Cross-Dataset Benchmark

Congenital heart disease (CHD) diagnosis and surgical planning often require patient-specific 3D anatomical models, but manual segmentation is labor-intensive, particularly in complex anatomies. Although deep-learning methods can automate this process, they are typically evaluated in-distribution, despite clinically relevant shifts in scanner, protocol, institution, population, and imaging modality. We present, to our knowledge, the first systematic evaluation of out-of-distribution (OOD) generalization in CHD segmentation, using ImageCHD as a held-out target cohort. We compare representative segmentation architectures under combined CT and CMR training, CT-only training, self-supervised pretraining, and limited target-domain adaptation. In-distribution performance proves to be a poor indicator of cross-cohort robustness: nnU-Net achieves the highest validation Dice (0.77) but falls to 0.51 on ImageCHD, while SwinUNETR generalizes substantially better, reaching 0.67 Dice. MAE and JEPA pretraining provide only modest additional benefit, suggesting that architecture contributes more to robustness than the tested pretraining strategies in this setting. When limited target-domain supervision is introduced, all SwinUNETR variants exceed 0.76 Dice with only 11 labeled ImageCHD cases. These findings demonstrate that conventional in-distribution evaluation can obscure clinically important generalization failures and support explicit cross-dataset testing as a key component of CHD segmentation evaluation.

Deep learning has led to numerous proposed architectures for Automated Emotion Recognition (AER) from electrocardiogram (ECG) data, but inconsistencies in preprocessing, training, and evaluation make direct comparisons difficult. Most studies train and validate models on individual datasets collected under homogeneous conditions, limiting variability and raising concerns about generalizability. Cross-dataset validation is sometimes used but primarily assesses model adaptability rather than true generalization. This study presents a comparative analysis of prominent deep learning architectures in AER, emphasizing model generalization over dataset adaptability. To enable this benchmark, we introduce two open-source frameworks: Affective Research on Representations and Classifications (ARRC), a standardized benchmarking toolkit, and Affective Research Dataset Toolkit (ARDT), a framework for inter-dataset training and validation. Using ARDT, we consolidate three publicly available AER datasets, CUADS, ASCERTAIN, and DREAMER, into a single dataset, increasing variability in sensor types, recording conditions, and participant demographics. We then use ARRC to evaluate three widely studied deep learning models and two CNN baselines through hyperparameter optimization and 10-fold cross-validation. Our findings provide insights into the trade-offs between classification accuracy and model complexity, establishing a reproducible benchmark for AER research. All source code for ARRC, ARDT, and model evaluation is publicly available to ensure transparency and facilitate further research.

Heart rate is a critical biomarker of health, and remote photoplethysmography (rPPG) enables its contactless estimation from video data for telemedicine applications. Recent advancements in deep learning based rPPG methods achieve state-of-the-art results, outperforming classical signal-processing methods in complex scenarios. However, deep learning methods depend on datasets with precise synchronization between videos and ground truth signals collected via contact sensors, whereas signal-processing-based methods do not. To address this dependence on labeled datasets, which are labor-intensive to collect, we investigate under which circumstances pseudo-labels extracted using unsupervised signal-processing methods can replace contact sensors labels for training deep learning methods. Our systematic evaluations found that for datasets with imperfect synchronization, the pseudo-label approach outperforms supervised training on contact sensors. For datasets with good synchronization, results are mixed: within-dataset evaluation shows no significant difference between training methods, while cross-dataset evaluation favors supervised training. However, removing a single outlier participant significantly improves the pseudo-label approach's cross-dataset performance, highlighting the importance of label quality. These results demonstrate that signal-processing methods can generate valid training signals for deep learning models, reducing dependency on labor-intensive dataset collection while maintaining competitive performance.

The growing realism and accessibility of manipulated and generated faces threaten the trustworthiness of digital media. To detect such forgeries, deepfake detectors based on vision foundation models have shown promising performance, but they typically rely on a single pretrained representation and are prone to overfitting to particular training distributions. To improve generalization to unseen forgeries, we propose UCF-Net, an uncertainty-aware cascaded fusion network that harnesses CLIP's language-aligned semantic priors and DINO's self-supervised visual-structure priors. UCF-Net extracts hierarchical features across Transformer depths, uses layer-wise expert aggregation to adaptively combine each encoder's multi-level cues, and performs weighted fusion of the resulting representations based on entropy-derived uncertainty. We further consolidate public deepfake datasets into a unified benchmark of approximately 4M images and construct a separate cross-generator evaluation set with over 8K face images from eight recent generators. On the unified benchmark, UCF-Net achieves the best mean AUC among the evaluated methods in both in-domain and cross-domain evaluations. On the cross-generator set, it adapts effectively with limited target-domain data, although zero-shot transfer remains challenging.

Background. Remote photoplethysmography estimates the cardiovascular pulse from facial video, and its explanations have rested on inspecting heatmaps rather than on quantitative evidence about where a model reads it. We quantified the explanations and asked whether such explanations transfer between datasets and track model performance. Method. We trained eight condition-specific RhythmFormer models on NCKU-rPPG, recorded under three illumination levels, speaking, rotation, and cycling, estimated one heart rate per 5.12-second clip, and set them beside a UBFC-rPPG reproduction. Raw attention, rollout, attention flow, and Beyond Intuition were assessed by skin coverage and the Salience-guided Faithfulness Coefficient (SaCo). Results. Beyond Intuition ranked highest on both datasets, at median coverage 0.789 and SaCo 0.837 on Static level 3 against 0.826 and 0.917 on UBFC-rPPG; lower ranks differed. Within one participant of one condition, neither measure was related to a clip's heart-rate error, waveform correlation, or signal-to-noise ratio on either dataset: 186 of the 252 coefficients fell below $|ρ|=0.10$ and 28 reached $p<0.05$ against the 13 expected by chance. Across the eight scenarios only Beyond Intuition's coverage followed the three performance measures, at $ρ=-0.43$, $+0.57$, and $+0.43$, while the attention-only methods' SaCo ran opposite to each. It failed at 40 lux alone, its median coverage falling to 0.180 and its median SaCo to $-0.178$, whereas motion degraded the estimates far more without such a drop. Conclusions. Skin coverage and SaCo carry information complementary to the performance measures rather than a proxy for them: attributing to the skin does not guarantee an accurate estimate. What an attribution reveals about a condition is where the model looks rather than how faithfully its map is ordered.

Brain tumor segmentation in magnetic resonance imaging (MRI) is a critical task for diagnosis and treatment planning. Despite the success of deep learning architectures such as U-Net and its variants, performance degradation across datasets remains a major challenge, particularly under domain shift and limited annotated data. To address this issue, this study systematically evaluates how individual MRI sequences influence model robustness across two well-known datasets. A ResUNet-based framework is employed, where each modality is trained independently to isolate its effect under a controlled cross-dataset evaluation protocol with tumor size stratification, without target-domain training, or with limited domain adaptation. Results show that the T2f/FLAIR sequence achieves the best cross-dataset performance, with Dice scores exceeding 75%. It consistently outperforms other modalities across most tumor size ranges, while multi-sequence training further improves performance. Additionally, even limited target-domain adaptation yields rapid initial gains, reducing the need for extensive annotations and costly retraining. Our source code is publicly available at https://github.com/henrique-zan/brain_tumor_segmentation/.

Modern image classification models excel when trained on single task-specific datasets but often struggle to generalize across domains and difficulty levels. We propose ARMDIL, an Adaptive Router for Multi-Domain Image classification with LLMs. ARMDIL is an ensemble that uses a multimodal large language model (MLLM) agent to dynamically route each image to the most suitable vision backbone. Our diverse ensemble employs convolutional neural networks (ResNets), self-supervised representation learners (SSL), and vision-language models (VLMs), each trained on a unified label space constructed from multiple image datasets with differing distributions and characteristics. Empirical evaluations illuminate the distinct capabilities and vulnerabilities of each architecture across disparate visual domains. Crucially, we show that ARMDIL effectively navigates these trade-offs, performing competitively with specialized training-based routers. Furthermore, it drastically improves adaptability by allowing new information to be integrated via simple prompt modifications, while enhancing interpretability through natural language reasoning traces. These advances in cross-dataset image classification pave the way for more reliable general-purpose vision systems such as AI assistants and autonomous robots.

