Variational inference in probabilistic graphical models aims to approximate fundamental quantities such as marginal distributions and the partition function. Popular approaches are the Bethe approximation, tree-reweighted, and other types of convex free energies. These approximations are efficient but can fail when the model is complex and highly interactive. In this work, we analyze two classes of approximations that generalize the above methods as special cases: one resulting from changing the state energy of the model, and the other from choosing a different entropy approximation. We discuss benefits and limitations of each approximation, identify favorable regimes for their parameters, and propose ADAPT-$c$ and ADAPT-$\zeta$ to automatically detect these optimal settings for a given pairwise binary graphical model. In experiments, we demonstrate their effectiveness for approximating marginals and the partition function.

The efficient multiangle centered discrete fractional Fourier transform (MA-CDFRFT) [1] has proven to be a useful tool for time-frequency analysis; in this paper, we generalize the MA-CDFRFT to general M-periodic transforms, which, among others, include the standard discrete Fourier, discrete sine, discrete cosine, Hadamard and discrete Hartley transform. Furthermore, we exploit the symmetries inherent to the MA-CDFRFT and our novel multiangle standard discrete fractional Fourier transform (MA-DFRFT) to halve the number of FFTs needed to compute these transforms, which paves the way for applications in resource-constrained environments.

In this paper, we propose a novel method for frequency modulated continuous wave (FMCW) radar mutual interference mitigation (IM) based on the discrete fractional Fourier transform (DFrFT). Interference chirps are detected and mitigated by compression and zeroing in the fractional domain. We provide an efficient implementation that can deal with multiple interferers, where we perform consecutive DFrFTs utilizing its angle-additivity property. For that purpose, we generalize and reduce the computational complexity of the multi-angle centered discrete fractional Fourier transform [1]. Our algorithm is designed to be simple and fast such that it can be implemented in hardware. We evaluate our algorithm on a synthetic I/Q-modulated dataset and outperform reference methods in terms of the mean squared error, signal-to-interference-plus-noise ratio, error vector magnitude, true positive rate, false alarm rate and F1-score.

The widespread adoption of large language models (LLMs) has increased concerns about their robustness. Vulnerabilities in perturbations of tokenisation of the input indicate that models trained with a deterministic canonical tokenisation can be brittle to adversarial attacks. Recent studies suggest that stochastic tokenisation can deliver internal representations that are less sensitive to perturbations. In this paper, we analyse how stochastic tokenisations affect robustness to adversarial attacks and random perturbations. We systematically study this over a range of learning regimes (pre-training, supervised fine-tuning, and in-context learning), data sets, and model architectures. We show that pre-training and fine-tuning with uniformly sampled stochastic tokenisations improve robustness to random and adversarial perturbations. Evaluating on uniformly sampled non-canonical tokenisations reduces the accuracy of a canonically trained Llama-1b model by 29.8

Electro-laryngeal (EL) speech is characterized by constant pitch, limited prosody, and mechanical noise, reducing naturalness and intelligibility. We propose a lightweight adaptation of the state-of-the-art StreamVC framework to this setting by removing pitch and energy modules and combining self-supervised pretraining with supervised fine-tuning on parallel EL and healthy (HE) speech data, guided by perceptual and intelligibility losses. Objective and subjective evaluations across different loss configurations confirm their influence: the best model variant, based on WavLM features and human-feedback predictions (+WavLM+HF), drastically reduces character error rate (CER) of EL inputs, raises naturalness mean opinion score (nMOS) from 1.1 to 3.3, and consistently narrows the gap to HE ground-truth speech in all evaluated metrics. These findings demonstrate the feasibility of adapting lightweight voice conversion architectures to EL voice rehabilitation while also identifying prosody generation and intelligibility improvements as the main remaining bottlenecks.

