Large language models (LLMs) excel at explicit reasoning, but their implicit computational strategies remain underexplored. Decades of psychophysics research show that humans intuitively process and integrate noisy signals using near-optimal Bayesian strategies in perceptual tasks. We ask whether LLMs exhibit similar behaviour and perform optimal multimodal integration without explicit training or instruction. Adopting the psychophysics paradigm, we infer computational principles of LLMs from systematic behavioural studies. We introduce a behavioural benchmark - BayesBench: four magnitude estimation tasks (length, location, distance, and duration) over text and image, inspired by classic psychophysics, and evaluate a diverse set of nine LLMs alongside human judgments for calibration. Through controlled ablations of noise, context, and instruction prompts, we measure performance, behaviour and efficiency in multimodal cue-combination. Beyond accuracy and efficiency metrics, we introduce a Bayesian Consistency Score that detects Bayes-consistent behavioural shifts even when accuracy saturates. Our results show that while capable models often adapt in Bayes-consistent ways, accuracy does not guarantee robustness. Notably, GPT-5 Mini achieves perfect text accuracy but fails to integrate visual cues efficiently. This reveals a critical dissociation between capability and strategy, suggesting accuracy-centric benchmarks may over-index on performance while missing brittle uncertainty handling. These findings reveal emergent principled handling of uncertainty and highlight the correlation between accuracy and Bayesian tendencies. We release our psychophysics benchmark and consistency metric (https://bayes-bench.github.io) as evaluation tools and to inform future multimodal architecture designs.

As generative models become powerful, concerns around transparency, accountability, and copyright violations have intensified. Understanding how specific training data contributes to a model's output is critical. We introduce a framework for interpreting generative outputs through the automatic construction of ontologyaligned knowledge graphs (KGs). While automatic KG construction from natural text has advanced, extracting structured and ontology-consistent representations from visual content remains challenging -- due to the richness and multi-object nature of images. Leveraging multimodal large language models (LLMs), our method extracts structured triples from images, aligned with a domain-specific ontology. By comparing the KGs of generated and training images, we can trace potential influences, enabling copyright analysis, dataset transparency, and interpretable AI. We validate our method through experiments on locally trained models via unlearning, and on large-scale models through a style-specific experiment. Our framework supports the development of AI systems that foster human collaboration, creativity and stimulate curiosity.

Ensuring content safety in large language models (LLMs) is essential for their deployment in real-world applications. However, existing safety guardrails are predominantly tailored for high-resource languages, leaving a significant portion of the world's population underrepresented who communicate in low-resource languages. To address this, we introduce CREST (CRoss-lingual Efficient Safety Transfer), a parameter-efficient multilingual safety classification model that supports 100 languages with only 0.5B parameters. By training on a strategically chosen subset of only 13 high-resource languages, our model utilizes cluster-based cross-lingual transfer from a few to 100 languages, enabling effective generalization to both unseen high-resource and low-resource languages. This approach addresses the challenge of limited training data in low-resource settings. We conduct comprehensive evaluations across six safety benchmarks to demonstrate that CREST outperforms existing state-of-the-art guardrails of comparable scale and achieves competitive results against models with significantly larger parameter counts (2.5B parameters and above). Our findings highlight the limitations of language-specific guardrails and underscore the importance of developing universal, language-agnostic safety systems that can scale effectively to serve global populations.

Conformal prediction (CP) provides a comprehensive framework to produce statistically rigorous uncertainty sets for black-box machine learning models. To further improve the efficiency of CP, conformal correction is proposed to fine-tune or wrap the base model with an extra module using a conformal-aware inefficiency loss. In this work, we empirically and theoretically identify a trade-off between the CP efficiency and the entropy of model prediction. We then propose an entropy-constrained conformal correction method, exploring a better Pareto optimum between efficiency and entropy. Extensive experimental results on both computer vision and graph datasets demonstrate the efficacy of the proposed method. For instance, it can significantly improve the efficiency of state-of-the-art CP methods by up to 34.4%, given an entropy threshold.

Domain adaptation in object detection is critical for real-world applications where distribution shifts degrade model performance. Security X-ray imaging presents a unique challenge due to variations in scanning devices and environmental conditions, leading to significant domain discrepancies. To address this, we apply ALDI++, a domain adaptation framework that integrates self-distillation, feature alignment, and enhanced training strategies to mitigate domain shift effectively in this area. We conduct extensive experiments on the EDS dataset, demonstrating that ALDI++ surpasses the state-of-the-art (SOTA) domain adaptation methods across multiple adaptation scenarios. In particular, ALDI++ with a Vision Transformer for Detection (ViTDet) backbone achieves the highest mean average precision (mAP), confirming the effectiveness of transformer-based architectures for cross-domain object detection. Additionally, our category-wise analysis highlights consistent improvements in detection accuracy, reinforcing the robustness of the model across diverse object classes. Our findings establish ALDI++ as an efficient solution for domain-adaptive object detection, setting a new benchmark for performance stability and cross-domain generalization in security X-ray imagery.