Electroencephalography (EEG) decoding models often generalize poorly across datasets and subjects due to domain shifts in acquisition protocols and individual neurophysiology. We propose EEG-PRIME, a two-stage EEG foundation model for cross-dataset multi-task decoding. EEG-PRIME combines masked pretraining with prototype-aligned instruction tuning to enable instruction-aware and subject-invariant decoding across diverse BCI paradigms. During pretraining, an EEG encoder learns transferable representations through masked reconstruction with frequency-cutoff spectral augmentation. During instruction tuning, EEG-PRIME incorporates task-semantic, dataset-specific, and subject-invariant conditioning. The resulting conditioning signal modulates the Q-Former through Layer-wise Query Modulation, while frozen text embeddings of class labels serve as prototypes for cosine-similarity-based prediction across heterogeneous label spaces. Experiments on sixteen datasets covering motor imagery, emotion recognition, ADHD detection, covert speech, and mental workload show consistent improvements over state-of-the-art baselines and prior EEG foundation models under cross-subject settings. On two additional held-out datasets, EEG-PRIME achieves balanced accuracy comparable to within-session calibration models without target-domain optimization, calibration, or linear probing, demonstrating promising zero-shot transfer capability.

Robust comparisons of federated aggregation methods require joint consideration of predictive performance, threat definitions, metric semantics, and execution provenance. A 500-cell seed-1 evaluation matrix was reconstructed across five aggregation methods, five datasets, five architectures, and four recorded conditions: clean, sign-flipping, Gaussian, and BadNets. Successful execution logs were identified for 454 original runs and 36 repaired or rerun executions, whereas 10 clean SVHN cells were supported by summary-only provenance. Trimmed Mean achieved the highest clean macro-mean accuracy (76.02%) and the lowest mean within-task rank (1.70). Krum attained the highest recorded accuracy under both sign-flipping and Gaussian configurations. These relative rankings remained unchanged when analysis was restricted to 21 task pairs for which original successful logs were available for every method-condition combination. Audit of the supplied BadNets metric implementation established that every test input is triggered prior to target-label counting; consequently, the retained metric represents Triggered Target-Label Rate (TTLR) rather than a conventional target-excluding attack success rate. An audit of the supplied FedPARETO scaffold further identified a pathway in which predictive summaries may characterize an uncorrupted local model while the aggregation weight is applied to a separately corrupted update, introducing a potential discrepancy between reported predictive outcomes and the updates used for aggregation. The canonical matrix contains a single identified seed for each cell, and exact attack and configuration lineage is incomplete. Accordingly, the findings should be interpreted as descriptive comparisons within the recorded configurations and not as statistical estimates or universal claims regarding robustness.

Sign-language research for resource-constrained languages is often limited by the cost of dense linguistic labels such as glosses, temporal boundaries, and sign order. Broadcast news offers a practical alternative by pairing continuous signing with spoken-language transcripts, but this supervision is weak since text and signing are loosely aligned. Morphologically rich languages such as Turkish add further difficulty, as the same lexical meaning can appear in many inflected forms while some derived forms should remain distinct. We study whether weak transcript-based supervision can pretrain a reusable sign encoder in this setting, where poor text normalization can fragment pseudo-gloss targets and weaken representation learning. Unlike prior pseudo-gloss pipelines designed mainly to improve translation, we test whether the pretrained encoder transfers as a reusable representation for cross-dataset sign spotting. We pretrain on TSL-News, a new Turkish broadcast corpus, using pseudo-gloss labels derived from transcripts rather than manual annotation, comparing rule-based morphological lemmatization with constrained LLM-assisted normalization over a fixed vocabulary. We evaluate the learned representations via cross-dataset sign spotting on a new TSL Spotting Benchmark built from the TSL Dictionary corpus. The LLM-assisted encoder raises top-5 temporal localization mean IoU from 0.235 to 0.465, with 56.2% of examples reaching an IoU of at least 0.50; a frequency analysis suggests this gain is not mainly driven by memorizing frequent pseudo-gloss labels. In a downstream translation check, the same pretraining improves BLEU-4 from 9.60 to 11.04 and ROUGE from 23.48 to 27.43. These results show that loosely aligned broadcast data can provide effective weak supervision for learning sign representations that capture both lexical content and temporal structure.

Breast density classification is a critical component of breast cancer risk assessment, yet AI models often struggle to generalize across clinical sites due to vendor-specific acquisition styles. In this work, we introduce two new datasets, BreastMammo and DenseMammo, to facilitate robust multi-view mammography research. We propose a domain generalization framework that utilizes a foreground-only histogram matching protocol to resolve the domain shift issue arising from disparate clinical sources. Internal evaluation using a 5-fold cross-validation protocol demonstrates the efficacy of our approach, with the Swin Transformer backbone achieving a peak AUC of 98.32% for density classification. External evaluation on the TNMammo and LUMINA datasets demonstrates that the proposed approach consistently reduces domain shift, significantly outperforming prominent domain generalization paradigms, including MixStyle and Discrete-Fourier-Transform-based frameworks.

Face presentation attack detection (PAD) aims to reliably detect a wide range of presentation attacks. While PAD methods achieve strong performance within individual datasets, their performance degrades under cross-dataset evaluation. Variations in sensors or lighting conditions can reduce the effectiveness of detectors from near-perfect to nearly random. Foundation models (FMs) have emerged as a promising alternative because typical PAD datasets, such as the MCIO benchmarks (MSU-MFSD, CASIA-FASD, Replay-Attack, and OULU-NPU), are small relative to the scale used for web-based pretraining. However, existing PAD systems primarily focus on CLIP-based foundation models, while overlooking other FMs with different architectures and training procedures. This study addresses this question by systematically evaluating 32 FMs. Zero-shot prompting achieves performance near chance across model families and scales. The vision encoders, when low-rankadapted (LoRA) with fewer than 1% trainable weights, achieve below 2% intra-dataset ACER in most cases, while cross-dataset ACER is substantially higher. LoRA primarily refines the decision boundary within a dataset, suggesting that pretrained representations and the adaptation dataset play a larger role in cross-dataset generalization than the evaluated lightweight adaptation strategy.

Publicly available Arabic datasets for target-specific stance detection remain limited, particularly for evaluating cross-target generalization. This paper presents the Mawqif-XT, consisting of 996 manually annotated Arabic tweets collected from three public targets: Women Driving, E-Cars, and Trimester System. Each tweet is annotated with stance, sentiment, and sarcasm labels following the original Mawqif annotation scheme. The released extension is intended as a held-out evaluation set for assessing model generalization to both semantically related and previously unseen targets, while the original Mawqif dataset is used for training and development. In addition, we establish baseline results using several Arabic and multilingual transformer models, as well as zero-shot large language models (LLMs), to facilitate reproducible evaluation. Together with the original Mawqif dataset, the Mawqif-v2 Extension provides a benchmark for evaluating cross-target generalization in Arabic stance detection.

AI-generated image detectors generalize poorly when their training and test images originate from different generators or datasets. Despite the rich spatial representations produced by vision foundation models like DINO, existing detectors typically classify images using only the globally aggregated CLS token. We hypothesize that globally aggregating DINO features into a single CLS token obscures spatially distributed generation traces. To test this hypothesis, we introduce PatchHead, a lightweight spatial aggregation head that preserves the two-dimensional organization of DINO patch tokens and integrates evidence across neighboring regions. During training, we freeze the pretrained DINO backbone and optimize only the inserted LoRA adapters, PatchHead, and auxiliary projection head. Across nine cross-dataset benchmarks spanning manually curated and in-the-wild settings, PatchHead ranks first on seven datasets and second on the remaining two. It improves the strongest prior method from 91.6% to 94.6% in average balanced accuracy (+3.0 points) and raises the worst-case accuracy from 82.4% to 89.4% (+6.9 points), while introducing only 8.6% more trainable parameters and 0.08% additional FLOPs. Further qualitative analysis suggests that PatchHead (i) reduces class-conditional domain discrepancy, and (ii) redirects the representation from content-dominated saliency toward spatially distributed authenticity evidence. Together, these observations provide a representation-level account of why spatial patch aggregation transfers more reliably across generators and datasets than a single CLS-based global representation. Our code and models will be made available upon acceptance.

