measure based on Space Syntax analysis. It applies hierarchical clustering to that distance, compares the approach with prior StarCraft analysis, reports results on rooms from Super Metroid, and demonstrates a proof-of-concept level generator based on the segmentation. The supplied abstract does not specify comparative results or limitations.

### Source excerpt

arXiv:2610.38987v1 Announce Type: new Abstract: In procedural videogame level design, many existing techniques combine hand-authored level chunks. Some existing analysis techniques also rely on a higher-level graph structure which abstracts the level design. In this paper, we explore techniques for computing both of these things automatically from an existing design. We introduce the concept of perceptual distance, based on analysis techniques from the architecture subfield of Space Syntax. We use this perceptual distance to segment level designs with hierarchical clustering. We compare with past work on StarCraft analysis, show results on rooms from Super Metroid, and also demonstrate a proof-of-concept level generation algorithm based on this segmentation method.

## FDSML: Formula-Driven Supervised Metric Learning with Parameter-Aware Triplet Loss

DevFeed: [FDSML: Formula-Driven Supervised Metric Learning with Parameter-Aware Triplet Loss](<https://devfeed.tech/articles/fdsml-formula-driven-supervised-metric-learning-with-parameter-aware-triplet-loss-108191.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.39423>)

Author: Hina Otake, Shinei Arakawa, Keitaro Tanaka, Yoshihiro Fukuhara, Hirokatsu Kataoka, Shigeo Morishima

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Training AI Models](<https://devfeed.tech/topics/training-ai-models.md>), [image recognition](<https://devfeed.tech/topics/image-recognition.md>)

Tags: [classification](<https://devfeed.tech/tags/classification.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ne](<https://devfeed.tech/tags/cs-ne.md>), [image-generation](<https://devfeed.tech/tags/image-generation.md>), [metric-learning](<https://devfeed.tech/tags/metric-learning.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [supervision](<https://devfeed.tech/tags/supervision.md>)

### AI overview

FDSML uses relationships between image-generation parameters to provide synthetic vision pre-training with more supervision than discretized class labels. It pairs the FineFractalDB dataset, designed for higher within-class variation and more stable structures, with a Parameter-Aware Triplet Loss that selects positive and negative examples by distance in Iterated Function System parameter space. In reported ResNet-50 experiments, FDSML pre-training raised Recall@1 over FDSL from 65.6% to 74.0% on Stanford Online Products and from 56.2% to 78.4% on Stanford Cars. The abstract also reports competitive downstream classification and improved robustness under several corruptions; it does not provide further details on those evaluations.

### Source excerpt

arXiv:2610.39423v1 Announce Type: cross Abstract: Formula-Driven Supervised Learning (FDSL) enables vision pre-training without real images, but existing methods cast pre-training as classification on discretized labels, discarding continuity in the generative process. We propose Formula-Driven Supervised Metric Learning (FDSML), which derives relational supervision from image-generation parameters beyond discrete class labels. FDSML combines FineFractalDB, our refined fractal dataset with higher intra-class variation and more stable structures, with a Parameter-Aware Triplet (PA-Triplet) Loss that assigns positives and negatives according to distances in the underlying Iterated Function System parameter space. This objective trains embeddings to respect parameter-derived relative similarity orderings. Experiments show that FDSML improves transferability to image retrieval while remaining competitive on downstream classification and improving robustness under several corruption settings. On ResNet-50, Recall@1 improves from 65.6% to 74.0% on Stanford Online Products and from 56.2% to 78.4% on Stanford Cars over FDSL pre-training. These results suggest that parameter-derived relational supervision provides a stronger synthetic pre-training signal than label-based classification for transfer beyond recognition.