In this paper, we present a variational inference algorithm that decomposes a signal into multiple groups of related spectral lines. The spectral lines in each group are associated with a group parameter common to all spectral lines within the group. The proposed algorithm jointly estimates the group parameters, the number of spetral lines within a group, and the number of groups exploiting a Bernoulli-Gamma-Gaussian hierarchical prior model which promotes sparse solutions. Aiming to maximize the evidence lower bound (ELBO), variational inference provides analytic approximations of the posterior probability density functions (PDFs) and also gives estimates of the additional model parameters such as the measurement noise variance. While the activation variables of the groups and the associated group parameters (such as fundamental frequencies and the corresponding higher order harmonics) are estimated as point estimates, the remaining parameters such as the complex amplitudes of the spectral lines and their precision parameters are estimated as approximate posterior PDFs. We demonstrate the versatility and performance of the proposed algorithm on three different inference problems. In particular, the proposed algorithm is applied to the multi-pitch estimation problem, the radar signal-based extended object estimation problem, and variational mode decomposition (VMD) using synthetic measurements and to real multi-pitch estimation problem using the Bach-10 dataset. The results show that the proposed algorithm outperforms state-of-the-art model-based and pre-trained algorithms on all three inference problems.

Child presence detection (CPD) is motivated by the need for safety and well-being of unattended infants in vehicles. This paper explores the detection of unattended infants using ultra-wideband radar with a low-resource hardware setup in realistic in-car scenarios. An infant dummy, capable of simulating breathing as well as limb and head motion, is used to test various occupancy and activity states in combination with human adults. Additional realistic external factors like passing individuals outside the car are introduced. We propose features extracted from radar data combined with a two-stage gradient boosting classifier. For comparison, convolutional neural networks are directly employed on radar maps. Experiments across multiple vehicles and individuals achieve F1-scores on the test data of up to 97.0 % for cars of similar sizes and 93.2 % for a vehicle of a different size.

Test-time adaptation (TTA) aims to improve model robustness under domain shifts without access to source data–an essential capability for real-world applications such as autonomous driving and robotics. Existing TTA methods for semantic segmentation often rely on stochastic techniques like Monte Carlo dropout or augmentation-averaged predictions to estimate uncertainty or stabilize outputs. However, these approaches typically require multiple forward passes, which are computationally expensive and limit real-time applicability. We propose GaPaTTA, a lightweight and deterministic TTA framework built on SegFormer. Unlike previous methods, GaPaTTA adopts a single forward pass with a traditional augmentation strategy, avoiding repeated inference required by ensemble-based TTA approaches. Key innovations include: (1) Grad-CAM-based global prompt placement identifies the most relevant encoder layers for adaptation; (2) Gaussian entropy-guided local prompt injection selects the top-K most uncertain pixels; (3) Shannon entropy-based filtering suppresses unreliable pseudo-labels; and (4) cross-stage consistency aligns mid- and high-level features for structural coherence. Experiments on ACDC (A-Fog, A-Night, A-Rain, A-Snow), Cityscapes-Foggy (CS-Fog) and Cityscapes-Rainy (CS-Rain) demonstrate that GaPaTTA consistently outperforms previous TTA methods in mean intersection over union (mIoU) while reducing inference time by over 50 https://github.com/ml4papers/GaPaTTA .

Test-time adaptation (TTA) aims to improve the robustness of segmentation models to an unlabeled target domain without requiring access to the source data. While existing TTA methods have achieved promising results on image classification, they often fail to translate effectively to semantic segmentation due to the spatial complexity and fine-grained nature of dense predictions. We propose PromptCAL, a lightweight and effective TTA framework tailored for semantic segmentation, built upon the SegFormer architecture. Our method addresses two central challenges: (1) Which model component to adapt remains underexplored. Using Grad-CAM visualization and sensitivity analysis, we identify Stage 2 of the transformer backbone as the most domain-sensitive and restrict adaptation to this stage. (2) How to identify reliable supervision during adaptation is critical. We introduce a confidence-aware self-training mechanism based on per-pixel entropy filtering to guide pixel selection for model adaptation, ensuring label quality and model transferability. In addition, we incorporate lightweight prompt injection to enhance the adaptability of mid-level features. Our method achieves competitive improvements over the state-of-the-art while maintaining high adaptation efficiency and significantly reducing runtime overhead. Extensive experiments on corrupted semantic segmentation benchmarks, including ACDC (A-fog, A-night, A-rain, and A-snow), Cityscapes-foggy (CS-fog) and Cityscapes-rainy (CS-rain) demonstrate that PromptCAL achieves comparable or superior accuracy to state-of-the-art TTA baselines, while reducing adaptation time by over 50% per domain. This makes it a practical solution for efficient TTA in smart cities and edge-deployed vision systems. The source code is available at https://github.com/ml4papers/PromptCAL.