Public infant cry corpora are small, label-incompatible, and almost always evaluated one corpus at a time. We ask what this practice hides and what fixes it. Across four cry corpora screened by a multi-level leakage audit (byte-level and embedding-level deduplication plus a within-corpus train-test near-duplicate audit), we probe four frozen encoders and a handcrafted baseline under a unified five-class need ontology and shared task formulations. The audit exposes what single-corpus evaluation conceals: within-domain macro-F1 swings by 0.57-0.80 for the same encoder, cross-corpus transfer is negative on average (negative-transfer ratio 0.19-0.35, significant in 18 of 30 directed cells, BH-FDR), and 349 content-identical clip groups carry conflicting metadata labels across corpus distributions. The same audit, however, reveals a consistent way forward. Transfer into the noisiest corpus is consistently positive in effect size at matched training size and after near-duplicate removal, offering a practical recipe for small, noisy corpora. Frozen probes saturate at modest label budgets, while stabilized fine-tuning wins with full labels; domain-adaptive pretraining significantly beats stabilized fine-tuning at 5-10-shot (the 1-shot advantage is not robust to optimization-seed variance) but shows no significant advantage at 50-shot or beyond. In the tested binary, shared-label settings, ontology-mapped joint training wins in all four encoder-by-target combinations, whereas naively merging unmapped labels costs up to 37 F1 points. We release the ontology, mapping code, and audit pipeline, turning incompatible cry corpora into a usable joint-training resource.

Comparisons between AutoML systems at short time budgets -- tens of seconds rather than hours -- are common in tool READMEs and workshop papers, and they are easy to get wrong. We report a case study in which a simple AutoML engine, Orcetra, appeared to beat FLAML and AutoGluon on 513 OpenML datasets, winning 57.1% of them at a nominal 60-second budget and 78.4% of datasets against FLAML alone at 30 seconds. Both margins came from protocol defects that a results table cannot show. The search loop scored every candidate on the test split and reported the best, making the headline metric a maximum over dozens of noisy estimates while the baselines selected on training data and touched the test set once; and the budget was checked before launching a candidate but never enforced during one, so the system consumed a median of 120 s against a 60-second budget, 2.24x the wall-clock AutoGluon used. Re-running with selection moved to a validation split, the deadline enforced externally and every framework pinned to an equal share of the machine, Orcetra's win rate on the re-run subset falls from 59.4% to 34.3% and no pairwise difference against either competitor remains significant. Recording both estimands inside a single search lets us attribute the collapse: the selection rule accounts for 4.8 percentage points and unequal compute for most of the rest. The same traces give the selection bias as a function of budget, measured rather than assumed: it grows with $K$ but reaches only 0.27 accuracy points, about five times below the $σ\sqrt{2\ln K}$ bound a marginal-standard-error argument predicts, because candidates scored on shared test rows cancel most of the noise. We close with a checklist for short-budget comparisons. Code, per-dataset results and the scripts that regenerate every number and figure in the paper are released with it.

Functional near-infrared spectroscopy (fNIRS) is a promising modality for autism spectrum disorder (ASD) classification, yet existing approaches assume temporally aligned evaluation. In practice, the optimal observation window varies across subjects due to differences in hemodynamic delay and neurovascular coupling, creating a temporal distribution shift that degrades performance. We formalize this as a \textit{cross-time-window transfer problem}, introducing a protocol that varies window length (2.5--10,s) and offset within biological motion trials. Using topographic map representations of fNIRS recordings, we benchmark three vision architectures under two zero-shot baselines and eight adaptation strategies under leave-one-subject-out cross-validation ($N{=}124$). Key findings: (1) zero-shot cross-window accuracy is near chance (54--69%); (2) ${\approx}5\%$ subject-specific fine-tuning recovers 90--96%, while a subject-specific upper bound reaches 97--100%, identifying inter-subject variability as the dominant barrier; (3) domain-adversarial and self-supervised strategies achieve 78--90% without target-subject data; and (4) discriminative information is recoverable from windows as short as 2.5,s. These findings provide a practical roadmap for deploying fNIRS-based ASD classifiers under realistic temporal variability.

Numerous unsupervised domain adaptation (UDA) algori-thms exist, but for clinical practice, selecting the best-suited one along with proper hyperparameters often remains unclear, as the unlabeled deployment (target) domain prevents direct evaluation. We propose a label-free criterion that jointly selects the algorithm and hyperparameters for UDA. Given a pool of candidate models from multiple algorithms trained with different hyperparameters, our approach scores each candidate against an agreement reference, and selects the one with the highest score. The agreement reference is constructed in two levels without using target labels. First, we leverage multiple label-free selection signals, using each to nominate a model within every algorithm. Second, the nominated models are aggregated across algorithms to form a reference prediction for each unlabeled target sample. The candidate whose predictions agree most with this reference is then selected for deployment. Experimental results on four brain MRI and four chest X-ray datasets across seven clinically relevant transfer scenarios show that our method achieves better selection performance than other methods and remains effective across different algorithm pools. Our approach takes a step towards practical, label-free algorithm selection for clinical deployment of UDA.

Face presentation attack detection (PAD) remains challenging under cross-dataset evaluation, where domain shift degrades models trained on a single dataset. The scarcity of large-scale labeled data motivates adapting pretrained vision models rather than training task-specific architectures from scratch, raising a fundamental question: do general-purpose vision foundation models encode PAD-relevant information accessible with minimal task-specific training? To investigate, we systematically evaluate 24 frozen encoders, including self-supervised vision transformers, vision-language encoders, and supervised CNNs, using a unified linear-probing protocol on the MCIO benchmark (MSU-MFSD, CASIA-FASD, Replay-Attack, OULU-NPU). The backbone remains fixed, and only a lightweight linear head is trained to isolate the PAD information already present in the pretrained representation. Results show that frozen foundation-model representations can support strong intra-dataset PAD performance with only a linear classifier, but this performance does not reliably transfer across datasets. Model scale is beneficial within several families, although the effect is not monotonic and is strongly mediated by architecture and pretraining. InternViT-6B achieves the lowest mean intra-dataset error, whereas CLIP ViT-B/32 offers the most favorable cross-dataset transfer-compute trade-off among the evaluated probes. These findings suggest that while pretrained representations contain PAD-relevant information, explicit adaptation remains necessary to address domain shift.

Large-scale calcium imaging has created an opportunity to build foundation-style models for neural population dynamics, but a central question remains unresolved: \textbf{whether a model pretrained on one collection of recordings can generalize to new datasets, experimental paradigms, and even species.} Existing approaches are often designed for specific tasks and evaluated on a single dataset, making it unclear whether their learned representations are reusable for new calcium trace datasets. To tackle this gap, we present \textbf{CAPT}, a \textbf{C}ontinuous \textbf{A}utoregressive \textbf{P}opulation \textbf{T}ransformer for calcium population dynamics. CAPT models continuous calcium traces directly through a continuous patch tokenization strategy and is trained autoregressively, enabling end-to-end pretraining and adaptation to diverse downstream tasks. We first pretrain CAPT on a large-scale mouse calcium imaging dataset and evaluate its transferability across independent mouse, larval zebrafish, and \textit{C. elegans} datasets collected by different laboratories. In these transfer settings, the pretrained backbone is frozen and only adaptation modules are updated. Across neural population forecasting and behavior decoding tasks, CAPT consistently outperforms specialized and general-purpose baselines. Alongside predictive performance, multimodal analyses using NeuroPAL annotations in \textit{C. elegans} datasets show that CAPT embeddings form a shared functional space across datasets and capture anatomical cell-identity-related structure. These results suggest that the continuous autoregressive modeling opens up possibilities for a simple route towards general-purpose neural foundation models for calcium imaging, which can generalize across datasets, experimental paradigms, and species. Code is available at https://github.com/TSuXinH/CAPT.

Recent fake audio detection methods often leverage large speech models to achieve robust speech representations. These models are typically very deep, providing multiple layer-wise representations. However, current works often rely solely on single layer representation or feature fusion to extract one utterance-level representation for decision making. These methods risk underutilizing rich information from multiple layers and might induce feature collapse. We propose a novel layer-wise decision fusion method that applies fusion after per-layer decision making and achieves the best cross-dataset performance on In-the-Wild dataset (EER 6.90%) compared to other strong baselines. Our model design also makes the model more transparent, allowing us to conduct detailed analysis to reveal the underlying mechanism of decision making.

