Unlike text, speech conveys information about the speaker, such as gender, through acoustic cues like pitch. This gives rise to modality-specific bias concerns. For example, in speech translation (ST), when translating from languages with notional gender, such as English, into languages where gender-ambiguous terms referring to the speaker are assigned grammatical gender, the speaker's vocal characteristics may play a role in gender assignment. This risks misgendering speakers, whether through masculine defaults or vocal-based assumptions. Yet, how ST models make these decisions remains poorly understood. We investigate the mechanisms ST models use to assign gender to speaker-referring terms across three language pairs (en-es/fr/it), examining how training data patterns, internal language model (ILM) biases, and acoustic information interact. We find that models do not simply replicate term-specific gender associations from training data, but learn broader patterns of masculine prevalence. While the ILM exhibits strong masculine bias, models can override these preferences based on acoustic input. Using contrastive feature attribution on spectrograms, we reveal that the model with higher gender accuracy relies on a previously unknown mechanism: using first-person pronouns to link gendered terms back to the speaker, accessing gender information distributed across the frequency spectrum rather than concentrated in pitch.

Transformer-based models have become state-of-the-art tools in various machine learning tasks, including time series classification, yet their complexity makes understanding their internal decision-making challenging. Existing explainability methods often focus on input-output attributions, leaving the internal mechanisms largely opaque. This paper addresses this gap by adapting various Mechanistic Interpretability techniques; activation patching, attention saliency, and sparse autoen-coders, from NLP to transformer architectures designed explicitly for time series classification. We systematically probe the internal causal roles of individual attention heads and timesteps, revealing causal structures within these models. Through experimentation on a benchmark time series dataset, we construct causal graphs illustrating how information propagates internally, highlighting key attention heads and temporal positions driving correct classifications. Additionally, we demonstrate the potential of sparse autoencoders for uncovering interpretable latent features. Our findings provide both methodological contributions to transformer interpretability and novel insights into the functional mechanics underlying transformer performance in time series classification tasks.

Translating non-invasive signals such as photoplethysmography (PPG) and ballistocardiography (BCG) into clinically meaningful signals like arterial blood pressure (ABP) is vital for continuous, low-cost healthcare monitoring. However, temporal misalignment in multimodal signal transformation impairs transformation accuracy, especially in capturing critical features like ABP peaks. Conventional synchronization methods often rely on strong similarity assumptions or manual tuning, while existing Learning with Noisy Labels (LNL) approaches are ineffective under time-shifted supervision, either discarding excessive data or failing to correct label shifts. To address this challenge, we propose ShiftSyncNet, a meta-learning-based bi-level optimization framework that automatically mitigates performance degradation due to time misalignment. It comprises a transformation network (TransNet) and a time-shift correction network (SyncNet), where SyncNet learns time offsets between training pairs and applies Fourier phase shifts to align supervision signals. Experiments on one real-world industrial dataset and two public datasets show that ShiftSyncNet outperforms strong baselines by 9.4%, 6.0%, and 12.8%, respectively. The results highlight its effectiveness in correcting time shifts, improving label quality, and enhancing transformation accuracy across diverse misalignment scenarios, pointing toward a unified direction for addressing temporal inconsistencies in multimodal physiological transformation.

The widespread use of neural surrogates in automotive aerodynamics, enabled by datasets such as DrivAerML and DrivAerNet++, has primarily focused on bluff-body flows with large wakes. Extending these methods to aerospace, particularly in the transonic regime, remains challenging due to the high level of non-linearity of compressible flows and 3D effects such as wingtip vortices. Existing aerospace datasets predominantly focus on 2D airfoils, neglecting these critical 3D phenomena. To address this gap, we present a new dataset of CFD simulations for 3D wings in the transonic regime. The dataset comprises volumetric and surface-level fields for around $30,000$ samples with unique geometry and inflow conditions. This allows computation of lift and drag coefficients, providing a foundation for data-driven aerodynamic optimization of the drag-lift Pareto front. We evaluate several state-of-the-art neural surrogates on our dataset, including Transolver and AB-UPT, focusing on their out-of-distribution (OOD) generalization over geometry and inflow variations. AB-UPT demonstrates strong performance for transonic flowfields and reproduces physically consistent drag-lift Pareto fronts even for unseen wing configurations. Our results demonstrate that AB-UPT can approximate drag-lift Pareto fronts for unseen geometries, highlighting its potential as an efficient and effective tool for rapid aerodynamic design exploration. To facilitate future research, we open-source our dataset at https://huggingface.co/datasets/EmmiAI/Emmi-Wing.

Readability assessment aims to evaluate the reading di ffi culty of a text. In recent years, while deep learning technol - ogy has been gradually applied to readability assessment, m ost approaches fail to consider either the length of the text or the ordinal relationship of readability labels. This pap er proposes a bidirectional readability assessment mechan ism that captures contextual information to identify regions w ith rich semantic information in the text, thereby predicti ng the readability level of individual sentences. These sente nce-level labels are then used to assist in predicting the ov erall readability level of the document. Additionally, a pairwis e sorting algorithm is introduced to model the ordinal relationship between readability levels through label subtrac tion. Experimental results on Chinese and English datasets demonstrate that the proposed model achieves competitive p erformance and outperforms other baseline models. Introduction Automatic Text Readability (ARA) research originated in th e early 20th century, aiming to evaluate text reading di ffi culty and assist educators in recommending appropriate rea ding materials for learners [ 1 ]. Readability assessment approaches are generally classified into three paradig ms: human evaluation, co-selection-based analysis, and content-based analysis. Human evaluation involves expert annotation or reader surveys; co-selection methods leverage user interaction data such as reading time or choices [ 2 ]; and content-based approaches infer readability using linguistic, syntactic, or semantic features extracted fro m the text itself. Early studies predominantly relied on experts' subjective evaluations and simple statistical feat ures, such as sentence length and word complexity.

