# ImageNet

ImageNet is an image dataset organized according to the WordNet hierarchy, providing image data for training large-scale object-recognition models.

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

## Where Global Context and Weight Sharing Work in Compact Vision Backbones

DevFeed: [Where Global Context and Weight Sharing Work in Compact Vision Backbones](<https://devfeed.tech/articles/where-global-context-and-weight-sharing-work-in-compact-vision-backbones-142656.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23446>)

Author: Michael Schroeder

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 recognition](<https://devfeed.tech/topics/image-recognition.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [batchnorm](<https://devfeed.tech/tags/batchnorm.md>), [block](<https://devfeed.tech/tags/block.md>), [blocks](<https://devfeed.tech/tags/blocks.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [classification](<https://devfeed.tech/tags/classification.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [convolution](<https://devfeed.tech/tags/convolution.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>), [efficientnet](<https://devfeed.tech/tags/efficientnet.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [global](<https://devfeed.tech/tags/global.md>), [global-context](<https://devfeed.tech/tags/global-context.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [scale](<https://devfeed.tech/tags/scale.md>), [self-attention](<https://devfeed.tech/tags/self-attention.md>), [sharing](<https://devfeed.tech/tags/sharing.md>), [stage](<https://devfeed.tech/tags/stage.md>), [vision](<https://devfeed.tech/tags/vision.md>), [weight](<https://devfeed.tech/tags/weight.md>), [work](<https://devfeed.tech/tags/work.md>)

### AI overview

Controlled EfficientNet-B0 experiments on ImageNet-1k suggest that global context helps when added to an intact MBConv trunk: zero-gated residual attention adapters at P3 (windowed) and P4/P5 (global) scored 73.88% versus 72.92% for a matched plain-B0 control, with attention amplitude increasing at deeper stages. A BiFPN wrapper alone scored 69.99-71.81%, below plain B0's 72.86%; adding stage attention raised it to 73.15%. For weight sharing, averaging kernels collapsed accuracy to 17.03%, while tying one template across a stage reduced parameters 27.6% but reached only 66.10%. Alternating two shared templates in Stage 6 cut parameters 10.9% (582K weights) and reached 70.92% after one recovery epoch, 1.84 points below the unshared control. Findings are limited to the described experiments and training schedule.

### Source excerpt

arXiv:2611.23446v1 Announce Type: cross Abstract: Compact convolutional backbones such as EfficientNet achieve strong parameter efficiency through local depthwise separable convolutions, yet two structural questions remain central to hybrid design: where does global self-attention improve recognition over pure convolution, and can repeated blocks share weights without losing representational capacity? We investigate both questions on ImageNet-1k (50,000-image validation) using controlled EfficientNet-B0 experiments. First, we show that wrapping a classification backbone in multi-scale feature pyramids (BiFPN) fails to outperform a plain MBConv trunk (69.99% - 71.81% vs. 72.86% plain B0) until paired with stage attention (73.15%). In contrast, keeping the MBConv trunk intact and inserting lightweight zero-gated residual self-attention adapters at stages P3 (28x28 windowed), P4 (14x14 global), and P5 (7x7 global) reaches 74.02% standalone and outperforms a bit-for-bit matched plain-B0 control by +0.954 percentage points (73.88% vs. 72.92%) across an 88-epoch stateless replay schedule, with learned attention amplitude rising monotonically with semantic depth (27.2% at P3 -> 45.4% at P4 -> 66.3% at P5). Second, we evaluate recurrent weight sharing across repeated MBConv blocks with pass-specific BatchNorm statistics (+1,536 parameters per pass). While post-hoc kernel averaging degrades accuracy to 17.03% and single-template stage tying (-27.6% parameters) plateaus at 66.10%, alternating two shared block templates ([B1, A, B, A]) in Stage 6 cuts total model parameters by 10.9% (582K weights) and recovers by +10.82 points in a single epoch to 70.92% (-1.84 points vs. unshared control).

## Isthmus: Bridging Supervision Islands Restores Uncertainty and Stability in One-Step Flows

DevFeed: [Isthmus: Bridging Supervision Islands Restores Uncertainty and Stability in One-Step Flows](<https://devfeed.tech/articles/isthmus-bridging-supervision-islands-restores-uncertainty-and-stability-in-one-step-flows-143415.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23616>)

Author: Luming Yang, Siqing Li, Haoxian Liu, Zhenyu Bu, Jiazheng Zhu, Alper Yilmaz

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

Content type: article

Language: en

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

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [bounded](<https://devfeed.tech/tags/bounded.md>), [brain](<https://devfeed.tech/tags/brain.md>), [bridging](<https://devfeed.tech/tags/bridging.md>), [cifar-10](<https://devfeed.tech/tags/cifar-10.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [consistency](<https://devfeed.tech/tags/consistency.md>), [correlation](<https://devfeed.tech/tags/correlation.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [edge](<https://devfeed.tech/tags/edge.md>), [error](<https://devfeed.tech/tags/error.md>), [failures](<https://devfeed.tech/tags/failures.md>), [flow](<https://devfeed.tech/tags/flow.md>), [generation](<https://devfeed.tech/tags/generation.md>), [identity](<https://devfeed.tech/tags/identity.md>), [inference](<https://devfeed.tech/tags/inference.md>), [island](<https://devfeed.tech/tags/island.md>), [less](<https://devfeed.tech/tags/less.md>), [one](<https://devfeed.tech/tags/one.md>), [supervision](<https://devfeed.tech/tags/supervision.md>), [uncertainty](<https://devfeed.tech/tags/uncertainty.md>)

### AI overview

One-step flow samplers can lose sample variance and correlation relative to multi-step samples; directly supervising the one-step jump worsens collapse and can destabilize extended training. The paper attributes both problems to supervision gaps in the start/end-time triangle: the one-step pair is unsupervised or lies on an island disconnected from data-targeted diagonal pairs by the self-consistency identity. Its "isthmus" assigns 5% of training pairs along the edge linking that pair to the diagonal, bounding one-step map error by self-consistency residual along the edge. The authors report improved one-step sampling across several flow-map objectives and tasks, with network, loss, and inference code unchanged and no added cost. In brain MRI, calibration reaches 0.96 and correlation with multi-step samples 0.93; both vary less across seeds, with agreement maintained at three times the training length.

### Source excerpt

arXiv:2611.23616v1 Announce Type: cross Abstract: One-step flow maps are increasingly used as posterior samplers, where the point of sampling is the uncertainty that the samples carry. We find that the one-step sampler loses this uncertainty and becomes less stable. Compared with multi-step samples of the same network, its samples lose variance and decorrelate. Supervising the one-step jump directly makes this worse; the samples collapse further toward the conditional mean and break down when training is extended. We trace both failures to the locations where training places its pairs of start and end times. Only pairs on the diagonal of the time triangle receive targets from data, and the pair used for one-step sampling is either left unsupervised or supervised on an island, a set of pairs that the self-consistency identity does not connect to the diagonal, where the objective neither determines the one-step map nor anchors its training. We bridge the island with an isthmus, which places 5% of the training pairs along the whole edge of the triangle that joins the one-step pair to the diagonal. The error of the one-step map is then bounded by the self-consistency residual along this edge. With the network, the loss and the inference code unchanged and at no added cost, the isthmus improves one-step sampling across several flow-map objectives, in MRI reconstruction, face inpainting, and unconditional generation on CIFAR-10 and ImageNet 32x32. On brain MRI, it restores one-step calibration to 0.96 and the correlation with multi-step samples to 0.93; both vary far less across seeds, and the agreement holds when training is extended to three times its length.

## Dual-Manifold Variational Autoencoders

DevFeed: [Dual-Manifold Variational Autoencoders](<https://devfeed.tech/articles/dual-manifold-variational-autoencoders-142732.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23964>)

Author: Hongxu Chen, Chun Wang, Yanghao Wang, Hongxiang Li, Zhen Wang, Zhixiang Zhang, Long Chen

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

Content type: article

Language: en

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

Topics: [VAE](<https://devfeed.tech/topics/vae.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [autoencoders](<https://devfeed.tech/tags/autoencoders.md>), [cfg](<https://devfeed.tech/tags/cfg.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [dimension](<https://devfeed.tech/tags/dimension.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [domains](<https://devfeed.tech/tags/domains.md>), [drift](<https://devfeed.tech/tags/drift.md>), [features](<https://devfeed.tech/tags/features.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generator](<https://devfeed.tech/tags/generator.md>), [guidance](<https://devfeed.tech/tags/guidance.md>), [inference](<https://devfeed.tech/tags/inference.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [space](<https://devfeed.tech/tags/space.md>), [subspace](<https://devfeed.tech/tags/subspace.md>), [training](<https://devfeed.tech/tags/training.md>), [ts](<https://devfeed.tech/tags/ts.md>)

### AI overview

DM-VAE addresses off-manifold drift in high-dimensional VAE latents by pairing a reconstructive branch with a low-dimensional branch that supplies a smooth topological skeleton to anchor the latent manifold. It avoids relying on a frozen foundation model as its primary regularizer. Topological representation alignment (TRA) supervises intermediate generator features during training; internal guidance with topological skeleton (IG_TS) guides structural components at inference while preserving high-frequency detail. On ImageNet-1K at 256x256, the foundation-model-free method reports rFID 0.12 and FID50K 1.28 without classifier-free guidance (CFG), or 1.20 with CFG. Combined orthogonally with DINOv2, it reports FID50K 1.11.