Cross-dataset generalisation remains a major barrier to clinical deployment of echocardiographic left ventricular segmentation, yet the sources of this shift are rarely disentangled. We examined whether transfer degradation could be estimated before deployment using handcrafted ultrasound descriptors, VAE latent features, and segmentation-derived latent features across six echocardiographic datasets. Geometry-aware preprocessing substantially improved several poor transfer cases, suggesting that much of the apparent domain shift reflects field-of-view and framing inconsistencies rather than intrinsic acoustic differences alone. Intensity z-normalisation changed dataset separability by less than 0.005, indicating that brightness and contrast are not the dominant shift axis. Absolute Dice drop on held-out source-target pairs was predicted with an R-squared value of 0.612, an MAE of 0.082, and a Spearman rho of 0.681. The variant without LV and fan-shaped features retained approximately 70% of this explanatory power, supporting mask-free transfer-risk monitoring. The most informative discrepancy measure depended on the representation, with CMD strongest in z-normalised handcrafted features, with an absolute r of approximately 0.86 and an R-squared value of approximately 0.70; log-Wasserstein strongest in VAE space, with an r of approximately -0.90 and an R-squared value of approximately 0.81; and log-MMD strongest in LV-segmentation latent features, with an r of approximately -0.92 and an R-squared value of approximately 0.84. Apparent vendor effects were largely dataset-confounded. Echocardiographic domain shift is therefore structured and measurable, and its impact on segmentation can be partly reduced through geometry-aware preprocessing and anticipated using representation-specific transfer-risk estimation.

Breast MRI is highly sensitive for detecting breast tumors, but exams contain many slices and require substantial reading time. Deep learning models often perform well on internal splits but can fail across institutions because of domain shift and dataset-origin bias. We study this failure mode for binary breast MRI tumor classification. EfficientNet-B3 and WaveViT-Small are trained using Duke Breast Cancer MRI and fastMRI, and evaluated only on the independent multi-center MAMA-MIA cohort. In a deliberately confounded setup, where label is perfectly correlated with dataset origin, external accuracy is near chance (0.5048--0.5265), despite very high recall. We then construct a mixed training set in which each class contains samples from both Duke and fastMRI, while preserving patient-level splitting, augmentation, and leakage controls. On MAMA-MIA, dataset mixing improves accuracy/F1 to 0.8463/0.8625 for WaveViT-Small and 0.8884/0.8994 for EfficientNet-B3. These results show that controlling dataset-origin bias is important for reliable breast MRI classification.

Crowdsourced labeling provides valuable labeled data for domains across natural language processing, computer vision, and video. Label aggregation aims to infer latent true labels from noisy and biased annotations, with the key lying in annotator reliability estimation. Despite promising progress, existing approaches struggle with one real-world bottleneck: most individual annotators label only a small subset of tasks, making accurate annotator estimation highly intractable. In this paper, we focus on the considerably more challenging multi-class label aggregation and propose AHEAD (cross-Annotator learning and High-confidEnce Annotator-guideD label aggregation), a cross-annotator learning framework that advances annotator reliability estimation by leveraging the population-level data. Specifically, AHEAD first learns high-dimensional cross-annotator contexts via a graph neural network, deriving multi-view, complementary annotator embeddings by aggregating individual-level annotator features with contextual information. These embeddings are then decoded into interpretable annotator-specific confusion matrices to fit the observed labels. We formulate a composite objective incorporating high-confidence annotators to alleviate the unsupervised training issues faced by prior models. Experiments on 10 real-world datasets spanning NLP, CV, Video, and Audio show that AHEAD substantially improves label accuracy, increasing average accuracy from 68.75% to 73.23%, with gains of up to 14.9% in the best case. Meanwhile, scalability experiments on the largest dataset further demonstrate the overall superiority of our method.

Understanding human emotions in spoken conversations is a key challenge in affective computing, with applications in empathetic AI, human computer interaction, and mental health monitoring. However, existing datasets vary in scale, emotion distribution, modality alignment, and data partitioning strategies, which can influence reliable cross-dataset generalization and minority-emotion modeling. We introduce SpEmoC a Speaking segment Emotion for Conversations comprising 306,544 raw clips from 3,100 English language movies and TV series. From these, 30,000 high quality, class balanced clips are curated, featuring synchronized visual, audio, and textual modalities annotated for seven emotions through a hybrid pipeline that integrates pretrained models with human validation. SpEmoC uses strict movie- and series-level splits to prevent content overlap between split sets, allowing more reliable evaluation of model generalization. The dataset also maintains a near-balanced distribution across seven emotions, including minority classes such as Fear and Disgust, which supports more balanced learning across categories. Extensive experiments, including in-domain benchmarking, cross-dataset transfer, low-data training, class-imbalance analysis, and modality transfer show that balanced data and careful splitting lead to more stable performance across emotions when models are evaluated on other datasets. These results highlight the importance of dataset design for robust and transferable multimodal emotion recognition.

Translating complex biomedical data into patient-friendly narratives is central to modern biomedical informatics. This study presents a comparative analysis of training small language models (SLMs) in specialized biomedical datato-text generation tasks. We explore widely adopted post-training methods including supervised fine-tuning (SFT), direct preference optimization (DPO), odds ratio preference optimization (ORPO), and group relative policy optimization (GRPO) with Qwen-based SLMs on a medicine package leaflets dataset. To assess cross-dataset generalizability, we also curated drug label data from openFDA. We evaluate models using both standard lexical overlap metrics like ROUGE as well as semantic similarity measures. Across our experiments, the results show that (1) the aligned SLMs outperform proprietary models like GPT-5; (2) ORPO outperforms the SFTbaselines; (3) GRPO yields the most robust cross-dataset performance among the alignment methods tested as well as GPT-5.

Introduce a MinHash-based similarity scheduling framework that constructs a progressive curriculum over neural architecture code for LLM-based neural architecture search (NAS). Using 128-permutation MinHash signatures over normalised 7-gram source code shingles, we partition the reference pool into similarity bands and present them in increasing architectural heterogeneity, with the best LoRA adapter from each stage merged cumulatively into the backbone. We evaluate the framework on OlympicCoder-7B within the LEMUR benchmark on CIFAR-10 image classification, generating N =15 candidate architectures per epoch across six progressive fine-tuning steps. The curriculum achieves 60% peak success rate at the high-similarity level without post-processing repair. A 2*2 ablation at the most diverse level curriculum versus base model, with versus without partial interface repair reveals that without repair the base model (47% peak SR) substantially outperforms the curriculum model (7% SR), while adding partial repair brings both to 53% SR. This pattern is consistent with merge-level weight drift progressively erasing evaluator-interface priors, and suggests that interface repair and curriculum scheduling target distinct failure modes. We further report a cross-dataset transfer observation on SVHN, where direct base-model generation without curriculum warmup yields 27% peak SR at substantially lower accuracy (60.5%) than the CIFAR-10 equivalent, consistent with the increased synthesis difficulty of the unq-family anchor architecture.

Foundation models are increasingly used as image feature extractors for mammography, but their robustness under external domain shift remains unclear. We benchmark 15 foundation-model backbones across breast density, BI-RADS severity, and cancer status using a unified frozen-backbone linear-probe protocol, training on 3 source datasets and evaluating on 12 task-compatible out-of-distribution (OOD) datasets after label harmonization. Mammography-specific vision-language models (Mammo-FM and MaMA) provide the strongest mean OOD performance, but robustness is not explained by mammography exposure alone. DINOv3 remains a competitive vision-only baseline, and mammography-adapted pretraining does not consistently improve generalization. Dataset-level analysis further shows that even leading models show heterogeneous performance across datasets. Feature-space inspection reveals that useful representations can preserve clinical signal while retaining dataset and acquisition structure. These findings highlight dataset-level OOD evaluation as a central criterion for assessing mammography representations. Our code is publicly available: https://github.com/biomedia-mira/mammo-ood.

Knowledge graph foundation models such as Ultra and Trix achieve strong inductive transfer by learning relation-graph representations that generalise to unseen entities and relations. Extending this transferability to temporal knowledge graphs (TKGs) remains challenging: existing temporal models tie their parameters to dataset-specific entities, relations, or timestamps and are not designed to transfer to TKGs with disjoint vocabularies. We propose GRATE (Gated Rotary Attention for Temporal Encoding), an entity-side message function that adds no learnable parameters and encodes time through relative time differences by rotating each edge message according to its time gap to the query and applying a query-conditioned gate to select temporally relevant signals. GRATE integrates into NBFNet-style KG foundation models while preserving structural transferability. Existing TKG benchmarks evaluate within shared train/test vocabularies and cannot directly test cross-dataset temporal transfer; we therefore construct GDELTIndT and WIKIIndT, inductive transfer benchmark suites with disjoint entities, relations, and timestamps spanning both interpolation and extrapolation. Across these benchmarks and held-out forecasting datasets, a single jointly pretrained GRATE checkpoint improves over the static base model in most settings.