In the COVID-19 pandemic, a rigorous testing scheme was crucial. However, tests can be time-consuming and expensive. A machine learning-based diagnostic tool for audio recordings could enable widespread testing at low costs. In order to achieve comparability between such algorithms, the DiCOVA challenge was created. It is based on the Coswara dataset offering the recording categories cough, speech, breath and vowel phonation. Recording durations vary greatly, ranging from one second to over a minute. A base model is pre-trained on random, short time intervals. Subsequently, a Multiple Instance Learning (MIL) model based on self-attention is incorporated to make collective predictions for multiple time segments within each audio recording, taking advantage of longer durations. In order to compete in the fusion category of the DiCOVA challenge, we utilize a linear regression approach among other fusion methods to combine predictions from the most successful models associated with each sound modality. The application of the MIL approach significantly improves generalizability, leading to an AUC ROC score of 86.6% in the fusion category. By incorporating previously unused data, including the sound modality ’sustained vowel phonation’ and patient metadata, we were able to significantly improve our previous results reaching a score of 92.2%.

Variational inference in probabilistic graphical models aims to approximate fundamental quantities such as marginal distributions and the partition function. Popular approaches are the Bethe approximation, tree-reweighted, and other types of convex free energies. These approximations are efficient but can fail if the model is complex and highly interactive. In this work, we analyze two classes of approximations that include the above methods as special cases: first, if the model parameters are changed; and second, if the entropy approximation is changed. We discuss benefits and drawbacks of either approach, and deduce from this analysis how a free energy approximation should ideally be constructed. Based on our observations, we propose approximations that automatically adapt to a given model and demonstrate their effectiveness for a range of difficult problems.

This paper presents methods for prominence classification in conversational speech. Most existing tools rely on prosodic features extracted at syllable- or phone-level, performing well on read speech. This is not the case for conversational speech, where the quality of automatic segmentation is significantly worse. We introduce entropy-based chroma features, requiring only word-level segmentations. They perform equally well as a random forest classifier with prosodic features (requiring phone-level segmentation), with accuracies in the range of the human inter-rater agreement. We further use Bayesian deep learning to quantify the epistemic and aleatoric uncertainty of the prediction for prosodic and chroma features. Whereas the aleatoric uncertainty is, as expected, consistent with inter-rater agreement and similarly high for both feature sets, the epistemic uncertainty is lower for the classifier based on chroma features, indicating higher classification consistency across the corpus.

Optimizing sampling efficiency is crucial for solving complex material design challenges, especially with a limited experimental budget. This study focuses on improving sampling efficiency by reducing the search space for carbide-free bainitic steels through the uncertainty-aware modeling of constraints. These constraints include avoiding the formation of undesirable competing phases such as carbides, ferrite, and martensite, as well as accounting for practical limitations on phase transformation durations. Experimental data, obtained through dilatometry and metallography, inform most constraints, except for the presence of carbides. To model these constraints, we use machine learning (ML) models trained on a combination of newly acquired experimental data and experimental data from the literature. Predicting properties in unexplored regions of the design space can lead to inaccuracies. Thus, reliable uncertainty quantification is essential to avoid excluding parts of the design space due to overconfident erroneous predictions. To address this, we employ conformal prediction (CP), a distribution-free framework that provides calibrated post-hoc uncertainty estimates for the different ML models, ensuring reliable extrapolations without prematurely excluding viable design regions. This approach achieves a reduction ranging from 80% to more than 99% depending on the strictness of the employed criteria reduction in the search space, greatly enhancing sampling efficiency without compromising reliability.