### Source excerpt

arXiv:2611.23964v1 Announce Type: cross Abstract: In latent diffusion models, expanding the latent dimension of variational autoencoders (VAEs) is essential to push the upper bound of reconstruction quality. However, without explicit structural constraints, such unregularized expansion easily yields a rugged and ill-conditioned latent manifold, causing generated latents to drift off the manifold during generation (i.e., off-manifold problem). Recent works smooth this manifold by using external foundation models (e.g., DINOv2) as a topological regularizer. However, we argue that tying the latent space to frozen external priors inherently restricts generalization: bound by their pre-training distributions and semantic biases, they may fail to capture faithful visual details on out-of-distribution domains. Observing that low-dimensional latents naturally form a smoother, generation-friendly manifold, we propose dual-manifold VAE (DM-VAE) to leverage internal representations as a topological regularizer. Specifically, DM-VAE pairs a high-dimensional reconstructive branch with a low-dimensional branch that extracts a clean topological skeleton, anchoring the high-dimensional manifold to this smooth subspace. Building upon this framework, we further introduce topological representation alignment (TRA) to supervise intermediate generator features during training, and internal guidance with topological skeleton (IG_TS) to guide structural components while preserving high-frequency details at inference. On ImageNet-1K (256x256), our foundation-model-free DM-VAE achieves a state-of-the-art rFID of 0.12 and a generation FID50K of 1.28 (w/o CFG) and 1.20 (w/CFG), and further reaches 1.11 when orthogonally combined with DINOv2.

## ResoLUT: Distilling Residual Networks into Lookup-Table Networks

DevFeed: [ResoLUT: Distilling Residual Networks into Lookup-Table Networks](<https://devfeed.tech/articles/resolut-distilling-residual-networks-into-lookup-table-networks-141824.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22955>)

Author: Katarzyna Fojcik, Aidas Mikalauskas, Matthias Kampa, Renaldas Zioma, Jogundas Armaitis

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

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

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [cifar-10](<https://devfeed.tech/topics/cifar-10.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [Embedded Systems](<https://devfeed.tech/topics/embedded-systems.md>)

Tags: [batchnorm](<https://devfeed.tech/tags/batchnorm.md>), [boolean](<https://devfeed.tech/tags/boolean.md>), [budget](<https://devfeed.tech/tags/budget.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distilling](<https://devfeed.tech/tags/distilling.md>), [fpga](<https://devfeed.tech/tags/fpga.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [inference](<https://devfeed.tech/tags/inference.md>), [learning](<https://devfeed.tech/tags/learning.md>), [logic](<https://devfeed.tech/tags/logic.md>), [lookup](<https://devfeed.tech/tags/lookup.md>), [lookup-tables](<https://devfeed.tech/tags/lookup-tables.md>), [lut](<https://devfeed.tech/tags/lut.md>), [networks](<https://devfeed.tech/tags/networks.md>), [optimizations](<https://devfeed.tech/tags/optimizations.md>), [optimized](<https://devfeed.tech/tags/optimized.md>), [residual](<https://devfeed.tech/tags/residual.md>), [resnet](<https://devfeed.tech/tags/resnet.md>), [results](<https://devfeed.tech/tags/results.md>), [scale](<https://devfeed.tech/tags/scale.md>)

### AI overview

ResoLUT converts a trained ResNet layer by layer into lookup-table trees, taking connectivity from the teacher instead of learning it. It preserves the multi-bit residual stream and BatchNorm, and uses hard LUT decisions during both training and inference. The reported results include 95.66% on CIFAR-10 for ResoLUT-20 with twice the trees, and 78.87% on CIFAR-100 and 53.69% on 64px ImageNet for ResoLUT-56. The most optimized ResoLUT-8 uses one eighth of the trees, fits on an FPGA, and reaches 90.04% on CIFAR-10 at 189,000 images per second. These figures are abstract-level claims; the supplied text does not include experimental details.

### Source excerpt

arXiv:2611.22955v1 Announce Type: new Abstract: Logic networks built from Boolean gates or lookup tables (LUTs) promise efficient hardware inference, but training them at depth and learning their connectivity remain difficult. We do not train logic as logic. Instead, ResoLUT distills a trained residual network (ResNet) layer by layer into LUT trees, preserving its multi-bit residual stream and BatchNorm while extracting connectivity from the teacher rather than learning it. To our knowledge, ResoLUT is the first convolutional logic network in computer vision to retain a ResNet-style multi-bit residual stream. Hard LUT decisions are used throughout training and inference, avoiding a discretization gap in the logic. ResoLUT-20 reaches 95.66% on CIFAR-10 with twice the trees, and ResoLUT-56 reaches 78.87% on CIFAR-100 and 53.69% on ImageNet at 64px, the best results of logic and lookup-table networks on CIFAR-10 and CIFAR-100, showing that LUT networks can scale in depth. We explore and quantify the role of the residuals and the tree budget and shape. The most optimized model, a ResoLUT-8 with one eighth of the trees and further optimizations, fits on an FPGA and reaches 90.04% at 189k images per second, the highest throughput among the compared models above 85% on CIFAR-10.

## Correcting Flows with Marginal Matching

DevFeed: [Correcting Flows with Marginal Matching](<https://devfeed.tech/articles/correcting-flows-with-marginal-matching-141851.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23134>)

Author: Meital Bojan, Nadav Bojan Sellam, Ev Zisselman, Aviv Tamar

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

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

Topics: [cifar-10](<https://devfeed.tech/topics/cifar-10.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [cifar-10](<https://devfeed.tech/tags/cifar-10.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [matching](<https://devfeed.tech/tags/matching.md>), [ode](<https://devfeed.tech/tags/ode.md>)

### AI overview

Marginal Matching adds a training-based predictor-corrector framework to flow matching models by using the reversibility of their ODE: at a forward time step, it corrects samples toward the matching marginal distribution of the backward ODE. The authors say this reduces a bound on the distance between the data and learned distributions, and introduce two implementations of the correction function. They report improved sample quality on CIFAR-10 and ImageNet-64 with minimal added computation time. The supplied text does not describe the two implementations or provide quantitative results.

### Source excerpt

arXiv:2611.23134v1 Announce Type: new Abstract: Flow matching models, ODE-based generative models, generate samples by gradually morphing a simple source distribution into a target distribution. However, these models still face challenges in optimally balancing computational efficiency with generation fidelity. While predictor-corrector schemes have been proven to help address this issue in diffusion models effectively, their application to flow matching models remains under-explored. In this work we present a novel training-based predictor-corrector framework for flow matching, which exploits the fact that the ODE can be easily reversed. Our key idea, termed Marginal Matching, is to correct samples at a particular time step in the forward ODE towards a matching marginal distribution of the backward ODE. This scheme proves to reduce a bound on the distance between the data distribution and the learned distribution, motivating two different implementations for the correction function. We show that our proposed method improves sample quality on CIFAR-10 and ImageNet-64, with minimal overhead in computation time.

## A tiny object detection dataset for high-altitude open-water search and rescue

DevFeed: [A tiny object detection dataset for high-altitude open-water search and rescue](<https://devfeed.tech/articles/a-tiny-object-detection-dataset-for-high-altitude-open-water-search-and-rescue-140082.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22825>)

Author: Hui Xu, Chenfei Hou, Henglai Wei, Hao Wang, Yidong Jin, Xiulin Niu, Haoyang Zhang, Hui Zhang, Tianshi Xu

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

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

Topics: [object-detection](<https://devfeed.tech/topics/object-detection.md>), [Image](<https://devfeed.tech/topics/image.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [object-detection](<https://devfeed.tech/tags/object-detection.md>), [tiny](<https://devfeed.tech/tags/tiny.md>), [water](<https://devfeed.tech/tags/water.md>)

### AI overview

TinyWOD targets a gap in open-water search and rescue imaging at 1.5-2.0 km, where lifebuoys may occupy only a few pixels. The dataset contains 11,359 manually annotated images and 19,505 instances across varied environments, lighting, water colors, and surface ripples. Its average equivalent side length is 12.7 pixels, and 76.96% of instances are under 15 pixels. The authors introduce APtiny and benchmark mainstream and state-of-the-art detection networks with it and conventional metrics; they report that all evaluated networks effectively detect tiny waterborne objects. The abstract presents TinyWOD as a resource for high-altitude detection research, but does not provide comparative scores or details about dataset access.