We present MulTTiPop, a dataset of pop music segments and their associated multitrack MIDI recordings for the evaluation of automatic music transcription models. MulTTiPop contains 572 segments of popular music totaling 3.5 hours of audio, and contains songs from diverse genres and decades from the 1930s to 2000s. To collect this dataset, we perform metadata-based matching on song segments from the Lakh MIDI and TheoryTab datasets, manually identify an anchor beat between the audio and MIDI, then use beat tracking on the audio and warp the MIDI to match its tempo and timing. We evaluate state-of-the-art automatic music transcription models on MulTTiPop and find substantial room for improvement, with the best model achieving 38% Onset F1. More details and sound examples of MulTTiPop are available at https://gclef-cmu.org/multtipop.

Deep learning-based computer-aided diagnosis (CAD) systems have shown strong performance in breast cancer diagnosis, particularly for classification tasks in mammography. However, domain shifts across multi-site datasets remain a challenge, especially when models are applied to unseen domains. In this work, we proposed a calcification classification framework to improve malignant versus benign breast disease classification across multi-site mammography datasets. The framework consisted of two components: (1) an unsupervised domain adaptation module based on style transfer models (AdaIN and CycleGAN) to generate vendor-specific and technique-specific training samples without additional annotations, and (2) a supervised classification module using Swin Transformer V2 as the backbone. We evaluated the proposed method on three datasets: cross-validation on OPTIMAM (National Health Service, United Kingdom; n=2994), followed by external validation on EMBED (Emory University; n=125), and Duke Calcification Dataset v1 (n=788). These datasets cover multiple vendors and include both full-field digital mammography and synthetic 2D images derived from digital breast tomosynthesis. The proposed framework improved cross-site performance for both EMBED (AUC 0.68 to 0.72) and the Duke Calcification Dataset (AUC 0.68 to 0.73). These findings indicate that domain adaptation can reduce domain shifts and improve the generalization for calcification classification across multi-site datasets.

First-shot anomalous sound detection in DCASE Challenge Task 2 must flag anomalies of unseen machine types with a single threshold, without knowing whether a test clip comes from the data-rich source domain (990 normal training clips) or the data-scarce target domain (10). Two organizer-reported problems remain open: source- and target-domain AUC are negatively correlated across systems, and development-set performance does not predict evaluation-set performance. We address both with a training-free post-hoc layer over frozen audio embeddings: (i) per-domain quantile calibration shrunk toward a pooled map by a prior strength m, tracing a source/target balance frontier, and (ii) a label-free cross-validated domain-balance criterion that ranks candidate configurations from training normals only, paired with a coarse development-labeled viability veto. On DCASE 2025, the criterion rank-predicts the official evaluation score across a 45-configuration grid (Spearman rho = +0.91; family-block bootstrap 95% CI [+0.83, +0.95]) while development score is uninformative (+0.06). Criterion-based selection raises the evaluation score from 55.83 to 59.34 (jackknife CI [2.2, 4.8]) and, on an extended grid, to 61.05 -- retrospectively fourth of 35 teams. Replicating on DCASE 2023 and 2024 bounds the claim: development score is uninformative in all three years and degenerate configurations recur (vetoed every time), but under family-clustered uncertainty the criterion's predictive evidence survives only in 2025; in both replication years a fixed full-equalization default matches or beats criterion-based selection. A DCASE 2026 forward test is frozen before the 2026 evaluation ground truth is released; all headline numbers are reproduced by the official evaluator.

Accurate and efficient dataset valuation is essential for enabling fair and transparent data marketplaces, especially when multiple contributors provide data for training multi-task models. Most existing valuation methods, however, are limited to single-task settings, overlooking scenarios where a buyer aims to optimize performance across multiple downstream tasks. Moreover, traditional valuation approaches, such as Shapley-based or retraining-based methods, are computationally expensive and poorly suited for decentralized environments without a trusted central coordinator and with strict privacy constraints. We propose DMVM (Decentralized Multi-task Valuation via Model Merging), a novel framework that bypasses retraining and data sharing by leveraging task arithmetic to infer dataset contributions directly from model combinations. Instead of retraining or sharing raw data, DMVM quantifies how models trained on different datasets combine in parameter space to infer each dataset's marginal utility across multiple tasks. This formulation yields a valuation process that is scalable, computationally efficient, and explicitly aligned with multi-task generalization behavior. To support decentralized deployment, we introduce a secure aggregation protocol that enables collaborative valuation without revealing individual model parameters or private data. We also provide theoretical error bounds characterizing the approximation quality of DMVM and validate our framework through comprehensive experiments on computer vision and natural language processing tasks.

Deploying 3D point cloud analysis in privacy-sensitive, resource-constrained settings faces two barriers: data cannot be centralized, and models must run on limited edge hardware. We present a multi-seed benchmark jointly evaluating federated learning (FL) and knowledge distillation (KD) for 3D point cloud classification. It spans 13 FL algorithms and 10 KD objectives (a 130-pair cross-product) across 504 training runs, evaluated on ModelNet40 and a clinical craniosynostosis dataset. We report three findings. First, under extreme non-IID label skew, standalone FL degrades sharply: on ModelNet40, the strongest method reaches 76.32% against a 92.26% centralized reference; on clinical data, the best reaches 75.83% against 100%. Second, distillation successfully compresses the teacher into a student 74.51% smaller and roughly twice as fast at inference, often matching or surpassing the teacher. Third, the combined pipeline exposes an evaluation pitfall: when distillation keeps a hard-label cross-entropy term on a labeled proxy split, a collapsed federated teacher (8.50%) paired with Logit-MSE still yields a 92.94% student. This 84.4-point gap reflects the proxy labels rather than the federated model, reusing the very labels whose privacy motivated federation. Objectives without hard labels instead track teacher quality ($r \approx 0.99$) and collapse when the teacher does. We therefore recommend evaluating FL-KD pipelines with label-free distillation so reported accuracy reflects the federated teacher, not the proxy.

The growing availability of trajectory datasets has fueled major advances in data-driven motion prediction. Yet, models trained on one dataset often fail to generalize beyond their training domain as a result of differences in scene layouts, agent behaviors, and sensing conditions. A framework that learns latent representations of datasets and quantifies their similarity using distributional metrics is presented. This large-scale study covers 24 major datasets, including the most widely used motion-prediction benchmarks, and shows that the resulting transferability scores strongly correlate with cross-dataset model performance. The results provide practical guidance for dataset selection, pretraining, and large-scale foundation models for motion prediction, paving the way toward more generalizable and robust predictive systems.

The increasing connectivity of modern vehicles has made securing in-vehicle communication networks a critical challenge. Intrusion Detection Systems (IDS) have been widely studied as a defense mechanism for detecting malicious activities on the Controller Area Network (CAN) bus. However, the evaluation of CAN IDS methods remains difficult due to inconsistencies in experimental setups and the lack of standardized benchmarking frameworks. As a result, reported performance often depends on dataset-specific characteristics and may not reflect how detection methods behave in different environments. This work introduces a benchmarking framework for consistent evaluation of CAN IDSs across multiple datasets. Using the proposed framework, we integrate seven publicly available CAN IDS datasets collected under different experimental conditions and perform cross-dataset evaluation of five conceptually different IDS approaches. Our results highlight how detection performance can vary significantly across datasets, demonstrating the importance of cross-dataset benchmarking for assessing the robustness and generalization capabilities of CAN IDS methods.

Automated "suspicious behavior" flagging is a headline promise of AI surveillance, and the field reports high frame-level ROC-AUC on standard video anomaly detection benchmarks. Those numbers are measured by training and testing on the same camera and scene. We audit what happens when that assumption is dropped. We build an unsupervised normality model from the all-normal training frames of one dataset, using frozen off-the-shelf embeddings (CLIP, DINOv2, ResNet-50, EfficientNet-B0) and a nearest-neighbour distance, and score the test frames of the same and of other datasets. Across 4 real datasets (UCSD Ped1, UCSD Ped2, CUHK Avenue, ShanghaiTech) and 4 backbones, same-dataset AUC averages 0.704 but cross-dataset AUC averages 0.499, which is chance: a detector calibrated on one scene is no better than a coin flip on another, and in several pairs it is below chance. The strongest backbone makes this worse, not better: DINOv2 has the best same-dataset AUC (up to 0.901 on Ped2) and the largest cross-dataset drop. The collapse is not an artefact of the scoring rule: replacing the nearest-neighbour detector with a PaDiM-style Mahalanobis detector reproduces it almost exactly (cross-dataset gap 0.202 versus 0.208). Even at a favourable operating point the false-alarm rate is on the order of 31,931 per hour. We conclude that the benchmark numbers quoted for surveillance anomaly detection describe a calibrated laboratory setting and overstate deployable reliability by a wide margin, and we release the code that reproduces every number.