## GeoLEON-CLIP: Graph-Anchored Multimodal Contrastive Learning for Geographic Region Representation

DevFeed: [GeoLEON-CLIP: Graph-Anchored Multimodal Contrastive Learning for Geographic Region Representation](<https://devfeed.tech/articles/geoleon-clip-graph-anchored-multimodal-contrastive-learning-for-geographic-region-representation-99599.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32311>)

Author: Radoslaw Malek, Szymon Soltysiak, Aleksandra Kawala-Sterniuk, Piotr Szymanski

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [geospatial](<https://devfeed.tech/topics/geospatial.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Graphs](<https://devfeed.tech/topics/graphs.md>)

Tags: [airbnb](<https://devfeed.tech/tags/airbnb.md>), [align](<https://devfeed.tech/tags/align.md>), [anchor](<https://devfeed.tech/tags/anchor.md>), [clip](<https://devfeed.tech/tags/clip.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [framework](<https://devfeed.tech/tags/framework.md>), [geospatial](<https://devfeed.tech/tags/geospatial.md>), [graph](<https://devfeed.tech/tags/graph.md>), [graph-embeddings](<https://devfeed.tech/tags/graph-embeddings.md>), [housing](<https://devfeed.tech/tags/housing.md>), [image](<https://devfeed.tech/tags/image.md>), [improvements](<https://devfeed.tech/tags/improvements.md>), [intermediate](<https://devfeed.tech/tags/intermediate.md>), [language](<https://devfeed.tech/tags/language.md>), [learning](<https://devfeed.tech/tags/learning.md>), [map](<https://devfeed.tech/tags/map.md>), [maps](<https://devfeed.tech/tags/maps.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [openstreetmap](<https://devfeed.tech/tags/openstreetmap.md>), [region](<https://devfeed.tech/tags/region.md>)

### AI overview

GeoLEON-CLIP aligns three views of each H3-indexed region: an OpenStreetMap-derived graph embedding, a satellite image crop, and a textual description. It uses frozen pretrained LEON graph embeddings as an anchor, with frozen graph, image, and text encoders linked by trainable residual projection heads and optimized using a symmetric tri-modal contrastive objective. The abstract reports strong graph-image and graph-text alignment, with graph-image retrieval performing best in a mixed-resolution pool. Chained retrieval through one graph cell improves R@1@1 over direct image-text retrieval by 4.8 points for image-to-text and 4.4 points for text-to-image. Downstream gains vary by task and are largest on Airbnb and Beijing housing.

### Source excerpt

arXiv:2610.32311v1 Announce Type: cross Abstract: Geographic regions are multimodal: satellite imagery captures visual morphology, text conveys human-interpretable semantics, and map structures describe urban context. Existing geospatial contrastive methods typically align image-text or image-location pairs, underusing structured map data. We introduce GeoLEON-CLIP, a graph-anchored contrastive framework that aligns three views of each H3-indexed region: an OpenStreetMap-derived graph embedding, a satellite image crop, and a textual description. Frozen pretrained LEON graph embeddings provide a semantic anchor between vision and language. Frozen graph, image, and text encoders are connected by trainable residual projection heads and trained with a symmetric tri-modal contrastive objective. Experiments show strong graph-image and graph-text alignment; graph-image retrieval achieves the best cross-modal performance in a mixed-resolution pool. To evaluate the anchoring hypothesis, we propose chained cross-modal recall $\mathrm{R}@A@B$, which routes retrieval through the graph modality and compares it with direct retrieval under a matched candidate budget. At $\mathrm{R}@1@1$, image$\rightarrow$graph$\rightarrow$text and text$\rightarrow$graph$\rightarrow$image outperform direct image-text retrieval by 4.8 and 4.4 points, respectively, despite using a single intermediate cell. Downstream probes show task-dependent gains, with the largest improvements on Airbnb and Beijing housing. These results indicate that OpenStreetMap graph structure can anchor multimodal representations combining maps, imagery, and language.