### Source excerpt

arXiv:2611.22825v1 Announce Type: cross Abstract: Large aerial platforms often operate above 1.5 km to improve the efficiency of open water search and rescue (OW-SAR). However, long imaging distances, wide fields of view, and limited sensor resolution reduce objects to a few pixels, increasing their susceptibility to environmental interference and making detection harder than for conventional small objects. To our knowledge, existing datasets focus on low-altitude scenarios and do not support automatic detection at 1.5-2.0 km. We therefore present TinyWOD, a tiny object detection dataset for high-altitude OW-SAR using lifebuoys as detection targets. It comprises 11,359 manually annotated images and 19,505 annotated instances spanning diverse environments, lighting conditions, water colors, and surface ripples. Compared with existing datasets, TinyWOD contains much smaller objects and a substantially higher proportion of tiny instances; its average equivalent side length (ESL) is 12.7 pixels, with 76.96% of in stances having ESLs below 15 pixels, reflecting actual high-altitude water-rescue imaging conditions. We define APtiny, a tiny object detection metric, and benchmark mainstream and state-of-the-art networks using this and conventional metrics. All evaluated networks effectively detect tiny waterborne objects, demonstrating TinyWOD's reusability and utility. We expect this dataset to support research on tiny object detection in high-altitude OW-SAR.

## How Far Is Too Far? Object Recognition Declines Monotonically with Semantic Distance

DevFeed: [How Far Is Too Far? Object Recognition Declines Monotonically with Semantic Distance](<https://devfeed.tech/articles/how-far-is-too-far-object-recognition-declines-monotonically-with-semantic-distance-137490.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19747>)

Author: Domenic Bersch, Hai Van Tran, Gemma Roig

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

Content type: article

Language: en

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

Topics: [image recognition](<https://devfeed.tech/topics/image-recognition.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [aggregation](<https://devfeed.tech/tags/aggregation.md>), [axis](<https://devfeed.tech/tags/axis.md>), [background](<https://devfeed.tech/tags/background.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [context](<https://devfeed.tech/tags/context.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [convolutional-neural-networks](<https://devfeed.tech/tags/convolutional-neural-networks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [failure](<https://devfeed.tech/tags/failure.md>), [models](<https://devfeed.tech/tags/models.md>), [object](<https://devfeed.tech/tags/object.md>), [patch](<https://devfeed.tech/tags/patch.md>), [prediction](<https://devfeed.tech/tags/prediction.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [recognition](<https://devfeed.tech/tags/recognition.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [transformer](<https://devfeed.tech/tags/transformer.md>), [transformers](<https://devfeed.tech/tags/transformers.md>), [vision](<https://devfeed.tech/tags/vision.md>), [vision-transformers](<https://devfeed.tech/tags/vision-transformers.md>)

### AI overview

On the ImageNet-OOC1k benchmark, recognition accuracy declined monotonically as object-background semantic dissimilarity increased across 35 convolutional, transformer, and vision-language models. Vision Transformer analyses found patterns consistent with object evidence being present in image patches but not consolidated into the final prediction. A spatial re-ranking strategy recovered some errors, improving in-sample accuracy by up to 5.7 percentage points. The findings characterize context sensitivity as a graded, predictable behavior; the proposed aggregation explanation is suggestive, and the reported improvement is limited to in-sample results.

### Source excerpt

arXiv:2611.19747v1 Announce Type: cross Abstract: Vision models are sensitive to object-background context, yet it remains unclear how recognition changes as the semantic distance between objects and backgrounds increases. We introduce ImageNet-OOC1k, a benchmark enabling graded control of object-background relationships along a continuous consensus visual-semantic dissimilarity axis. This formulation reveals a highly consistent monotonic decline in recognition accuracy across 35 models spanning convolutional neural networks, Vision Transformers, and vision-language models. Internal analyses of Vision Transformers reveal patterns consistent with aggregation failure, where object evidence is present at the patch level but not consolidated into the final prediction. Using this characterization as a diagnostic probe, a simple spatial re-ranking strategy recovers a subset of these errors and improves in-sample accuracy by up to 5.7 percentage points. These results establish context sensitivity as a continuous and predictable property of modern vision systems, with Vision Transformer analyses suggesting that aggregation limitations may contribute to some failure types even when object evidence remains available.

## Hard or Mislabeled? Failure Screens Enrich Vision Benchmarks for Label Errors

DevFeed: [Hard or Mislabeled? Failure Screens Enrich Vision Benchmarks for Label Errors](<https://devfeed.tech/articles/hard-or-mislabeled-failure-screens-enrich-vision-benchmarks-for-label-errors-137384.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19044>)

Author: Haolin Guo

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

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: [Computer vision](<https://devfeed.tech/topics/computer-vision.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [cifar-10](<https://devfeed.tech/topics/cifar-10.md>)

Tags: [audit](<https://devfeed.tech/tags/audit.md>), [audits](<https://devfeed.tech/tags/audits.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cifar-10](<https://devfeed.tech/tags/cifar-10.md>), [classifier](<https://devfeed.tech/tags/classifier.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.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>), [enrichment](<https://devfeed.tech/tags/enrichment.md>), [errors](<https://devfeed.tech/tags/errors.md>), [failure](<https://devfeed.tech/tags/failure.md>), [screens](<https://devfeed.tech/tags/screens.md>), [vision](<https://devfeed.tech/tags/vision.md>)

### AI overview

Failure screens for hard vision benchmarks can disproportionately select mislabeled images. On ImageNet's twice-relabeled validation set, 61% of the 1,892 images missed by all 66 supervised classifiers had wrong labels, versus 11.1% overall. As screens become stricter, the mislabeled share generally rises along a curve that predicted held-out model-family screens with a 2.4 percentage-point mean error; replays of ImageNet-D and ImageNet-Hard were within 0.1 and 6.3 points. The curve did not transfer uniformly: strict screens on ImageNet-V2 and CIFAR-10 showed stronger enrichment than predicted, while ImageNet-A's screen, which restricts choices to 200 of 1,000 classes, fell below it. Human relabeling found 67% of the all-66-failure ImageNet-V2 images mislabeled, compared with 46% predicted. The abstract also reports that 32% of ImageNet-Hard's ImageNet images have no correct label by ReImageNet. ScreenAudit and corrected answer sets are described as planned releases.

### Source excerpt

arXiv:2611.19044v1 Announce Type: cross Abstract: Hard vision benchmarks often keep images that strong classifiers get wrong. Since a classifier is also marked wrong when the label is wrong, such failure screens may collect label errors. We measure this on ImageNet's twice-relabeled validation set with 103 classifiers. Among the 1,892 images that all 66 supervised classifiers get wrong, 61% have a wrong label, versus 11.1% overall. The fewer images a screen lets through, the more of them are mislabeled, along approximately one curve. Fit on one model family, this curve predicts the screens of two held-out model families with a mean error of 2.4 percentage points, and replays of ImageNet-D's and ImageNet-Hard's screens within 0.1 and 6.3 points. On new pools, ImageNet-V2 and CIFAR-10, enrichment at the strictest screens is stronger than the curve predicts: on ImageNet-V2, by human relabeling, 67% of the images that all 66 classifiers get wrong are mislabeled, against 46% predicted. Screens that let models choose among only some of the classes, as ImageNet-A's does (200 of 1,000), fall below the curve. By ReImageNet, 32% of ImageNet-Hard's ImageNet images have no correct label. ScreenAudit, which comprises an audit tool, a script that rebuilds corrected answer sets for the 5,464 decided ImageNet-Hard images (of 5,646) from ReImageNet's gated release, and per-image audits of 1,198 benchmark images, will be released later.

## InvarOrd: Self-Supervised Ordinal Knowledge Distillation with Rank-Invariant Alignment for Label-Efficient Cross-Species Hyperspectral Fish Freshness Assessment

DevFeed: [InvarOrd: Self-Supervised Ordinal Knowledge Distillation with Rank-Invariant Alignment for Label-Efficient Cross-Species Hyperspectral Fish Freshness Assessment](<https://devfeed.tech/articles/invarord-self-supervised-ordinal-knowledge-distillation-with-rank-invariant-alignment-for-label-efficient-cross-species-hyperspectral-fish-freshness-assessment-137452.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19472>)

Author: Kazi Nabiul Alam, Omenuko Joseph Nwebonyi, Pooneh Bagheri Zadeh, Akbar Sheikh-Akbari

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

Content type: article

Language: en

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

Topics: [Post-training optimization](<https://devfeed.tech/topics/post-training-optimization.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [Model Optimizer](<https://devfeed.tech/topics/model-optimizer.md>)

Tags: [absolute](<https://devfeed.tech/tags/absolute.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [assessment](<https://devfeed.tech/tags/assessment.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>), [distillation](<https://devfeed.tech/tags/distillation.md>), [fish](<https://devfeed.tech/tags/fish.md>), [invariant](<https://devfeed.tech/tags/invariant.md>), [knowledge-distillation](<https://devfeed.tech/tags/knowledge-distillation.md>), [rank](<https://devfeed.tech/tags/rank.md>)

### AI overview

InvarOrd adapts hyperspectral fish-freshness models to a new species using 30 labelled cubes (five per freshness rank). It combines self-supervised spectral pretraining on unlabelled source data, covariance alignment only within matched ranks, and ordinal distillation into a compact student. The alignment is invariant to rank means, helping preserve freshness order. On paired salmon and cod data (800 cubes per species across 16 days), a source-trained teacher transferred poorly: 1.6-1.9 days MAE and negative QWK. The 423K-parameter student achieved 1.476 days MAE and 56.1% accuracy within one day, outperforming target-only training by 0.29 days (p = 0.031) and matching ConvNeXt-T's 1.503-day MAE with 66 times fewer parameters. Rank-invariant alignment retained more ordinal structure than global discrepancy (rank correlation 0.183 and QWK 0.112 versus 0.049 and zero), with only 0.014 days higher MAE for global discrepancy. Reported inference is 1.44 ms per cube on a GPU, with five-second readaptation.