Pathological speech, caused by dysphonia or produced via electro-larynx devices, often suffers from poor intelligibility and unnatural prosody. In this paper, we investigate the potential of four state-of-the-art voice conversion models: FreeVC, QuickVC, LLVC, and XVC for restoring healthy-sounding speech. All models are fine-tuned on Austrian-German datasets and evaluated using objective and subjective metrics. Results show substantial gains in intelligibility, naturalness, and perceived vocal health. QuickVC, FreeVC, and XVC perform similarly and achieve the highest preference scores, exceeding unprocessed pathological speech by up to 200%. These findings highlight the potential to improve communication for individuals with voice disorders and motivate further development of efficient, high-quality conversion systems.

Profiled power analysis is one of the most powerful forms of passive side-channel attacks. Over the last two decades, many works have analyzed their impact on cryptographic implementations as well as corresponding countermeasure techniques. To date, the most advanced variants of profiled power analysis are based on Soft-analytical Side-Channel Attacks (SASCA). After the initial profiling phase, a SASCA adversary creates a probabilistic graphical model, called a factor graph, of the target implementation and encodes the results of the previous step as prior information. Then, an inference algorithm such as loopy Belief Propagation (BP) can be used to recover the distribution of a target variable in the graph, i.e., sensitive data/keys. Designers of cryptographic implementations aim to reduce information leakage as much as possible and assess how much leakage can be allowed without compromising security requirements. Despite the existence of many works on profiled power analysis, it is still notoriously difficult to state under which conditions a cryptographic implementation provides sufficient protection against a profiling attacker with certain capabilities. In particular, it is unknown when a BP-based attack is optimal or whether tuning some heuristics in that algorithm may significant strengthen the attack. This knowledge gap led us to investigate the effectiveness of BP for SASCAs by studying the modes of failures of BP in the context of the SASCA, and systematically analyzing the behavior of BP on practically-relevant factor graphs. We use exact inference to gauge the quality of the approximation provided by BP. Through this assessment, we show that there exists a significant disparity between BP and exact inference in terms of guessing entropy when performing SASCAs on several classes of factor graphs. We further review and analyze various BP improvement heuristics from the literature.

Test-time adaptation (TTA) aims to adjust the parameters of a pre-trained source model using samples from the target domain, without requiring access to the source data. While recent studies have shown the potential of TTA across various computer vision tasks, most TTA methods are limited to uni-modal adaptation, and the domain shift caused by unimodal data corruption in multimodal tasks is not adequately addressed. Although some recent approaches have reduced cross-modal information discrepancy through modality-sharing modules, the domain adaptation for modality-specific modules has been overlooked. In this paper, we introduce a two-level test-time adaptation method (2LTTA) that accounts for both intra-modal distribution shifts and cross-modal reliability bias in multimodal learning (MML). Unlike conventional TTA methods, which focus primarily on fine-tuning normalization layers, 2LTTA modulates all normalization layers, self-Attention modules of the encoder related to the corrupted modality, and the modality-sharing block. Additionally, we design a two-level objective function that addresses both intra-modal distribution shift and cross-modal reliability bias in the modality fusion block. First, Shannon entropy with sample reweighting is used to mitigate intra-modal distribution shifts caused by data corruption. Second, a diversity-promoting loss is incorporated to reduce cross-modal information discrepancy. Our experiments show that 2LTTA outperforms baseline methods across various datasets.