## PlayerEmbed: Learning Behavioral Fingerprint Representations from Raw Input Sequences

DevFeed: [PlayerEmbed: Learning Behavioral Fingerprint Representations from Raw Input Sequences](<https://devfeed.tech/articles/playerembed-learning-behavioral-fingerprint-representations-from-raw-input-sequences-99648.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32727>)

Author: Tobias Schneider, Lorenz Sparrenberg, Priya Tomar, Shahzeb Qamar, Tobias Deu{\\ss}er, Rafet Sifa

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [generalization in machine learning](<https://devfeed.tech/topics/generalization-in-machine-learning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [analytics](<https://devfeed.tech/tags/analytics.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [cnn](<https://devfeed.tech/tags/cnn.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [deep](<https://devfeed.tech/tags/deep.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [fingerprint](<https://devfeed.tech/tags/fingerprint.md>), [framework](<https://devfeed.tech/tags/framework.md>), [games](<https://devfeed.tech/tags/games.md>), [hybrid](<https://devfeed.tech/tags/hybrid.md>), [input](<https://devfeed.tech/tags/input.md>), [learning](<https://devfeed.tech/tags/learning.md>), [lstm](<https://devfeed.tech/tags/lstm.md>), [metric](<https://devfeed.tech/tags/metric.md>), [metric-learning](<https://devfeed.tech/tags/metric-learning.md>), [model](<https://devfeed.tech/tags/model.md>), [patterns](<https://devfeed.tech/tags/patterns.md>), [performance](<https://devfeed.tech/tags/performance.md>), [raw](<https://devfeed.tech/tags/raw.md>), [reference](<https://devfeed.tech/tags/reference.md>), [sequences](<https://devfeed.tech/tags/sequences.md>)

### AI overview

PlayerEmbed learns compact player representations from noisy raw video-game input sequences using deep metric learning and a hybrid CNN-LSTM encoder trained on data from 1,000 players. A nearest-centroid classifier using reference embeddings reaches balanced accuracy of 0.950 on held-out sequences from training identities and 0.879 on sequences from 1,000 identities absent from training. Increasing the number of training players substantially improves generalization to unseen identities, which the authors interpret as learning transferable behavioral patterns beyond simple fingerprinting.

### Source excerpt

arXiv:2610.32727v1 Announce Type: new Abstract: Raw player input sequences in video games contain fine-grained behavior, but are high-dimensional and noisy, which makes large-scale analytics challenging. To address this, we introduce PlayerEmbed, a framework that learns compact and discriminative player embeddings using deep metric learning. Our approach uses a hybrid CNN-LSTM encoder trained on raw input sequences from 1,000 players. We evaluate the model on held-out sequences from these players and on sequences from 1,000 players entirely unseen during training to assess generalization to unseen identities. For player identification, a nearest-centroid classifier built from reference embeddings achieves strong performance, reaching balanced accuracies of 0.950 for players seen during training and 0.879 for players unseen during training. Systematically increasing the number of training players substantially improves generalization to unseen players, indicating a transition from simple fingerprinting to learning transferable behavioral patterns.

## Latent space configuration for neural network finetuning, fast ranking, and embedding dimensionality reduction

DevFeed: [Latent space configuration for neural network finetuning, fast ranking, and embedding dimensionality reduction](<https://devfeed.tech/articles/latent-space-configuration-for-neural-network-finetuning-fast-ranking-and-embedding-dimensionality-reduction-99864.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.29295>)

Author: Nikita Gabdullin, Anton Raskovalov

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [AI-driven information retrieval](<https://devfeed.tech/topics/ai-driven-information-retrieval.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [accelerate](<https://devfeed.tech/tags/accelerate.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [complexity](<https://devfeed.tech/tags/complexity.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [neural-network](<https://devfeed.tech/tags/neural-network.md>), [reduction](<https://devfeed.tech/tags/reduction.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [search](<https://devfeed.tech/tags/search.md>)

### AI overview

The paper combines Latent Space Configuration (LSC) finetuning of ArcFace-pretrained IResNet50 models with search algorithms that exploit structured embedding spaces to narrow top-K retrieval to a small candidate subset with stated O(1) complexity. Against exhaustive search, the authors report up to 3x speed-up on GPU and 40% on CPU relative to the baseline method's GPU speed. They report results with both original and reduced embedding dimensions; dimensionality reduction further accelerates search and can improve rank-N metrics. These are abstract-level reported findings, with no benchmark details provided in the supplied text.