Deep learning models for diabetic foot ulcer (DFU) segmentation routinely report high accuracy, but they are almost always trained and tested on the same dataset, leaving their behaviour on data from a different clinical source largely unmeasured. We benchmark three representative segmentation architectures -- U-Net and DeepLabV3+ (convolutional) and SegFormer-B2 (Transformer) -- under an identical, leakage-screened protocol: training on the combined FUSeg/AZH wound data and evaluating, without fine-tuning, on two independent external datasets (DFUC2022 and Medetec). All models achieve strong in-domain performance (Dice 0.80--0.83) but degrade substantially across datasets. The degradation is, however, architecture-dependent: SegFormer-B2 generalizes best on both external sets (DFUC2022 Dice 0.557, Medetec Dice 0.786), outperforming both convolutional models, while the more complex DeepLabV3+ generalizes worse than the simpler U-Net. Per-image failure analysis on 2,160 images across both external test sets confirms that SegFormer-B2 produces the fewest catastrophic failures on DFUC2022 (31.1%), compared with U-Net (38.5%) and DeepLabV3+ (43.0%). The consistent ranking across two independent external sources, confirmed by Wilcoxon signed-rank tests (p < 0.001 on both datasets), indicates that architecture family, not model complexity, drives cross-hospital generalization.

Data scarcity is a major bottleneck in medical Multiple Instance Learning (MIL), especially for rare diseases or expensive modalities. We introduce a statistically grounded patient augmentation approach that generates realistic patients directly in embedding space. Using Gaussian Mixture Models as a probabilistic clustering approach on pooled instance embeddings from all patients, our method learns disease-specific "recipes"-statistical distributions of instances across unsupervised clusters. New patients are then generated by sampling embeddings from clusters based on learned recipes. Unlike existing methods that require examples from all categories, our method can generate patients offline by re-mixing pooled embeddings. Generated patients are further selected based on uncertainty quantification to improve MIL performance. We evaluate our method across three clinically relevant scarcity scenarios: (i) cross-dataset transfer, where an entirely missing "healthy" class is generated using statistics from an external cohort; (ii) low-data regimes, where class sizes are extremely limited; and (iii) small-cohort non-image tasks, including single-cell RNA-seq and flow cytometry. Across all experiments, our method improves performance over baseline, often outperforming other bag-mixing strategies. Notably, in the missing-class scenario, a performance comparable to full-dataset training is achieved, demonstrating its potential for rare disease diagnostic and privacy-preserving patient augmentation. The code is available at https://github.com/marrlab/RECIPE

Foundation models trained on biased datasets often rely on spurious correlations between target labels and non-causal attributes, resulting in poor generalization on minority groups. Bias mitigation remains challenging due to two fundamental issues. First, when group labels are unavailable, existing group-unsupervised methods typically infer spurious attributes implicitly from model behavior, making it difficult to identify spurious factors that are semantically aligned with real-world biases. Second, even with pseudo spurious supervision, most existing debiasing methods follow a single-branch design that operates within a single shared feature space, where target and spurious attributes are intrinsically entangled. To address the first challenge, we introduce Confidence-guided Bias Concept Mining (CBCM), which leverages diffusion-disentangled, semantically grounded concept representations to identify reliable spurious attributes without attribute annotations. To address the second challenge, we propose Dual-branch Cross-projection Debiasing (DCD), a prompt-tuning framework that separates target and spurious representations into two branches and explicitly removes spurious information through cross null-space projection while preserving target-relevant semantics. Extensive experiments on four benchmark datasets show that our method achieves state-of-the-art worst group accuracy among group-unsupervised approaches, while tuning at most 0.22% of the model parameters. The source code is available in the supplementary materials.

A Prior-data fitted Network learns the posterior predictive induced by its training prior; bringing this paradigm to multivariate time-series classification therefore calls for a synthetic generator that produces complete labelled datasets with temporal structure. We introduce a causal prior that synthesizes each dataset from a randomly sampled DAG over typed nodes across two modalities (tabular attributes and time series), natively producing multivariate, multi-class TSC datasets with cross-modal causal structure across channels, timesteps and labels, a regime not addressed by existing synthetic priors. To validate the prior, we finetune TabPFN v2.5 with minimal adaptations and evaluate on 75 UCR/UEA datasets within TabPFN's operating regime. Finetuning on our generator significantly outperforms both the unmodified upstream model and a tabular-only ablation of the same prior (Wilcoxon signed-rank $p=3.0\times 10^{-8}$ on ROC-AUC), isolating the contribution of the cross-modal temporal structure.

Continual Test-Time Adaptation (CTTA) aims to maintain model performance under evolving target domains by adapting online without labeled data. However, practical deployments often cannot retain the source dataset due to privacy or licensing constraints, and purely source-free CTTA methods tend to become unstable under long-term distribution shift, suffering from compounding self-training errors and catastrophic forgetting. We introduce DO-ALL (Distill Once, Adapt Life-Long), a plug-and-play framework that revisits source information in a compact and privacy-conscious form via Dataset Distillation (DD). Before deployment, DO-ALL performs DD to produce a small set of synthetic distilled anchors that summarize the source distribution. During adaptation, each target sample is matched with its most semantically aligned anchor, which provides a stable reference for various CTTA via source replay, representation alignment, and manifold-smoothing regularization. DO-ALL can be seamlessly integrated into existing CTTA algorithms, consistently improving long-term robustness across CIFAR100-C, ImageNet-C, and the CCC benchmark. This demonstrates the potential of leveraging DD to enable stable and continuous adaptation without retaining raw source data. The code is available at https://github.com/blue-531/DOALL.

The challenge associated with recognizing dysarthric speech primarily arises from pronounced acoustic variability attributed to impaired articulatory precision. Past research has demonstrated improved recognition through the use of hybrid DNN/HMM sequence discriminative training. This paper presents a comprehensive investigation of various combinations of acoustic features tailored to different Acoustic Models, offering suitable feature selections for each. The incorporation of Pitch features notably improved recognition performance, especially for sentence recognition tasks involving dysarthric speech. Through a systematic examination of the TORGO database, we have demonstrated the potential to enhance the performance of the state-of-the-art Factorized Time Delay Neural Network (F-TDNN) model for recognizing dysarthric speech. Our methods, implemented with the F-TDNN model, resulted in a 4.65% relative improvement in isolated word recognition and a 4.63% relative improvement in sentence recognition for dysarthric speech, compared to previous research. This improvement effectively compensates for speech variability, attributable to our deliberate selection of the number of overlapping frames between consecutive training example chunks.

Audio deepfakes generated by neural text-to-speech and voice-cloning systems threaten speaker verification and public discourse at scale. The core challenge is cross-dataset generalization: detectors trained on one synthesis pipeline collapse on unseen forgeries. We argue that this failure is primarily because of structural synthetic speech artifacts which are multi-timescale trajectory anomalies. Though every existing detector aggregates a fixed-window frame statistics, this misaligns the architecture with the signal. We propose FlowFake, a Liquid Time-Constant (LTC) architecture whose hidden state evolves via a learned ODE, with per-neuron adaptive time constants simultaneously resolving spectral (10ms) and prosodic (2s) cues. At only 34K parameters FlowFake achieves formal BIBO stability and O(dt^4) integration error. On a four-dataset cross domain benchmark (ASVspoof2019-LA, FakeOrReal, InTheWild, MLAAD), FlowFake reaches 75.29% on ASVspoof2019 trained only on FakeOrReal and 79.97% trained only on MLAAD. It outperforms RawGAT-ST and Whisper-DF on every evaluated pair and matching SSL Wav2vec2 (300x larger) at 0.01% of its parameter count. The source code is available on : https://github.com/GhostRider2023/FlowFake

Training-free anomalous sound detection (ASD) scores a test clip against a memory bank of normal embeddings from a frozen pretrained audio encoder. Recent work attributes domain-shift robustness mainly to how frame-level features are pooled over time; the scoring backend applied on top of the pooled embedding has received far less systematic attention. Using a single frozen BEATs encoder on the DCASE 2023 Task 2 development set (all seven machine types), we cross four classical backends -- nearest-neighbor cosine distance, Mahalanobis distance, locally density-normalized kNN, and PCA-subspace reconstruction residual -- with three temporal poolings (mean, GeM, max). Switching the backend moves target-domain AUC by 13.8 points on average (up to 53.8), whereas switching the pooling moves it by only 3.2 points: in this training-free regime, the backend, not the pooling, dominates domain-shift robustness. No backend wins everywhere, but the machine-dependent pattern reproduces on the DCASE 2025 development data (fan, bearing). Exploiting this, we propose a label-free score fusion that z-normalizes each backend with its training-bank self-scores and takes the minimum; it reaches a harmonic-mean target AUC of 63.3% versus 64.4% for the per-machine oracle, surpassing every fixed single backend while preserving source-domain accuracy. We also report a negative result: selecting a backend by source-domain pseudo-validation with proxy outliers fails, because all backends saturate on the proxy task.