Ensemble learning improves classification performance by combining multiple base classifiers. While increasing the number of classifiers generally enhances accuracy, excessively large ensembles can lead to computational inefficiency and diminishing returns. This paper investigates the relationship between ensemble size and performance through the lens of linear independence among classifier votes in data streams. We propose that ensembles composed of linearly independent classifiers maximize representational capacity, particularly under a geometric model. We then generalize the importance of linear independence to the weighted majority voting problem. By modeling the probability of achieving linear independence among classifier outputs, we derive a theoretical framework that explains the trade-off between ensemble size and accuracy. Our analysis leads to a theoretical estimate of the ensemble size required to achieve a user-specified probability of linear independence. We validate our theory through experiments on both real-world and synthetic datasets using two ensemble methods, OzaBagging and GOOWE. Our results confirm that this theoretical estimate effectively identifies the point of performance saturation for robust ensembles like OzaBagging. Conversely, for complex weighting schemes like GOOWE, our framework reveals that high theoretical diversity can trigger algorithmic instability. Our implementation is publicly available to support reproducibility and future research.

Ensuring the safety of embodied AI agents during task planning is critical for real-world deployment, especially in household environments where dangerous instructions pose significant risks. Existing methods often suffer from either high computational costs due to preference alignment training or over-rejection when using single-agent safety prompts. To address these limitations, we propose MADRA, a training-free Multi-Agent Debate Risk Assessment framework that leverages collective reasoning to enhance safety awareness without sacrificing task performance. MADRA employs multiple LLM-based agents to debate the safety of a given instruction, guided by a critical evaluator that scores responses based on logical soundness, risk identification, evidence quality, and clarity. Through iterative deliberation and consensus voting, MADRA significantly reduces false rejections while maintaining high sensitivity to dangerous tasks. Additionally, we introduce a hierarchical cognitive collaborative planning framework that integrates safety, memory, planning, and self-evolution mechanisms to improve task success rates through continuous learning. We also contribute SafeAware-VH, a benchmark dataset for safety-aware task planning in VirtualHome, containing 800 annotated instructions. Extensive experiments on AI2-THOR and VirtualHome demonstrate that our approach achieves over 90% rejection of unsafe tasks while ensuring that safe-task rejection is low, outperforming existing methods in both safety and execution efficiency. Our work provides a scalable, model-agnostic solution for building trustworthy embodied agents.

Large language models excel on broad multilingual benchmarks but remain to be evaluated extensively in figurative and culturally grounded reasoning, especially in low-resource contexts. We present BengaliFig, a compact yet richly annotated challenge set that targets this gap in Bengali, a widely spoken low-resourced language. The dataset contains 435 unique riddles drawn from Bengali oral and literary traditions. Each item is annotated along five orthogonal dimensions capturing reasoning type, trap type, cultural depth, answer category, and difficulty, and is automatically converted to multiple-choice format through a constraint-aware, AI-assisted pipeline. We evaluate eight frontier LLMs from major providers under zero-shot and few-shot chain-of-thought prompting, revealing consistent weaknesses in metaphorical and culturally specific reasoning. BengaliFig thus contributes both a diagnostic probe for evaluating LLM robustness in low-resource cultural contexts and a step toward inclusive and heritage-aware NLP evaluation.

Self-assessment is a key aspect of reliable intelligence, yet evaluations of large language models (LLMs) focus mainly on task accuracy. We adapted the 10-item General Self-Efficacy Scale (GSES) to elicit simulated self-assessments from ten LLMs across four conditions: no task, computational reasoning, social reasoning, and summarization. GSES responses were highly stable across repeated administrations and randomized item orders. However, models showed significantly different self-efficacy levels across conditions, with aggregate scores lower than human norms. All models achieved perfect accuracy on computational and social questions, whereas summarization performance varied widely. Self-assessment did not reliably reflect ability: several low-scoring models performed accurately, while some high-scoring models produced weaker summaries. Follow-up confidence prompts yielded modest, mostly downward revisions, suggesting mild overestimation in first-pass assessments. Qualitative analysis showed that higher self-efficacy corresponded to more assertive, anthropomorphic reasoning styles, whereas lower scores reflected cautious, de-anthropomorphized explanations. Psychometric prompting provides structured insight into LLM communication behavior but not calibrated performance estimates.

Payment networks form the backbone of modern commerce, generating high volumes of transaction records from daily activities. Properly modeling this data can enable applications such as abnormal behavior detection and consumer-level insights for hyper-personalized experiences, ultimately improving people's lives. In this paper, we present TREASURE, TRansformer Engine As Scalable Universal transaction Representation Encoder, a multipurpose transformer-based foundation model specifically designed for transaction data. The model simultaneously captures both consumer behavior and payment network signals (such as response codes and system flags), providing comprehensive information necessary for applications like accurate recommendation systems and abnormal behavior detection. Verified with industry-grade datasets, TREASURE features three key capabilities: 1) an input module with dedicated sub-modules for static and dynamic attributes, enabling more efficient training and inference; 2) an efficient and effective training paradigm for predicting high-cardinality categorical attributes; and 3) demonstrated effectiveness as both a standalone model that increases abnormal behavior detection performance by 111% over production systems and an embedding provider that enhances recommendation models by 104%. We present key insights from extensive ablation studies, benchmarks against production models, and case studies, highlighting valuable knowledge gained from developing TREASURE.