Large language models (LLMs) are known to inherit and even amplify societal biases present in their pre-training corpora, threatening fairness and social trust. To address this issue, recent work has explored ``editing'' LLM parameters to mitigate social bias with model merging approaches; however, there is no empirical comparison. In this work, we empirically survey seven algorithms: Linear, Karcher Mean, SLERP, NuSLERP, TIES, DELLA, and Nearswap, applying 13 open weight models in the GPT, LLaMA, and Qwen families. We perform a comprehensive evaluation using three bias datasets (BBQ, BOLD, and HONEST) and measure the impact of these techniques on LLM performance in downstream tasks of the SuperGLUE benchmark. We find a trade-off between bias reduction and downstream performance: methods achieving greater bias mitigation degrade accuracy, particularly on tasks requiring reading comprehension and commonsense and causal reasoning. Among the merging algorithms, Linear, SLERP, and Nearswap consistently reduce bias while maintaining overall performance, with SLERP at moderate interpolation weights emerging as the most balanced choice. These results highlight the potential of model merging algorithms for bias mitigation, while indicating that excessive debiasing or inappropriate merging methods may lead to the degradation of important linguistic abilities.

Large language models have demonstrated remarkable capabilities across many tasks, yet face significant challenges when dealing with recursive reasoning problems, those requiring the resolution of nested hierarchical structures. While prior research has extensively studied length generalization (a model's ability to handle longer sequences than seen during training), we investigate a distinct and underexplored limitation: depth generalization. Here, depth refers to the number of nested levels in a hierarchical problem, such as the layers of parentheses in a mathematical expression or the nesting of logical clauses in a Boolean formula. Our work reveals that standard transformer architectures struggle with problems involving deeper recursion than encountered during training, even when they perform well on longer but non-nested sequences. This limitation stems from their inability to maintain stack-like behavior, the capacity to track and resolve multiple levels of nested dependencies. Through systematic analysis, we demonstrate how this architectural constraint leads to rapid performance decay as the depth of the recursion increases. To address this challenge, we develop a novel looped locate-and-replace pipeline that decomposes recursive problems into manageable subcomponents. The approach employs two specialized models: a locator that identifies solvable subexpressions and a replacer that evaluates these components while preserving the overall structure. We evaluated this method in three carefully designed domains: Boolean algebra, recursive arithmetic, and propositional logic, each with a controllable depth of recursion. We show that our method effectively alleviates the performance decay when tested on out-of-distribution recursion depth.

This paper presents a unified study of four distinct modeling approaches for classifying dysarthria severity in the Speech Analysis for Neurodegenerative Diseases (SAND) challenge. All models tackle the same five class classification task using a common dataset of speech recordings. We investigate: (1) a ViT-OF method leveraging a Vision Transformer on spectrogram images, (2) a 1D-CNN approach using eight 1-D CNN's with majority-vote fusion, (3) a BiLSTM-OF approach using nine BiLSTM models with majority vote fusion, and (4) a Hierarchical XGBoost ensemble that combines glottal and formant features through a two stage learning framework. Each method is described, and their performances on a validation set of 53 speakers are compared. Results show that while the feature-engineered XGBoost ensemble achieves the highest macro-F1 (0.86), the deep learning models (ViT, CNN, BiLSTM) attain competitive F1-scores (0.70) and offer complementary insights into the problem.

The de novo generation of molecules with desirable properties is a critical challenge, where diffusion models are computationally intensive and autoregressive models struggle with error propagation. In this work, we introduce the Graph VQ-Transformer (GVT), a two-stage generative framework that achieves both high accuracy and efficiency. The core of our approach is a novel Graph Vector Quantized Variational Autoencoder (VQ-VAE) that compresses molecular graphs into high-fidelity discrete latent sequences. By synergistically combining a Graph Transformer with canonical Reverse Cuthill-McKee (RCM) node ordering and Rotary Positional Embeddings (RoPE), our VQ-VAE achieves near-perfect reconstruction rates. An autoregressive Transformer is then trained on these discrete latents, effectively converting graph generation into a well-structured sequence modeling problem. Crucially, this mapping of complex graphs to high-fidelity discrete sequences bridges molecular design with the powerful paradigm of large-scale sequence modeling, unlocking potential synergies with Large Language Models (LLMs). Extensive experiments show that GVT achieves state-of-the-art or highly competitive performance across major benchmarks like ZINC250k, MOSES, and GuacaMol, and notably outperforms leading diffusion models on key distribution similarity metrics such as FCD and KL Divergence. With its superior performance, efficiency, and architectural novelty, GVT not only presents a compelling alternative to diffusion models but also establishes a strong new baseline for the field, paving the way for future research in discrete latent-space molecular generation.

Large language models (LLMs) are now used in settings such as Google's AI Overviews, where it summarizes multiple long documents. However, it remains unclear whether they weight all inputs equally. Focusing on abortion-related news, we construct 40 pro-neutral-con article triplets, permute each triplet into six input orders, and prompt Gemini 2.5 Flash to generate a neutral overview. We evaluate each summary against its source articles using ROUGE-L (lexical overlap), BERTScore (semantic similarity), and SummaC (factual consistency). One-way ANOVA reveals a significant primacy effect for BERTScore across all stances, indicating that summaries are more semantically aligned with the first-seen article. Pairwise comparisons further show that Position 1 differs significantly from Positions 2 and 3, while the latter two do not differ from each other, confirming a selective preference for the first document. The findings present risks for applications that rely on LLM-generated overviews and for agentic AI systems, where the steps involving LLMs can disproportionately influence downstream actions.