### Source excerpt

arXiv:2611.19472v1 Announce Type: cross Abstract: Hyperspectral models of fish freshness rarely transfer between species, and labelling a new species is slow and destructive. Freshness is also ordinal, and standard domain alignment can erase that order. We present InvarOrd, which adapts to a new species from 30 labelled cubes (5 per rank) in three stages: self-supervised spectral pretraining on unlabelled source cubes, cross-species alignment by a shrinkage covariance discrepancy computed only within matched ranks, and ordinal distillation into a compact student. The alignment term is provably invariant to the rank means, so it gives no incentive to collapse the ordinal axis. On paired salmon and cod data (800 cubes each, 16 days), a source-trained teacher reaches 1.6 to 1.9 days mean absolute error (MAE) on the other species, with quadratic weighted kappa (QWK) below zero. The 423K-parameter student, 2.54 times smaller than its teacher, attains the lowest MAE (1.476 days) and highest within 1-day accuracy (56.1%) among 14 configurations, matching the best of six ImageNet-pretrained backbones (ConvNeXt-T, 1.503 days) at 66 times fewer parameters, 1.44ms per cube on a GPU and 5s re-adaptation. InvarOrd beats target-only training by 0.29 days (p = 0.031). Among alignment objectives, rank-invariant alignment retains the most ordinal structure (embedding rank correlation 0.183, QWK 0.112), whereas global discrepancy lowers the rank correlation to 0.049 and QWK to zero at only 0.014 days higher MAE.

## Quantum-inspired orthogonal adapters for certifiably robust medical image classification

DevFeed: [Quantum-inspired orthogonal adapters for certifiably robust medical image classification](<https://devfeed.tech/articles/quantum-inspired-orthogonal-adapters-for-certifiably-robust-medical-image-classification-138786.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18775>)

Author: Olga Tsurkan, Laia Domingo, Victoria Orlova, Asel Sagingalieva, Iris Zalaudek, Raffaele Dante Caposiena Caro

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

Content type: article

Language: en

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

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [Medical imaging](<https://devfeed.tech/topics/medical-imaging.md>), [Adversarial attacks](<https://devfeed.tech/topics/adversarial-attacks.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Image](<https://devfeed.tech/topics/image.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [adversarial](<https://devfeed.tech/tags/adversarial.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [butterfly](<https://devfeed.tech/tags/butterfly.md>), [certificates](<https://devfeed.tech/tags/certificates.md>), [circuit](<https://devfeed.tech/tags/circuit.md>), [classification](<https://devfeed.tech/tags/classification.md>), [classification-accuracy](<https://devfeed.tech/tags/classification-accuracy.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [connectivity](<https://devfeed.tech/tags/connectivity.md>), [cost](<https://devfeed.tech/tags/cost.md>), [covering](<https://devfeed.tech/tags/covering.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [design](<https://devfeed.tech/tags/design.md>), [diagnosis](<https://devfeed.tech/tags/diagnosis.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [image](<https://devfeed.tech/tags/image.md>), [image-classification](<https://devfeed.tech/tags/image-classification.md>), [medical](<https://devfeed.tech/tags/medical.md>), [quant-ph](<https://devfeed.tech/tags/quant-ph.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [robust](<https://devfeed.tech/tags/robust.md>)

### AI overview

Quantum-inspired orthogonal butterfly layers adapt certified pretrained image classifiers while preserving their Lipschitz bounds, using O(n log n) trainable parameters. In MNIST and MedMNIST experiments they improved certified robustness over conventional alternatives; on DermaMNIST transfer, they resisted tested attacks better than LoRA. For trichoscopic alopecia classification, adapting a certified ImageNet-pretrained network yielded about twice the certified accuracy of conventional fine-tuning and LoRA, with comparable classification accuracy and similar resistance to tested attacks on the same backbone and data split. Broader certificates, including for certain image-processing transformations, came at the cost of reduced accuracy. The abstract describes efficient classical implementation and a possible future quantum path under suitable connectivity assumptions.

### Source excerpt

arXiv:2611.18775v1 Announce Type: cross Abstract: Medical image classifiers require not only high accuracy but also reliable predictions under small input perturbations. Certified robustness provides mathematical guarantees of prediction stability, but preserving these guarantees when adapting pretrained models to new tasks remains challenging. We propose a fine-tuning approach based on quantum-inspired orthogonal butterfly layers that preserves the Lipschitz bounds of pretrained certified networks. Inspired by quantum circuits of reconfigurable beam splitters, these layers preserve orthogonality by construction using only $\mathcal{O}(n\log n)$ trainable parameters. Their structured design enables efficient classical implementation and offers a path toward future quantum implementations with logarithmic circuit depth under suitable connectivity assumptions. Controlled experiments on MNIST and MedMNIST demonstrate improved certified robustness compared with conventional alternatives and, in the DermaMNIST transfer, greater resistance to the tested attacks than low-rank adaptation (LoRA). We then apply it to trichoscopic image classification for alopecia diagnosis and alopecia areata activity assessment. By adapting a certified ImageNet-pretrained network, butterfly fine-tuning achieves approximately twice the certified accuracy of conventional fine-tuning and LoRA, with comparable classification accuracy on the same backbone and data split and similar empirical resistance to the tested adversarial attacks. Larger robustness guarantees, including certificates covering certain image-processing transformations, can be obtained at the cost of reduced accuracy. These results highlight the potential of quantum-inspired butterfly layers for parameter-efficient adaptation of medical image classifiers while preserving the mathematical guarantees underlying certified robustness.

## Simple yet Effective Semi-supervised Knowledge Distillation from Vision-Language Models via Dual-Head Optimization

DevFeed: [Simple yet Effective Semi-supervised Knowledge Distillation from Vision-Language Models via Dual-Head Optimization](<https://devfeed.tech/articles/simple-yet-effective-semi-supervised-knowledge-distillation-from-vision-language-models-via-dual-head-optimization-138712.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2505.07675>)

Author: Seongjae Kang, Dong Bok Lee, Hyungjoon Jang, Sung Ju Hwang

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

Content type: article

Language: en

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

Topics: [Post-training optimization](<https://devfeed.tech/topics/post-training-optimization.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [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>), [datasets](<https://devfeed.tech/tags/datasets.md>), [knowledge-distillation](<https://devfeed.tech/tags/knowledge-distillation.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [out-of-distribution-generalization](<https://devfeed.tech/tags/out-of-distribution-generalization.md>), [performance](<https://devfeed.tech/tags/performance.md>), [ssl](<https://devfeed.tech/tags/ssl.md>), [student](<https://devfeed.tech/tags/student.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

### AI overview

Dual-Head Optimization (DHO) addresses gradient conflicts in semi-supervised knowledge distillation from vision-language models by using separate prediction heads for supervised and distillation signals. The abstract reports that this reduces conflicts and improves feature learning over single-head KD baselines, with minimal computational overhead and test-time hyperparameter tuning without retraining. Across 15 datasets, DHO reportedly outperforms KD baselines and often smaller student models outperform their teacher; it also reports state-of-the-art results for in-distribution ImageNet semi-supervised learning and out-of-distribution generalization across ImageNet variants.

### Source excerpt

arXiv:2505.07675v4 Announce Type: replace Abstract: Semi-supervised learning (SSL) has emerged as a practical solution for addressing data scarcity challenges by leveraging unlabeled data. Recently, vision-language models (VLMs), pre-trained on massive image-text pairs, have demonstrated remarkable zero-/few-shot performance that often surpasses SSL approaches due to their exceptional generalization capabilities. This gap motivates us to question: how can we effectively harness the powerful generalization capabilities of VLMs into task-specific models? Knowledge distillation (KD) offers a natural framework for transferring VLM capabilities, but we identify that it suffers from gradient conflicts between supervised and distillation losses. To address this challenge, we propose Dual-Head Optimization (DHO), which introduces dual prediction heads for each distinct signal. We observe that DHO mitigates gradient conflicts, enabling improved feature learning compared to single-head KD baselines, with practical benefits of minimal computational overhead and test-time hyperparameter tuning without retraining. Extensive experiments across 15 datasets show that DHO consistently outperforms KD baselines, often outperforming teacher models with smaller student models. DHO also achieves new state-of-the-art performance on both in-distribution ImageNet semi-supervised learning and out-of-distribution generalization across ImageNet variants. Code and model checkpoints are publicly available at https://github.com/erjui/DHO/tree/main.