Test-time adaptation (TTA) aims to adapt a model trained on a source domain to an unlabeled target domain without requiring source data. Traditional TTA assumes that the entire target domain data is available for observation during adaptation. In contrast, online test-time adaptation (OTTA) processes target domain data sequentially in mini-batches, enabling real-time model updates as new data arrives. While recent studies have demonstrated the potential of OTTA for various computer vision tasks, several challenges remain. This paper addresses two key issues: (1) While existing methods adapt different model components (e.g., batch normalization (BN) layers, fully connected (FC) layers, and the whole feature extractor), the most effective strategy is unclear. We propose a hybrid fine-tuning approach that modulates only the shallow convolutional (Conv) layers and all BN layers, enabling a more precise and effective adaptation compared to conventional methods. (2) The objective function plays a crucial role in enhancing prediction accuracy. Unlike previous TTA frameworks that focus solely on entropy minimization, we introduce a robust objective function that combines a diversity-promoting regularization term to balance certainty and diversity, along with a confidence-enhanced loss term that prioritizes high-confidence predictions. Our experiments on various corruption datasets, using both CNN and vision transformer (ViT)-based models, demonstrate that our approach outperforms state-of-the-art methods across a wide range of applications. Our implementation is publicly available.

Accurate temperature prediction in rotary cement kilns is crucial for process stability and equipment longevity. However, the repeated application of the predicted values creates an error accumulation over the length of the forecast, causing a domain drift of the predictions. This issue is exacerbated for image prediction, as more degrees of freedom lead to a higher sensitivity to small errors as local structures are lost. Using a vector-quantized autoencoder can mitigate the problem as it can map predictions back to the source domain, but it leads to almost constant predictions. Thus, we propose the usage of an additional diffusion model to avoid the local minimum of constant predictions. Our method maintains reliable in-domain predictions, preventing localized temperature peaks and ensuring stable kiln operation. Our experiments show, that the proposed vector-quantized diffusion model (VQ-Diff) can forecast much longer time sequences than reference methods with high accuracy, by being limited to the generation of in-domain images.

Real-world optimization problems often involve complex physical or industrial process constraints that must be met to ensure feasible solutions. Constrained Bayesian Optimization (CBO) provides an efficient framework for such optimization tasks that relies on uncertainty-aware surrogate models to approximate both the objective function and the constraints. Existing approaches for modeling the constraints often assume homoscedastic Gaussian noise and utilize Gaussian processes (GPs), which can lead to miscalibrated uncertainty estimates, particularly in the presence of model mismatch. To address these challenges, we propose a novel framework that improves both the calibration and adaptivity of constraint modeling. First, we use conformal prediction (CP) to construct prediction intervals with guaranteed calibration, independent of the noise distribution. Next, we improve local adaptivity by modeling residuals with a nonparametric kernel density estimator, enabling the intervals to adjust dynamically to heteroscedastic and non-Gaussian noise. Finally, we use a distance-based uncertainty heuristic to detect distribution shifts, improving robustness in regions with sparse or out-of-distribution samples. Our framework is model-agnostic and can be applied to any prediction model. We validate our method on a synthetic CBO task and on real-world data by modeling constraints for carbite-free bainitic steel optimization [1]. Our results show that standard GP models often produce miscalibrated uncertainty estimates, while our method yields improved calibration, and reduced prediction intervals.

Modern vehicles increasingly rely on sensors to monitor their environment and to support driver assistance and safety systems. Most vehicles use a variety of different sensors to improve robustness. A vital part of these is the radar sensor. It provides the vehicle not only with location but also with valuable velocity information from surrounding objects. The increasing usage of radar systems in road traffic also causes problems in terms of mutual interference between different radar sensors. This interference leads to broadband disturbances in the signal which must be mitigated to ensure reliable object detection and object angle estimation. In this article, we compare different variants of convolutional neural networks (CNNs) in their ability to mitigate mutual interference for multiantenna radar data. We analyze the potential of using multiantenna data for real-valued CNN (RVCNN) and complex-valued (CVCNN) models, comparing detection, phase reconstruction, and angle estimation performances. Furthermore, we propose a complex-valued CVCNN (CVCNN) architecture using a modified batch normalization method that omits activation scaling. Our experiments show, that using multiantenna data in combination with CVCNNs can greatly improve detection, phase, as well as angle estimation performance and that activation scaling is detrimental to our CVCNN architecture.