### Source excerpt

arXiv:2610.29295v2 Announce Type: replace-cross Abstract: Finding K most similar elements for a query in a large gallery of neural network (NN) embeddings is essential in various areas of computer science and industrial applications of artificial intelligence. This paper proposes search algorithms for structured latent spaces (LS) of NNs which allow to speed up top K item retrieval owing to O(1) complexity small candidate gallery subset search. ArcFace-pretrained IResNet50 models are finetuned using Latent Space Configuration (LSC) methodology to obtain new distribution of embeddings with desired geometric properties. Applying the proposed algorithms to embeddings obtained from finetuned models leads to several times speed-up over baseline exhaustive search. Specifically, we report up to 3 times speed-up on GPU and 40% speed-up on CPU compared to the baseline method GPU speed. These results are obtained for the original and reduced numbers of LS dimensions. The results show that LS dimensionality reduction allows to further accelerate the search and even improve rank-N metrics.

## Rethinking Post-hoc Debiasing of Large Pretrained Models

DevFeed: [Rethinking Post-hoc Debiasing of Large Pretrained Models](<https://devfeed.tech/articles/rethinking-post-hoc-debiasing-of-large-pretrained-models-94979.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.29688>)

Author: Divyaansh Mertia, Chirag Bhutra, Himanshu Shekhar, Shubham Bagwari, Pratik Mazumder

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [metric learning](<https://devfeed.tech/topics/metric-learning.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [all](<https://devfeed.tech/tags/all.md>), [annotations](<https://devfeed.tech/tags/annotations.md>), [approach](<https://devfeed.tech/tags/approach.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [based](<https://devfeed.tech/tags/based.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [bias](<https://devfeed.tech/tags/bias.md>), [body](<https://devfeed.tech/tags/body.md>), [class](<https://devfeed.tech/tags/class.md>), [clip](<https://devfeed.tech/tags/clip.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [dependency](<https://devfeed.tech/tags/dependency.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [example](<https://devfeed.tech/tags/example.md>), [large](<https://devfeed.tech/tags/large.md>), [models](<https://devfeed.tech/tags/models.md>), [post](<https://devfeed.tech/tags/post.md>)

### AI overview

Post-hoc debiasing methods for frozen pretrained models can suffer when they require identifying and expressing bias attributes in text. On UrbanCars, where two biases co-occur, recent methods underperform a vanilla classifier on biased frozen features, and ground-truth bias identities do not close the gap. The paper finds that cross-class training pairs nearest in the encoder's frozen feature geometry almost always share a spurious attribute, without needing bias labels. It uses such pairs to learn a single linear map for debiasing, reporting competitive performance across datasets without bias annotations or LLMs.

### Source excerpt

arXiv:2610.29688v1 Announce Type: cross Abstract: Large pretrained models such as CLIP are widely used as frozen feature extractors. However, their embeddings inherit spurious correlations from training data, introducing bias into the features. A growing body of research has sought to address this issue through post hoc debiasing, leaving the backbone untouched. We discover that despite differing mechanisms, these methods require the bias feature to be identified and expressed in text. This dependency is not innocuous. In this work, we experimentally demonstrate that this requirement incurs an unforeseen penalty in certain scenarios. For example, on the UrbanCars benchmark, having two co-occurring bias attributes, the recent methods underperform a vanilla classifier on biased frozenbackbone features, and even providing ground-truth bias identities does not close the gap. We show that the pattern that such methods seek is readily observable in the frozen geometry. Specifically, cross-class training pairs placed closest together by the frozen encoder almost always share a spurious attribute, regardless of the bias type, without requiring access to the bias labels. Based on this, we propose a novel approach to identify such pairs and learn a single linear map that transforms the features to perform post-hoc debiasing. The proposed approach achieves competitive performance across all datasets without access to bias attribute annotations or any LLMs.