## Analyzing Learning-Dynamics Across Loss and Prediction Choices for Diffusion Models

DevFeed: [Analyzing Learning-Dynamics Across Loss and Prediction Choices for Diffusion Models](<https://devfeed.tech/articles/analyzing-learning-dynamics-across-loss-and-prediction-choices-for-diffusion-models-135850.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17967>)

Author: Dongwoo Kim

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

Content type: article

Language: en

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

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [bottleneck](<https://devfeed.tech/tags/bottleneck.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [can](<https://devfeed.tech/tags/can.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [choice](<https://devfeed.tech/tags/choice.md>), [cifar-10](<https://devfeed.tech/tags/cifar-10.md>), [compression](<https://devfeed.tech/tags/compression.md>), [conversion](<https://devfeed.tech/tags/conversion.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [d](<https://devfeed.tech/tags/d.md>), [data](<https://devfeed.tech/tags/data.md>), [data-structure](<https://devfeed.tech/tags/data-structure.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [learning](<https://devfeed.tech/tags/learning.md>), [models](<https://devfeed.tech/tags/models.md>), [prediction](<https://devfeed.tech/tags/prediction.md>)

### AI overview

The abstract reports that diffusion-model prediction and loss spaces can affect denoising accuracy and sample quality despite algebraic relationships among the nine combinations. For a fixed prediction target, changing loss space is equivalent to timestep reweighting. It also derives a squared-error floor of D−m for noise and velocity prediction through a rank-m bottleneck on D-dimensional noisy input, while clean-data prediction can use low-dimensional data structure; this leads to different necessary bottleneck capacities. Experiments with U-Net and JiT on CIFAR-10, Imagenette, and ImageNet-256 support these predictions, with compressed noise-prediction errors close to the theoretical floor. The evidence is an abstract, so it does not provide details of methods or results beyond these claims.

### Source excerpt

arXiv:2611.17967v1 Announce Type: new Abstract: Denoising generative models predict clean data ($x$), noise ($\epsilon$), or velocity ($v$), while training losses can be measured in any of these spaces. Although the resulting nine combinations are algebraically related, they can yield substantially different denoising accuracy and sample quality. We disentangle the effects of prediction and loss spaces. For a fixed prediction target, we show that changing the loss space is exactly equivalent to timestep reweighting and provide a heuristic analysis of how this reweighting affects generation quality. We then show that a rank-$m$ linear bottleneck compressing a $D$-dimensional noisy input imposes an irreducible squared-error floor of $D-m$ on noise and velocity prediction, whereas clean-data prediction can exploit low-dimensional data structure. Combining this bound with prediction-dependent error conversion yields distinct necessary conditions on bottleneck capacity. Experiments with U-Net and JiT across CIFAR-10, Imagenette, and ImageNet-256 support these theoretical predictions. Under compression, prediction choice strongly affects both denoising accuracy and sample quality, with noise-prediction errors closely matching the theoretical lower bound. Our results provide a quantitative understanding of how architectural bottlenecks interact with prediction and loss choices in denoising generative models.

## Learning Anisotropic Prediction Targets for Flow Models

DevFeed: [Learning Anisotropic Prediction Targets for Flow Models](<https://devfeed.tech/articles/learning-anisotropic-prediction-targets-for-flow-models-135825.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17767>)

Author: Rayhan Zirvi, Hansheng Chen, Gordon Wetzstein, Leonidas Guibas

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

Content type: article

Language: en

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

Topics: [ai-image-generator](<https://devfeed.tech/topics/ai-image-generator.md>), [Model Development](<https://devfeed.tech/topics/model-development.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [analyze](<https://devfeed.tech/tags/analyze.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [better](<https://devfeed.tech/tags/better.md>), [changes](<https://devfeed.tech/tags/changes.md>), [choice](<https://devfeed.tech/tags/choice.md>), [component](<https://devfeed.tech/tags/component.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>), [easier](<https://devfeed.tech/tags/easier.md>), [error](<https://devfeed.tech/tags/error.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [flow](<https://devfeed.tech/tags/flow.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [learning](<https://devfeed.tech/tags/learning.md>), [models](<https://devfeed.tech/tags/models.md>), [prediction](<https://devfeed.tech/tags/prediction.md>)

### AI overview

AnisoFlow learns the flow model's prediction target by assigning nonnegative noise weights to directions in a jointly learned basis. A closed-form map exactly recovers velocity, so the architecture, flow-matching loss, and sampler remain unchanged while training chooses a target that is easier to predict. The analysis predicts that preferred directional weights scale near-linearly with data variance, consistent with learned pixel-space weights. On ImageNet 256x256, the method improves FID from 1.90 to 1.55 over the JiT baseline without representation alignment or architectural changes. For pixel-space text-to-image generation, the abstract reports improved image quality and fewer patch-seam artifacts under reduced-precision inference, attributed to better-conditioned velocity recovery.

### Source excerpt

arXiv:2611.17767v1 Announce Type: cross Abstract: Flow-based generative models are trained on a hand-chosen prediction target, such as the velocity, the noise, or the clean data. This choice strongly affects training and sample quality. We introduce Anisotropic Flow Models (AnisoFlow), which replaces a fixed prediction target with a fully learnable one. AnisoFlow learns to scale the noise component of the target anisotropically, assigning a separate nonnegative weight to each direction in a jointly learned basis. A closed-form map recovers the velocity from the network's prediction, leaving the architecture, flow-matching loss, and sampler unchanged. Since recovery is exact, training optimizes the target purely to make it easier for the network to predict. We theoretically analyze the resulting prediction error and predict that the preferred noise weight in each direction scales near-linearly with data variance, in agreement with the weights learned empirically in pixel space. AnisoFlow improves FID from 1.90 to 1.55 over the JiT baseline on ImageNet 256$\times$256, by learning only the prediction target, without representation alignment or architectural changes. In pixel-space text-to-image generation, AnisoFlow improves image quality and, through better-conditioned velocity recovery, largely removes patch-seam artifacts under reduced-precision inference.

## Adversarial Decoys: Misdirecting Attention-Based Defenses in ViT

DevFeed: [Adversarial Decoys: Misdirecting Attention-Based Defenses in ViT](<https://devfeed.tech/articles/adversarial-decoys-misdirecting-attention-based-defenses-in-vit-131948.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.07922>)

Author: Giulia Marchiori Pietrosanti, Giulio Rossolini, Giorgio Buttazzo

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

Content type: article

Language: en

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

Topics: [Adversarial attacks](<https://devfeed.tech/topics/adversarial-attacks.md>), [Transformer](<https://devfeed.tech/topics/transformer.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [attention](<https://devfeed.tech/tags/attention.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [defense](<https://devfeed.tech/tags/defense.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [image](<https://devfeed.tech/tags/image.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [ranking](<https://devfeed.tech/tags/ranking.md>), [target](<https://devfeed.tech/tags/target.md>), [vision-transformers](<https://devfeed.tech/tags/vision-transformers.md>)

### AI overview

The paper presents adversarial decoys that exploit attention-ranking defenses for Vision Transformers. The original patch causes misclassification, while a separate optimized patch raises attention on selected target tokens, diverting the defense from the true adversarial region. A layer-wise objective ranks those targets above competing tokens, and the decoy optimization can be combined with different adversarial patch attacks. ImageNet experiments across multiple ViT architectures and attacks report that decoys redirect high attention scores while preserving much of attack effectiveness. The findings indicate that attention alone may not reliably identify adversarially relevant tokens.

### Source excerpt

arXiv:2607.07922v2 Announce Type: replace-cross Abstract: Vision Transformers (ViTs) remain vulnerable to localized adversarial attacks, e.g., adversarial patches, while recent test-time defenses mitigate them by suppressing image tokens with abnormally high attention scores. These defenses exploit a strong coupling between attention and adversarial effectiveness: adversarial tokens often need to attract substantial attention to influence the prediction. We introduce adversarial decoys, image patches optimized to redirect attention, and therefore related defenses, towards selected target tokens. Rather than jointly optimizing misclassifications and defense evasion, our approach decouples the two objectives: the original adversarial region induces the incorrect prediction, while a separate decoy manipulates the attention ranking used by the defense. A layer-wise objective increases target-token attention and promotes these tokens above competing non-target ones. Since the decoy optimization is not tied to a specific attack formulation, the method can be easily integrated with any existing adversarial patch attack. Experiments on ImageNet across multiple ViT architectures and adversarial patch attacks show that decoys consistently redirect high attention scores away from the true adversarial region while preserving much of the attack effectiveness against attention-ranking defenses. These results reveal a fundamental limitation of using attention alone as an indicator of adversarial relevance.

## Winfree Oscillatory Neural Network

DevFeed: [Winfree Oscillatory Neural Network](<https://devfeed.tech/articles/winfree-oscillatory-neural-network-131917.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.20922>)

Author: Jiawen Dai, Yue Song

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

Tags: [architecture](<https://devfeed.tech/tags/architecture.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>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [image-recognition](<https://devfeed.tech/tags/image-recognition.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [mappings](<https://devfeed.tech/tags/mappings.md>), [neural-network](<https://devfeed.tech/tags/neural-network.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [scalability](<https://devfeed.tech/tags/scalability.md>), [synchronization](<https://devfeed.tech/tags/synchronization.md>)

### AI overview

The Winfree Oscillatory Neural Network (WONN) uses generalized Winfree dynamics to evolve representations on a torus through structured oscillatory interactions. Its hierarchical interaction mechanisms can use fixed trigonometric mappings or learnable neural networks. The abstract reports competitive or superior performance with strong parameter efficiency across CIFAR, ImageNet, Maze-hard, and Sudoku; it describes WONN as the first synchronization-based oscillatory architecture to scale competitively to ImageNet-1K. On Maze-hard, it reports 80.1% accuracy using 1% of the parameters of prior state-of-the-art models. These findings suggest oscillatory dynamics may offer a scalable, parameter-efficient alternative, though the supplied abstract does not provide detailed methods or comparisons.