Recent advances in generative AI, such as diffusion models and face-swapping tools, have enabled the creation of highly realistic deepfakes, leading to real-world harms including financial fraud and non-consensual explicit content. In response, deepfake detection has become an active research area, with recent methods increasingly focusing on improving generalization to unseen manipulations. This is typically evaluated using the Area Under the ROC Curve (AUC) measured separately across multiple datasets. However, such an evaluation fails to reflect real-world scenarios where detectors face a mixture of data sources and varying artifact types. To address this limitation, we introduce a novel metric, Cross-dataset AUC (Cross-AUC) that averages per-domain AUCs with a measure of prediction polarization for taking into account the robustness to domain shift. The polarization extent is quantified by the Wasserstein Distance between class score distributions. Cross-AUC not only assesses the generalization capabilities of deepfake detectors under domain shifts more realistically, but it is also interpretable as it better explains the reason behind a drop in performance. Experiments performed on seven benchmark datasets demonstrate its practical relevance.

Dataset distillation (DD) has emerged as a prominent approach in data centric machine learning, aiming to synthesize compact training sets for efficient training by compressing the information in large datasets into a small number of synthetic samples. However, DD methods are often evaluated under inconsistent evaluation protocols, ranging from standard ERM to single/multi-teacher supervision, making it difficult to isolate the effectiveness of distilled data from evaluation. Moreover, many prior methods claim that DD outperforms data pruning approaches such as coreset selection (CS), based on the assumption that restricting condensed datasets to subsets of real samples fundamentally limits their expressiveness. In this work, we critically evaluate DD methods through large-scale experiments using standardized datasets and evaluation protocols to assess their intrinsic effectiveness. We benchmark seven state-of-the-art (SOTA) DD methods on ImageNet-1K, ImageNet100, and ImageNette, using three widely adopted training protocols against three CS strategies. Our results show that while some DD methods fail to outperform even simple random subsets, the SOTA DD approaches are comparable to or worse than coresets on large-scale datasets and incur a substantially higher cost for construction. Beyond accuracy, we also evaluate the representativeness, diversity, and quality of condensed sets, and find that coresets consistently achieve better coverage of the original data distribution. These findings highlight the limited practical advantages of current DD methods and show that coresets remain competitive and are often a more computationally efficient alternative for data-centric learning.

Dynamic data pruning techniques aim to reduce computational cost while minimizing information loss by periodically selecting representative subsets of input data during model training. However, existing methods often struggle to maintain strong worst-group accuracy, particularly at high pruning rates, across balanced and imbalanced datasets. To address this challenge, we propose RCAP, a Robust, Class-Aware, Probabilistic dynamic dataset pruning algorithm for classification tasks. RCAP applies a closed-form solution to estimate the fraction of samples to be included in the training subset for each individual class. This fraction is adaptively adjusted in every epoch using class-wise aggregated loss. Thereafter, it employs an adaptive sampling strategy that prioritizes samples having high loss for populating the class-wise subsets. We evaluate RCAP on six diverse datasets ranging from class-balanced to highly imbalanced using five distinct models across three training paradigms: training from scratch, transfer learning, and fine-tuning. Our approach consistently outperforms state-of-the-art dataset pruning methods, achieving superior worst-group accuracy at all pruning rates. Remarkably, with only $10\%$ data, RCAP delivers $>1\%$ improvement in performance on class-imbalanced datasets compared to full data training while providing an average $8.69\times$ speedup. The code can be accessed at https://github.com/atif-hassan/RCAP-dynamic-dataset-pruning

The scale of speech anti-spoofing datasets has grown exponentially over the past decade, driven by the assumption that larger data leads to better performance. However, it remains unclear whether indiscriminate scaling commensurately improves model generalization. This study challenges the "scale-first" paradigm by decoupling the impacts of training data scale versus diversity. Through experiments on representative datasets, we report two key findings: (1) Larger is not always better. Expanding data scale excessively under fixed generation methods yields negligible returns and may even degrade cross-domain generalization due to overfitting.(2) Diversity outweighs scale. A smaller composite training set featuring diverse attacks significantly outperforms larger-scale datasets with limited diversity in cross-dataset evaluations. We conclude that future dataset construction should prioritize the diversity of generation methods over scale to effectively enhance model generalization.

Cross-domain shifts challenge Presentation Attack Detection (PAD) on ID Cards, given the restricted data available due to privacy concerns. This work proposes a compact multimodal model, based on new generative and discriminative blocks, which combines visual and textual data for PAD on genuine and synthetic ID images. While multimodal models exhibit strong generalisation after supervised fine-tuning, they fail in zero-shot settings. Our findings underscore that model capacity and real-world data are essential for reliable PAD, while existing synthetic datasets may not reflect real-world challenges. We argue for a re-evaluation of synthetic data as a benchmark and emphasise the need for more realistic, diverse datasets to advance PAD research.

Autonomous driving has shifted from modular perception-prediction-planning stacks toward end-to-end (E2E) models that map sensor inputs directly to vehicle control, often regularized by auxiliary tasks such as 3D detection, motion forecasting, and HD-map perception. Progress is driven by a fast-growing ecosystem of sensor-rich driving datasets, yet each ships its own file formats, APIs, coordinate conventions, and modality coverage, leaving cross-dataset experimentation and even basic per-dataset preprocessing to be re-implemented per project. We present StandardE2E, a framework that provides a single unified interface over E2E driving datasets. StandardE2E (i) standardizes per-dataset preprocessing under one shared data schema; (ii) combines multiple datasets in a single PyTorch DataLoader for cross-dataset pretraining, auxiliary-task supervision, and scenario-level filtering; and (iii) reduces adding a new dataset to a single per-dataset mapping from raw frames to the canonical schema, leaving the entire downstream pipeline unchanged. The framework supports six datasets out of the box: Waymo End-to-End, Waymo Perception, Argoverse 2 Sensor, Argoverse 2 LiDAR, NAVSIM (OpenScene-v1.1), and WayveScenes101, and is released as the open-source standard-e2e Python package, available at https://github.com/stepankonev/StandardE2E.

Models for AI-based skin cancer screening suffer a severe performance drop when shifting from expert dermoscopic (source) images to consumer-grade clinical (target) images, hindering real-world deployment. Existing domain adaptation methods often ignore crucial semantic invariants, such as clinical concepts. While new foundation models like MONET can provide this semantic information as dense, probabilistic scores, this metadata is unavailable at test time, creating a deployment paradox for practical image-only screening tools. We address this gap by proposing CoFiDA-M, a privileged information framework that learns from concepts at training time but deploys as an image-only model. Our method trains a teacher network that uses MONET concept probabilities to guide a FiLM modulator, transforming visual features into a semantically edited" feature space. A lightweight, image-only student is then trained to reproduce this edited representation, not just the teacher's final predictions. This distillation bakes" the clinical reasoning into the student's weights. On a challenging multi-dataset benchmark, our image-only student significantly outperforms state-of-the-art approaches, especially in melanoma recall. Our work provides a practical and generalizable framework for leveraging noisy, probabilistic metadata as privileged information, demonstrating strong cross-dataset robustness and potential for real-world deployment beyond dermatology. Implementation code is available at: https://github.com/mmu-dermatology-research/CoFiDA.git

The Artificial Intelligence (AI) life cycle requires a thorough understanding of the underlying data dynamics for robust, safe and cost-effective AI development and use. Dataset shifts are defined as changes between train and test data distributions. Whether occurring over time (temporal) or across different sites (multi-source), they can severely degrade model performance and compromise data quality. This is particularly important in health AI, where the safety and fundamental rights of patients can be severely affected by uncontrolled shifts both at training and operational stages. While the theoretical foundations of covariate, prior, and concept shifts are well established, there is a lack of accessible and comprehensive software tools to perform their analysis. We introduce dashi, an open-source Python library designed for the exploration, quantification, and characterization of dataset shifts. dashi provides a dual approach: an unsupervised approach that leverages information geometry and non-parametric statistical manifolds to data variability characterization and analysis (e.g., Information Geometric Temporal plots and Multi-Source Variability metrics like Global Probabilistic Deviation and Source Probabilistic Outlyingness), and a supervised approach that quantifies and characterizes model performance degradation. Both unsupervised and supervised approaches work across user-defined temporal and domain/source batches. We demonstrate the utility of dashi on three simulated and real-world health AI case studies on gestational diabetes mellitus, COVID-19 and emergency medical dispatch. By providing interactive visual analytics and variability metrics, dashi supports trustworthiness of AI life cycle stages enabling robust and safe machine learning pipelines through the assessment of data coherence and AI performance.