## Multi-Marginal Inverse Optimal Transport for Contrastive Learning Via Explicit Anchor-Positive-Negative Coupling

DevFeed: [Multi-Marginal Inverse Optimal Transport for Contrastive Learning Via Explicit Anchor-Positive-Negative Coupling](<https://devfeed.tech/articles/multi-marginal-inverse-optimal-transport-for-contrastive-learning-via-explicit-anchor-positive-negative-coupling-94166.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.33741>)

Author: Ngoc-Hai Nguyen, Thuan Nguyen, Prakash Ishwar, Shuchin Aeron

Published: 2026-10-21T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>)

Tags: [anchor](<https://devfeed.tech/tags/anchor.md>), [cl](<https://devfeed.tech/tags/cl.md>), [class](<https://devfeed.tech/tags/class.md>), [cost](<https://devfeed.tech/tags/cost.md>), [coupling](<https://devfeed.tech/tags/coupling.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [dimension](<https://devfeed.tech/tags/dimension.md>), [learning](<https://devfeed.tech/tags/learning.md>), [matching](<https://devfeed.tech/tags/matching.md>), [model](<https://devfeed.tech/tags/model.md>), [neural](<https://devfeed.tech/tags/neural.md>), [number](<https://devfeed.tech/tags/number.md>), [optimal-transport](<https://devfeed.tech/tags/optimal-transport.md>), [ot](<https://devfeed.tech/tags/ot.md>), [pair](<https://devfeed.tech/tags/pair.md>), [pairs](<https://devfeed.tech/tags/pairs.md>), [performance](<https://devfeed.tech/tags/performance.md>), [points](<https://devfeed.tech/tags/points.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [transport](<https://devfeed.tech/tags/transport.md>)

### AI overview

Neg-MMIOT-CL adds explicit negative examples to inverse optimal transport contrastive learning, which earlier pairwise OT methods omit. It learns from a multi-marginal coupling over anchor-positive-negative triplets using a designed triplet ground cost. Experiments on a latent class model indicate that it alleviates dimensional collapse; the authors also prove that, for a specific ground cost, its optimal representations have equiangular within-class and across-class structure, yielding Neural Collapse when representation dimension exceeds the number of classes minus one. Neg-IOT-CL-PushPull is presented as a more computationally efficient alternative that reduces the cost of computing multi-marginal OT plans. On synthetic and real-world datasets, the methods reportedly improve on existing OT-based contrastive learning methods.

### Source excerpt

arXiv:2609.33741v3 Announce Type: replace-cross Abstract: Inverse Optimal Transport (OT) based methods for representation learning learn representations such that the global OT coupling between a pair of data marginals in the representation space, concentrates on the positive pairs. This is in contrast to previous methods that primarily focused on pairwise matching. However, these methods $\textit{DO NOT}$ utilize negative pairs and hence are not truly contrastive in their approach. We show that this leads to issues of dimensional collapse and hence degraded downstream performance. To alleviate this, we develop a novel multi-marginal (MM) inverse OT (IOT) contrastive learning (CL) approach called Neg-MMIOT-CL, which learns representations such that the global multi-marginal OT (MMOT) coupling between a triple of data marginals, with respect to a carefully designed ground-cost between triplets of data points in the representation space, concentrates on the anchor-positive-negative $\textit{triplets}$. For a latent class model, we empirically show that Neg-MMIOT-CL alleviates dimensional collapse. Furthermore, for a specific choice of ground cost for all triplets in representation space, we prove that the optimal representation configuration for Neg-MMIOT-CL exhibits equiangular property for within-class and across-class representations, which translates to Neural-Collapse when the representation dimension is larger than the number of classes minus one -- a result that is $\textit{previously established only}$ for pairwise contrastive learning methods. Finally, we propose Neg-IOT-CL-PushPull, that is a computationally efficient alternative to Neg-MMIOT-CL, alleviating the high cost of computing MMOT plans needed during implementation. We apply these methods on both synthetic and real-world datasets and show significant improvements over existing OT-based contrastive learning methods.