### Source excerpt

arXiv:2605.20922v2 Announce Type: replace-cross Abstract: Oscillations and synchronization are widely believed to play a fundamental role in representation and computation. However, existing machine learning approaches based on synchronization dynamics have largely been confined to specialized settings such as object discovery, with limited evidence of scalability to standard vision benchmarks or logic reasoning tasks. We propose the Winfree Oscillatory Neural Network (WONN), a dynamical neural architecture based on generalized Winfree dynamics. WONN evolves representations on the torus $(S^1)^d$ through structured oscillatory interactions, combining phase-based inductive biases with flexible and hierarchical interaction mechanisms instantiated as either fixed trigonometric mappings or learnable neural networks. We evaluate WONN on image recognition and complex reasoning tasks, including CIFAR, ImageNet, Maze-hard, and Sudoku. Across these domains, WONN achieves competitive or superior performance with strong parameter efficiency. In particular, WONN is, to our knowledge, the first synchronization-based oscillatory architecture to scale competitively to ImageNet-1K. Furthermore, on Maze-hard, WONN achieves 80.1% accuracy using only 1% of the parameters of prior state-of-the-art models. These results suggest that structured oscillatory dynamics provide a scalable and parameter-efficient alternative to conventional neural architectures.

## EbbFlow: PINN-Inspired Progressive Curriculum for MeanFlow Training

DevFeed: [EbbFlow: PINN-Inspired Progressive Curriculum for MeanFlow Training](<https://devfeed.tech/articles/ebbflow-pinn-inspired-progressive-curriculum-for-meanflow-training-125340.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11220>)

Author: Mengfan Yuan, Xunheng Kuang, Jianbin Chen, Lei He

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

Content type: article

Language: en

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

Topics: [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [inference](<https://devfeed.tech/tags/inference.md>), [results](<https://devfeed.tech/tags/results.md>), [sampling](<https://devfeed.tech/tags/sampling.md>), [schedule](<https://devfeed.tech/tags/schedule.md>), [supervision](<https://devfeed.tech/tags/supervision.md>), [temporal](<https://devfeed.tech/tags/temporal.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

EbbFlow changes MeanFlow's temporal sampling schedule while keeping its loss, target, backbone, and inference procedure unchanged. Its Temporal Supervision Allocation begins with diagonal supervision, then introduces interior intervals and expands their maximum length using left-anchored intervals to preserve time coverage. In matched class-conditional ImageNet 256x256 runs (DiT-B/2, one seed per method, 400k updates, no classifier-free guidance), EbbFlow reports FID 39.65 with one network evaluation and 36.02 with two-evaluation ODE sampling, compared with MeanFlow's 44.82 and 39.86. Ablations support the scheduling and interval choice, but a fixed full-horizon variant does better at one evaluation. The results suggest benefits depend on the inference protocol.

### Source excerpt

arXiv:2611.11220v1 Announce Type: new Abstract: MeanFlow learns interval-average velocity for few-step generation, but its fixed temporal sampler mixes diagonal and interior supervision throughout training. We interpret temporal coordinates as collocation points: diagonal samples provide a stochastic boundary target, whereas finite intervals supply a differential-consistency target. The diagonal fraction controls how often the consistency component appears, while interval length enters its gradient through an explicit factor. Motivated by this distinction, EbbFlow uses Temporal Supervision Allocation (TSA) to schedule the diagonal fraction and interval geometry. It starts with diagonal supervision, then admits interior samples and expands their maximum interval; a left-anchored interval law preserves coverage across time. The per-sample loss, target, backbone, and inference procedure remain unchanged. In matched class-conditional ImageNet $256\times 256$ runs without classifier-free guidance (DiT-B/2, one training seed per method, 400k updates), EbbFlow achieves 50k-sample FID of 39.65 at one network evaluation and 36.02 at two-evaluation ODE sampling, versus 44.82 and 39.86 for MeanFlow. Gradient probes show a mid-training difference in component geometry. Ablations support allocation scheduling and left-anchored intervals, although a fixed full-horizon variant performs better at one evaluation. These results suggest that temporal sampling is a useful training control whose benefits depend on the inference protocol.

## Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection

DevFeed: [Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection](<https://devfeed.tech/articles/vendi-anomaly-scores-for-efficient-and-accurate-anomaly-detection-125597.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.12871>)

Author: Amey Pasarkar, Adji Bousso Dieng

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

Content type: article

Language: en

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

Topics: [outlier-detection](<https://devfeed.tech/topics/outlier-detection.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [AI for science](<https://devfeed.tech/topics/ai-for-science.md>)

Tags: [anomaly-detection](<https://devfeed.tech/tags/anomaly-detection.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diversity](<https://devfeed.tech/tags/diversity.md>)

### AI overview

The Vendi Anomaly Score (VAS) flags samples by measuring how much their removal changes dataset diversity, as measured by the Vendi score. The authors say it captures local redundancy and global structure without density estimation or partitioning heuristics, adapts to the data's spectral structure without hyperparameter tuning, and is non-parametric with linear scaling in dataset size. Across 148 embedding benchmarks from 10 source datasets and 4 architectures, it reportedly achieves state-of-the-art anomaly-detection performance; large-scale ImageNet experiments report robustness as contamination and dataset scale increase.

### Source excerpt

arXiv:2611.12871v1 Announce Type: new Abstract: Anomaly detection (AD) is a critical for trustworthy machine learning and scientific discovery. Existing methods typically define anomalies through local density, isolation heuristics, or distance from a fitted distribution, making them sensitive to neighborhood scale, partitioning choices, or restrictive distributional assumptions. In this work, we introduce a new paradigm by formulating anomaly detection from a diversity perspective. We propose the $\textit{Vendi Anomaly Score}$ (VAS), which detects anomalies by quantifying how much a sample changes the diversity of the dataset, as measured by the Vendi score, when removed. VAS captures both local redundancy and global data structure without relying on density estimation or partitioning heuristics. Moreover, VAS requires no hyperparameter tuning, instead adapting automatically to the spectral structure of the data. VAS is non-parametric and scales linearly with dataset size. Across 148 benchmark embedding datasets derived from 10 source datasets and 4 embedding architectures, VAS achieves state-of-the-art AD performance. We further validate VAS on large-scale ImageNet experiments, where it remains robust as both contamination rate and dataset scale increase.

## Self-Supervised Collapse Prevention with Score-Based Entropy Maximization

DevFeed: [Self-Supervised Collapse Prevention with Score-Based Entropy Maximization](<https://devfeed.tech/articles/self-supervised-collapse-prevention-with-score-based-entropy-maximization-125370.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11379>)

Author: Adarsh Kappiyath, Silpa Vadakkeeveetil Sreelatha, Xiatian Zhu, Yann LeCun, Lu Yin, Amir Bar

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

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>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [diversity](<https://devfeed.tech/tags/diversity.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [information](<https://devfeed.tech/tags/information.md>), [learning](<https://devfeed.tech/tags/learning.md>), [linear](<https://devfeed.tech/tags/linear.md>), [lower-bound](<https://devfeed.tech/tags/lower-bound.md>), [model](<https://devfeed.tech/tags/model.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [score](<https://devfeed.tech/tags/score.md>), [training](<https://devfeed.tech/tags/training.md>), [view](<https://devfeed.tech/tags/view.md>)

### AI overview

EBReg uses a jointly learned encoder and latent score model to discourage representational collapse in self-supervised learning. It aligns views while maximizing smoothed marginal entropy, using score estimation without explicitly evaluating the latent density. The same score model can refine corrupted representations at test time toward the training latent distribution. The abstract reports 75.57% ImageNet-1K linear-probe accuracy with ViT-L and an approximately 5% average relative accuracy gain on ImageNet-C from score-guided test-time adaptation across diverse architectures.

### Source excerpt

arXiv:2611.11379v1 Announce Type: new Abstract: Diffusion models have demonstrated the effectiveness of score-based objectives for generative modeling, yet their potential for self-supervised representation learning remains underexplored. We introduce Entropy-Based Regularization (EBReg), a framework that uses latent score estimation to regularize the latent distribution during self-supervised training. EBReg jointly learns an encoder and a latent score model, and utilizes the learned score to provide a diversity-preserving signal that prevents representational collapse. Motivated by a mutual-information lower bound, we formulate the objective as view alignment with smoothed marginal entropy maximization and optimize the marginal entropy without explicit density evaluation. The same score model can additionally be used at test time to refine corrupted representations toward the training latent distribution. Empirically, EBReg learns competitive representations on ImageNet-1K, achieving 75.57% linear-probe accuracy with ViT-L. Across diverse architectures, score-guided test-time adaptation further improves robustness on ImageNet-C, yielding an average relative accuracy gain of approximately 5%.