Can a dataset be recognized from the spurious correlations it induces during training? We argue that datasets leave dataset-specific traces in a model's learned semantic correlation structure: incidental regularities that are predictive within a dataset, but not causal for the underlying task, can be internalized during training. We use this insight to study dataset-level membership inference, moving beyond existing methods that rely on behavioral or distributional evidence such as confidence scores, losses, margins, generated samples, or query responses. We introduce a white-box semantic fingerprinting approach based on semantic correlation descriptors (SCDs), which capture the semantic correlation structure learned by a model and make it comparable across dataset mixtures. In a controlled leave-one-dataset-out diagnostic, SCDs recover dataset-specific changes and perfectly separate matching from non-matching dataset pairs. We then propose a practical SCD-based membership score that tests whether a target dataset is part of a model's training mixture using only the model's SCD and the target dataset's standalone SCD, without requiring leave-one-dataset-out models. Across three diverse experimental settings, with dataset groups for natural language inference, emotion classification, and medical text classification, we test both the advantages and limitations of SCD-based membership inference with different degrees of semantic separation and keyword support between dataset splits. On average, the classifier based on this score achieves the highest performance and the lowest std, outperforming black-box baselines RMIA, Attack-P, and LiRA, as well as the white-box SIF baseline. These results show that dataset membership can be traced through internal semantic correlations, with the largest relative gain exceeding 60% in ROC-AUC when dataset groups expose distinct semantic particularities.

AI-driven respiratory sound classification (RSC) is promising for automated pulmonary disease detection, yet multi-site deployment is hindered by inter-stethoscope variability. We introduce a federated domain generalization (FedDG) formulation for RSC under stethoscope-induced device shifts, where clients use heterogeneous devices and the model is evaluated on unseen devices. Our empirical analysis shows that stethoscope-induced style and disease-specific content are tightly entangled, making deterministic style removal unreliable. In response, we propose a causality-inspired multimodal FedDG framework that combines: (i) a causality-inspired device style intervention network that performs content-preserving style perturbations, (ii) counterfactual text augmentation that neutralizes metadata shortcuts, and (iii) gradient alignment that facilitates device-invariant representations across clients. Built on a multimodal language-audio pretraining model, it outperforms conventional data augmentation and federated learning baselines in leave-one-device-out validation on ICBHI and SPRSound datasets. Code will be released upon publication.

Ultrasound is widely used in obstetric care due to its safety, accessibility, and real-time imaging. However, interpretation remains operator-dependent and susceptible to noise and artifacts. Deep learning models have shown strong performance to solve these problem, but they typically require large annotated datasets that are difficult to obtain in clinical ultrasound. Foundation models (FMs) offer an alternative, using a large number of ultrasound images to learn transferable representations that can generalize with limited labeled data. This work presents a comprehensive benchmark of ultrasound-specific FMs for fetal plane classification. We evaluated four ultrasound FMs (USFM, MOFO, UltraSAM, FetalCLIP) against two CNN baselines (ResNet50, EfficientNet-V2) and a ViT (DINOv3) pretrained on natural images. We trained all models under two complementary settings: full fine-tuning and linear probing with a frozen encoder. All models were trained using 5-fold patient-level cross-validation on a Spanish fetal ultrasound dataset and tested on both in-domain data and an external African cohort to assess cross-population generalization. We found that FetalCLIP achieved the best results in the linear probing setting (F1 = 0.9261 for in-domain, F1 = 0.9731 for out-of-domain), while USFM performed best in the full fine-tuning setting (F1 = 0.9476 for in-domain, F1 = 0.9515 for out-of-domain). MOFO and UltraSAM degraded most in both settings, underperforming natural image pretrained models in some cases. These findings highlight how the choice of pretrained model strongly affects fetal plane classification performance, since different pretraining objectives lead to different levels of transferability.

Within-dataset class-split evaluation is widely used as a proxy for fully unconditional out-of-distribution anomaly detection. We show that this protocol can become ill-posed when the held-out anomaly class overlaps the normal mixture in representation space. In this regime, anomaly scores may collapse toward chance or even invert, and the preferred score direction can depend on the unknown anomaly class. We introduce a simple training-free diagnostic, neighborhood class leakage, and show that it predicts score-direction instability across Fashion-MNIST, CIFAR-10, and Imagenette, in both pixel and VAE latent spaces. Our results suggest that class-split AD benchmarks should be treated as geometry-dependent stress tests rather than unconditional evidence of anomaly-detection ability.

The growing availability of clinical data has increased the use of machine learning, yet centralized data aggregation is often infeasible for sensitive health information. Federated Learning (FL) offers a distributed alternative, but its adoption is limited by substantial heterogeneity across institutional datasets, making harmonization a critical but frequently overlooked prerequisite for multi-site analytics. We introduce PrivFusion, a privacy-preserving multi-agent framework that automates the harmonization of structured datasets prior to federated training. PrivFusion uses agents to analyze local data, cluster semantically similar features across sites, and provide iterative transformation recommendations until alignment is achieved. Evaluation across four heterogeneous COVID-19 datasets demonstrates that PrivFusion effectively and efficiently harmonizes multi-site data while substantially reducing manual effort.

Frozen Vision Foundation Models (VFMs) with lightweight classification heads are increasingly used in medical imaging because they offer efficient and reproducible deployment. Yet noisy-label learning methods for this frozen-feature regime remain poorly understood, and most existing methods still rely on a small-loss assumption inherited from end-to-end training. We present a controlled benchmark of eight noisy-label methods across five medical datasets, three backbones, two noise types, and five noise rates (150 conditions, 6,000 training runs), evaluated with balanced accuracy. The benchmark shows that there is no universal winner: Friedman ranking over the 150 conditions yields $χ^2 = 333.2$ ($p = 4.77 \times 10^{-68}$), ELR wins the most conditions (49/150), while CUFIT attains the best mean rank (2.51). The practical cost of method choice grows sharply with noise severity, from 4.5pp on clean data to 18.8pp at asymmetric 40% noise. To explain these benchmark-level patterns, we revisit the small-loss assumption in a representative high-risk regime. Under frozen DINOv2 features, clean and noisy loss distributions overlap by 53--61%, and matched-rate clean-sample detection shows that prediction agreement is markedly more stable than loss ranking under asymmetric noise (3pp vs.\ 13pp precision drop). On ISIC2019 with asymmetric 40% noise, Co-Teaching reaches 68% overall accuracy while collapsing to 35.1% balanced accuracy with zero recall on three minority classes. Together, these results recast noisy-label learning for frozen VFMs as a regime-aware method-selection problem rather than a search for a single dominant algorithm. We conclude with evidence-based guidance and a low-regret feature-space selector for practical recommendation.

Dataset distillation (DD) compresses a large training set into a small synthetic set for efficient training, but most DD methods optimize only clean accuracy and leave robustness uncontrolled. Recent robust DD methods improve robustness, yet they often suffer from a poor accuracy-robustness trade-off because they (i) treat all adversarially perturbed examples uniformly, despite robust risk being dominated by near-zero robust margins, and (ii) do not explicitly increase inter-class separation in the decision boundary where attacks concentrate. We present Contrastive Curriculum for Robust Dataset Distillation (C$^2$R), a framework that couples an attack-aware curriculum with a contrastive robustness objective. From a robust-margin perspective, we derive a perturbation score that approximates each sample's robust hinge, enabling a curriculum that prioritizes the smallest-margin adversaries that most directly drive robust error. In parallel, a class-balanced contrastive robustness loss enforces adversarial invariance while explicitly widening boundary separation across classes. Experiments on CIFAR-10/100, Tiny-ImageNet, and multiple ImageNet-1K subsets under six attacks show that C$^2$R achieves the best robust accuracy, outperforming prior robust DD by $2.8$% on average.