## Personalized Fishbowl Recommendations with Learned Embeddings: Part 2

DevFeed: [Personalized Fishbowl Recommendations with Learned Embeddings: Part 2](<https://devfeed.tech/articles/personalized-fishbowl-recommendations-with-learned-embeddings-part-2-22616.md>)

Original publisher: [Read original article](<https://medium.com/glassdoor-engineering/personalized-fishbowl-recommendations-with-learned-embeddings-part-2-78a16b04d396?source=rss----288d984af747---4>)

Author: Ahmad Khan

Published: 2022-04-05T00:24:03Z

Content type: article

Language: en

Sources: [Glassdoor Engineering](<https://devfeed.tech/sources/glassdoor-engineering.md>)

Topics: [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [recommendation systems](<https://devfeed.tech/topics/recommendation-systems.md>), [vector embeddings in AI](<https://devfeed.tech/topics/vector-embeddings-in-ai.md>), [personalization](<https://devfeed.tech/topics/personalization.md>), [metric learning](<https://devfeed.tech/topics/metric-learning.md>), [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [cosine similarity search](<https://devfeed.tech/topics/cosine-similarity-search.md>), [information retrieval](<https://devfeed.tech/topics/information-retrieval.md>), [dataset](<https://devfeed.tech/topics/dataset.md>)

Tags: [cosine-similarity](<https://devfeed.tech/tags/cosine-similarity.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [doc2vec](<https://devfeed.tech/tags/doc2vec.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [features](<https://devfeed.tech/tags/features.md>), [fishbowl](<https://devfeed.tech/tags/fishbowl.md>), [nlp](<https://devfeed.tech/tags/nlp.md>), [personalization](<https://devfeed.tech/tags/personalization.md>), [post](<https://devfeed.tech/tags/post.md>), [rank](<https://devfeed.tech/tags/rank.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [recommendations](<https://devfeed.tech/tags/recommendations.md>), [recommender-systems](<https://devfeed.tech/tags/recommender-systems.md>), [text](<https://devfeed.tech/tags/text.md>), [word-embeddings](<https://devfeed.tech/tags/word-embeddings.md>)

### AI overview

The article compares learned embeddings for Fishbowl recommendations when clickstream data lacks clear negative signals. Its GloVe method combines six text features for users and posts, adds popularity signals and liked-neighbor information, then uses PCA to reduce the representation to 128 dimensions. BERT uses the same features but averages outputs from its last four attention layers; the authors note Sentence BERT as a possible future approach. Siamese and graph convolutional networks learn user and post embeddings with triplet loss, while the graph model also pools sampled neighbors and tests harder negatives. In a 12-month training and one-week test evaluation, GloVe with PCA performed best and was shipped for production testing; the neural models did slightly worse and overfit quickly. The authors suggest simpler models suited their data, while better negative sampling might improve neural results.

### Source excerpt

Introduction In the previous blog post, we saw how we can utilize text based embeddings to help recommend posts to users on Fishbowl, a professional networking community that was recently acquired by Glassdoor, in which working professionals can have workplace related conversations with other peers in industry. On Fishbowl, users can anonymously write what's on their mind in posts and also comment on posts from other anonymous users in what we call "bowls" or "feeds"; a collection of posts related to a certain industry or topic. Previously we discussed how in the absence of clear negative signals from click stream data we cannot as easily use a supervised learning setup to rank items to users. Given the inputs we used were text based, we can instead use more unsupervised methods like Doc2Vec [1] to generate post text embeddings. We can treat the text of the posts as individual documents and use those to train a Doc2Vec model to generate a post text embedding. For users, we can take the average post text embedding of the posts the user liked and consider that as the user embedding. We can also add the user embeddings of other users who liked a post into our post embedding calculation so we incorporate some "collaborative" notion of what other similar users liked as opposed to a pure content similarity ranking. We can next compute the cosine similarity between the user and posts embeddings and use the similarity score to rank posts to recommend to users. Ranking via Embeddings Overview However, such an approach can have some shortcomings. First the text of the post is just one of many features we can use for personalization. We can also leverage other user provided information when users agree to sign up like the employer, job title, work city of a user as features. In addition, the bowl or feed name and its description can be considered important features for posts as posts under the same feed tend to be about similar topics. Second, the previous Doc2Vec approach tra