## Training-Free Active Test-Time Adaptation for Vision-Language Models

DevFeed: [Training-Free Active Test-Time Adaptation for Vision-Language Models](<https://devfeed.tech/articles/training-free-active-test-time-adaptation-for-vision-language-models-124617.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09084>)

Author: Jihwan Bang, Sumyeong Ahn, Hwanjun Song, Jae-Gil Lee

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

Content type: article

Language: en

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

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [budget](<https://devfeed.tech/tags/budget.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [labeling](<https://devfeed.tech/tags/labeling.md>), [memory](<https://devfeed.tech/tags/memory.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [training](<https://devfeed.tech/tags/training.md>), [tta](<https://devfeed.tech/tags/tta.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

### AI overview

Active VLM-TTA lets a pretrained vision-language model spend a labeling budget on selected test samples, with each prediction made before that sample's label arrives. The queried labels are delayed evidence for future samples, not corrections for the current prediction. ACTOR is a training-free plug-in that routes uncertain samples among the base prediction, a label query, and memory-based correction, checking consistency and reliability against previously queried labels. The abstract reports improvements over strong VLM-TTA baselines across ten cross-domain benchmarks and an ImageNet-based domain-transfer benchmark, on ResNet and ViT backbones, with negligible inference overhead. It does not provide benchmark-level results or details of the routing rule.

### Source excerpt

arXiv:2611.09084v1 Announce Type: cross Abstract: Test-time adaptation (TTA) for vision-language models (VLMs) is usually studied in a fully unsupervised regime, where adaptation relies on pseudo-labels, confidence scores, or entropy computed by the model itself. This is a severe restriction on hard shifted samples: the model is asked to repair its own uncertain predictions without any external evidence. At the other extreme, supervised online adaptation assumes labels can directly train or correct the model. We study the intermediate deployment regime between these two extremes. In Active Test-time Adaptation for VLMs (Active VLM-TTA), a pretrained VLM may spend an explicit labeling budget on selected test samples, but the prediction for the current sample must be emitted before its label can be used; queried labels are therefore costed, delayed evidence for future samples rather than free correction for the present one. Under this future-only constraint, the central question is how to reuse sparse labels conservatively. We instantiate this setting with ACTive VLM-TTA for Online Retrieval (ACTOR), a training-free plug-in module that routes uncertain samples among base prediction, label query, and memory-based correction using a consistency-and-reliability check over previously queried labels. Across ten cross-domain benchmarks and a domain-transfer benchmark based on ImageNet, ACTOR consistently improves strong VLM-TTA baselines on ResNet and ViT backbones with negligible inference overhead. Our implementation is publicly available at https://github.com/kaist-dmlab/ACTOR.

## A Theoretical Bridge Between Long-Tailed Recognition and Continual Learning

DevFeed: [A Theoretical Bridge Between Long-Tailed Recognition and Continual Learning](<https://devfeed.tech/articles/a-theoretical-bridge-between-long-tailed-recognition-and-continual-learning-123054.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2306.13275>)

Author: Mahdiyar Molahasani, Michael Greenspan, Ali Etemad

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

Content type: article

Language: en

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

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [cifar-10](<https://devfeed.tech/topics/cifar-10.md>)

Tags: [cl](<https://devfeed.tech/tags/cl.md>), [clip](<https://devfeed.tech/tags/clip.md>), [continual-learning](<https://devfeed.tech/tags/continual-learning.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>), [distributions](<https://devfeed.tech/tags/distributions.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [framework](<https://devfeed.tech/tags/framework.md>), [performance](<https://devfeed.tech/tags/performance.md>), [recognition](<https://devfeed.tech/tags/recognition.md>), [standard](<https://devfeed.tech/tags/standard.md>), [strategies](<https://devfeed.tech/tags/strategies.md>), [theoretical](<https://devfeed.tech/tags/theoretical.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

The paper connects long-tailed recognition (LTR) to continual learning (CL): it shows that training on data with imbalance factor IF moves parameters into an O(1/√IF) neighborhood of the dominant-class solution, and reformulates LTR as sequential learning. It proves that the CL objective upper-bounds balanced LTR loss up to a constant set by residual Head loss. The proposed CLTR framework sequentially learns Head and Tail classes with standard CL methods to mitigate catastrophic forgetting. Experiments on CIFAR100-LT, CIFAR10-LT, ImageNet-LT, and Caltech256 support the theoretical predictions and report strong benchmark performance. With a CLIP backbone, CLTR using DualPrompt outperforms standard adaptation baselines and is competitive with specialized methods that use external semantic supervision.

### Source excerpt

arXiv:2306.13275v3 Announce Type: replace Abstract: We theoretically and empirically establish a previously unexplored connection between Long-Tailed Recognition (LTR) and Continual Learning (CL). Specifically, we show that training on a long-tailed dataset with imbalance factor $\mathrm{IF}$ drives model parameters into an $O(1/\sqrt{\mathrm{IF}})$ neighborhood of the dominant-class solution, under both uniform and exponentially decaying cardinality distributions. Building on this result, we establish an objective-level reformulation of LTR as a sequential learning problem and prove that the CL objective upper-bounds the balanced LTR loss up to a constant determined by the residual Head loss. Motivated by this connection, we introduce Continual Learning for Long-Tailed Recognition (CLTR), a principled framework that leverages standard off-the-shelf CL methods to sequentially learn Head and Tail classes while mitigating catastrophic forgetting. Extensive experiments on CIFAR100-LT, CIFAR10-LT, ImageNet-LT, and Caltech256 validate our theoretical predictions and demonstrate strong performance across diverse LTR benchmarks. In the foundation-model regime, CLTR with DualPrompt on a CLIP backbone outperforms standard adaptation baselines and is competitive with specialized methods using external semantic supervision. Our work bridges LTR and CL both theoretically and empirically, providing a principled approach for addressing long-tailed learning using standard CL strategies.

## Improving Context-Shift Robustness of Convolutional Networks via Context-Regularized Cross-Entropy

DevFeed: [Improving Context-Shift Robustness of Convolutional Networks via Context-Regularized Cross-Entropy](<https://devfeed.tech/articles/improving-context-shift-robustness-of-convolutional-networks-via-context-regularized-cross-entropy-120712.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.08157>)

Author: Jinen Setpal, Chaoyue Liu

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

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

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [context](<https://devfeed.tech/tags/context.md>), [convolutional-neural-networks](<https://devfeed.tech/tags/convolutional-neural-networks.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [feature](<https://devfeed.tech/tags/feature.md>), [image-classification](<https://devfeed.tech/tags/image-classification.md>), [robustness](<https://devfeed.tech/tags/robustness.md>)

### AI overview

The paper proposes a training framework to reduce convolutional networks' reliance on contextual cues that may shift between training and test data. It introduces ContrastiveCAMs, an attribution method intended to produce more faithful maps of model predictions, then uses those maps in Context-Regularized Cross-Entropy (CR-CE), which modifies cross-entropy to suppress contextual regions and emphasize object-relevant features. The abstract reports consistent improvements in context-shift robustness on Waterbirds, Spawrious, ImageNet/ImageNet-BG, and Hard-ImageNet; it provides no quantitative results or implementation details.

### Source excerpt

arXiv:2611.08157v1 Announce Type: cross Abstract: Image classification models often encounter objects in diverse and unpredictable contexts at test time, which may differ substantially from those seen during training. This phenomenon, termed context shift, exposes a key fragility in neural networks: when models rely on non-causal contextual cues for prediction, shifts in context can lead to significant performance degradation. In this paper, we propose a novel training framework to improve the robustness of convolutional neural networks under context shift. Our approach leverages spatial feature attribution to guide models toward making predictions that rely less on contextual regions and more on object-relevant features. As a first step, recognizing a theoretical limitation of existing feature attribution methods we introduce a new variant, ContrastiveCAMs, which produces more faithful attribution maps of model predictions. Building on ContrastiveCAMs, we further propose Context-Regularized Cross-Entropy (CR-CE), a modification of the standard cross-entropy loss that regularizes the model's attention by suppressing the influence of contextual regions, thereby improving robustness to context shift. We evaluate the effectiveness of our approach on several medium-to-large scale datasets (Waterbirds, Spawrious, ImageNet/ImageNet-BG, Hard-ImageNet), and report consistent improvements in context-shift robustness.

## Taming the Entropy Cliff: Variable Codebook Size Quantization for Autoregressive Visual Generation

DevFeed: [Taming the Entropy Cliff: Variable Codebook Size Quantization for Autoregressive Visual Generation](<https://devfeed.tech/articles/taming-the-entropy-cliff-variable-codebook-size-quantization-for-autoregressive-visual-generation-120018.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.06207>)

Author: Bowen Zheng, Weijian Luo, Guang Yang, Colin Zhang, Tianyang Hu

Published: 2026-11-06T05:00:00Z

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: [quantization](<https://devfeed.tech/topics/quantization.md>), [Image](<https://devfeed.tech/topics/image.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [scaling laws](<https://devfeed.tech/topics/scaling-laws.md>)

Tags: [ar](<https://devfeed.tech/tags/ar.md>), [autoregressive](<https://devfeed.tech/tags/autoregressive.md>), [budget](<https://devfeed.tech/tags/budget.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [conditional](<https://devfeed.tech/tags/conditional.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>), [distribution](<https://devfeed.tech/tags/distribution.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [features](<https://devfeed.tech/tags/features.md>), [generation](<https://devfeed.tech/tags/generation.md>), [information](<https://devfeed.tech/tags/information.md>), [language](<https://devfeed.tech/tags/language.md>), [linear](<https://devfeed.tech/tags/linear.md>), [next-token-prediction](<https://devfeed.tech/tags/next-token-prediction.md>), [quantization](<https://devfeed.tech/tags/quantization.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [size](<https://devfeed.tech/tags/size.md>), [tokenizers](<https://devfeed.tech/tags/tokenizers.md>), [training](<https://devfeed.tech/tags/training.md>), [transformer](<https://devfeed.tech/tags/transformer.md>), [variable](<https://devfeed.tech/tags/variable.md>)

### AI overview

Variable Codebook Size Quantization (VCQ) addresses the "Entropy Cliff" in discrete visual tokenizers: with a shared codebook of size 16,384, ImageNet's empirical conditional entropy reportedly becomes nearly deterministic after about 2 of 256 positions. VCQ instead increases codebook size monotonically from 2 to a maximum, while keeping the loss, parameter count, and autoregressive training procedure unchanged. Using a vanilla autoregressive Transformer and standard next-token prediction, VCQ-Base reports gFID without classifier-free guidance of 14.80 on ImageNet 256x256, versus 27.98 and 18.16 for constant codebooks of 16,384 and 8,192. A scaled model reports gFID 1.71 with 684M autoregressive parameters and no added techniques such as semantic regularization or causal alignment. A linear probe on the first 10 positions reaches 43.8% top-1 accuracy, versus 27.1% with uniform codebooks. The evidence provided is the abstract, so details beyond these reported results are not established here.

### Source excerpt

arXiv:2605.06207v2 Announce Type: replace-cross Abstract: Most discrete visual tokenizers rely on a default design: every position in the sequence shares the same codebook. Such a constant-codebook design can consume the finite-data entropy budget very quickly. We observe that the per-position empirical conditional entropy decays so quickly along the sequence that, after a few positions, the conditional distribution becomes essentially deterministic. On ImageNet with $K=16384$, this happens after only about 2 out of 256 positions, when most prefixes already identify individual images in the dataset. We call this phenomenon the Entropy Cliff. Under full conditional utilization, larger codebooks exhaust the finite-data entropy budget in fewer positions. The empirical entropy decreases more gradually in our language comparison, where the effective entropy per position stays below the codebook capacity. To address this, we propose Variable Codebook Size Quantization (VCQ), where the codebook size $K_t$ grows monotonically along the sequence from $K_{\min}=2$ to $K_{\max}$, leaving the loss function, parameter count, and AR training procedure unchanged. With a vanilla autoregressive Transformer and standard next-token prediction, VCQ-Base achieves gFID w/o CFG 14.80 on ImageNet $256\times256$, compared with 27.98 and 18.16 for constant codebooks of size 16384 and 8192, respectively. Scaled up, it reaches gFID 1.71 with 684M autoregressive parameters, without any extra training techniques such as semantic regularization or causal alignment. The extreme information bottleneck at $K_{\min}=2$ encourages semantic concentration in early encoder features: a linear probe on features from only the first 10 positions reaches 43.8% top-1 accuracy on ImageNet, compared to 27.1% for uniform codebooks. These results show that what matters is not only the total capacity of the codebook, but also how that capacity is distributed and organized.

## Are We Overthinking One-Step Generation?

DevFeed: [Are We Overthinking One-Step Generation?](<https://devfeed.tech/articles/are-we-overthinking-one-step-generation-120507.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.20331>)

Author: Woobin Park, Jaeyeon Kim, Yilun Du, Sitan Chen, Bohyung Han

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

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

Topics: [AI Models](<https://devfeed.tech/topics/ai-models.md>), [ai-image-generator](<https://devfeed.tech/topics/ai-image-generator.md>), [World models](<https://devfeed.tech/topics/world-models.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [compute](<https://devfeed.tech/tags/compute.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [denoising](<https://devfeed.tech/tags/denoising.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [feature](<https://devfeed.tech/tags/feature.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [flow](<https://devfeed.tech/tags/flow.md>), [flow-matching](<https://devfeed.tech/tags/flow-matching.md>), [frontier](<https://devfeed.tech/tags/frontier.md>), [generation](<https://devfeed.tech/tags/generation.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generators](<https://devfeed.tech/tags/generators.md>), [image-generation](<https://devfeed.tech/tags/image-generation.md>), [one](<https://devfeed.tech/tags/one.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [quality](<https://devfeed.tech/tags/quality.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>), [training](<https://devfeed.tech/tags/training.md>), [world](<https://devfeed.tech/tags/world.md>)

### AI overview

Priming Models train one-step generators in two stages: first on a reconstruction objective, such as pixel-space denoising at a fixed noise level, then with feature-space distributional post-training. The authors report that the reconstruction stage makes distributional training feasible and improves the training-compute versus generation-quality tradeoff across class-conditional image generation, text-to-image generation, and world modeling. On ImageNet at 256 x 256, they report FID 1.19, using about one-tenth the training compute of prior methods with comparable FID. These are claims reported in the supplied abstract.

### Source excerpt

arXiv:2610.20331v2 Announce Type: replace Abstract: Diffusion and flow matching are now dominant frameworks for generative modeling in continuous domains, but they typically require multiple network evaluations at inference. This has motivated extensive research on one-step generative modeling primarily built on diffusion and flow, though existing approaches often rely on complex training machinery. In this work, we introduce Priming Models, a simple two-stage framework for training one-step generators. Motivated by the difficulty of distributional training from scratch, Priming Models are first trained on a reconstruction objective, e.g., pixel-space denoising at a fixed noise level, before undergoing feature-space distributional post-training. Our central scientific finding is that this simple reconstruction stage is already enough to entirely alleviate the difficulty of distributional training. Empirically, across class-conditional image generation, text-to-image generation, and world modeling, Priming Models significantly advance the Pareto frontier of training compute vs. generation quality. On ImageNet at $256 \times 256$ resolution, we achieve a state-of-the-art FID of $1.19$ among one-step generators using approximately one-tenth the training compute of prior methods with comparable FID. We open-source our codebase at https://github.com/JaeyeonKim01/priming.

## MONET: A Massive, Open, Non-redundant and Enriched Text-to-image dataset

DevFeed: [MONET: A Massive, Open, Non-redundant and Enriched Text-to-image dataset](<https://devfeed.tech/articles/monet-a-massive-open-non-redundant-and-enriched-text-to-image-dataset-118224.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.21272>)

Author: Benjamin Aubin, Gonzalo I\\~naki Quintana, Onur Tasar, Sanjeev Sreetharan, Urszula Czerwinska, Damien Henry, Cl\\'ement Chadebec

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

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

Topics: [ImageNet](<https://devfeed.tech/topics/imagenet.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>)

Tags: [apache-2-0](<https://devfeed.tech/tags/apache-2-0.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [content](<https://devfeed.tech/tags/content.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [domain](<https://devfeed.tech/tags/domain.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [image](<https://devfeed.tech/tags/image.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large](<https://devfeed.tech/tags/large.md>), [latent-diffusion-model](<https://devfeed.tech/tags/latent-diffusion-model.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [monet](<https://devfeed.tech/tags/monet.md>), [open](<https://devfeed.tech/tags/open.md>), [pairs](<https://devfeed.tech/tags/pairs.md>), [text](<https://devfeed.tech/tags/text.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>)

### AI overview

MONET is an Apache 2.0 text-to-image dataset containing 103.8 million image-text pairs selected from 2.9 billion raw pairs across open sources. Its curation pipeline applies safety and domain filtering, removes exact and near duplicates, and uses multiple vision-language models to produce captions ranging from short to long-form; synthetic samples further augment the collection. Images include precomputed embeddings and annotations. The authors report that a 4-billion-parameter latent diffusion model trained exclusively on MONET achieved competitive GenEval and DPG scores. These results suggest the dataset can support large-scale, reproducible research; the supplied abstract does not give the scores or evaluation details.

### Source excerpt

arXiv:2605.21272v2 Announce Type: replace-cross Abstract: Training large text-to-image models requires high-quality, curated datasets with diverse content and detailed captions. Yet the cost and complexity of collecting, filtering, deduplicating, and re-captioning such corpora at scale hinder open and reproducible research in the field. We introduce MONET, an open Apache 2.0 dataset of 103.8M image-text pairs collected from 2.9B raw pairs across heterogeneous open sources through successive stages of safety filtering, domain-based filtering, exact and near-duplicate removal, and re-captioning with multiple vision-language models covering short to long-form descriptions, and further augmented with synthetically generated samples. Each image is shipped with pre-computed embeddings and annotations to accelerate downstream use. To validate the effectiveness of MONET, we train a 4B-parameter latent diffusion model exclusively on it and reach competitive GenEval and DPG scores, demonstrating that our dataset lowers the barrier to large-scale, reproducible text-to-image research.

[Next page](<https://devfeed.tech/topics/imagenet.md?cursor=WyIyMDI2LTExLTA1VDA1OjAwOjAwKzAwOjAwIiwgImVkYWE0MjM5LTQ0MzMtNDQ5Ny05YTdkLWE2OWJiYmFmNjQ2MyJd>)