Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2608.27688v1 Announce Type: new Abstract: In this paper, we develop SafeStep, an interactive browser-based semantic communication platform for live pedestrian safety monitoring. SafeStep extrac...
arXiv:2608.28198v1 Announce Type: new Abstract: Privacy-preserving learning is often motivated by the idea that protecting users' data can preserve trust and thus participation, improving utility in...
Benchmarking competitions are central to AI development in medical imaging, but it is unclear if they provide representative, accessible, and reusable data for clinical relevance. This study systematically examined 249 challenges (458 tasks) across 19 imaging modalities, finding limited geographic, modality, and problem-type representation. Additionally, many datasets suffer from restrictive access, ambiguous licensing, and poor documentation, hindering reproducibility and long-term reuse.
The paper introduces Distortion Extenders (DEX), learnable parameters that adapt vision foundation models to fisheye cameras by modeling distortion coefficients and correcting distributional shifts between fisheye and perspective images. DEX is applied to monocular depth estimation and open‑vocabulary segmentation across convolutional and Transformer architectures, consistently outperforming baselines on indoor and outdoor fisheye datasets. Additionally, DEX activations can be decoded to obtain distortion coefficients, aiding camera calibration.
Computer use agents (CUAs) are vision-language models that perceive a screen and act on a real operating system through mouse, keyboard, and terminal, and they are increasingly deployed to automate ev...
Systematic generalization remains a significant challenge in deep learning. In particular, combinatorial generalization - generalizing to new configurations of known factors of variation - is effortle...
University students experience disproportionately high rates of common mental health conditions, such as depression, which can impair learning, social functioning, and overall well-being. Although lif...
Large language models (LLMs) demonstrate strong Helpfulness, Harmlessness, and Honesty (3H) alignment in English-centric settings, but these gains transfer poorly to low-resource languages due to cult...
Large Vision-Language Models (LVLMs) achieve strong performance across many multimodal tasks; however, they often exploit spurious object-background correlations, resulting in predictions driven by co...
Detecting hallucinations in Large Vision-Language Models (LVLMs) requires both accurate span localization and well-calibrated confidence scores. Fine-tuned generative VLMs excel at identifying halluci...
Personalized segmentation and personalized retrieval both aim to identify the same physical object across different images. While the former localizes the object within a target image, the latter retr...
Text-to-image models learn associations between concepts - in the case of this paper, people's professions, which we refer to as roles - and visual attributes. These associations can underpin many obs...
The same Persona behavior can be beneficial in one context but harmful in another, causing static Persona elicitation to perform inconsistently across tasks. We introduce the Persona Selection--Realiz...
The paper proposes five runtime primitives—discovery, identity, governance, attestation, and supply chain—to manage autonomous AI agents in enterprise settings. It argues that traditional control models fail because agents are transient, model-driven, and self‑discoverable, making runtime governance essential. The authors detail an implementation that mediates agent actions against policy, authorizes them via a per‑tenant vocabulary, and records them in a verifiable ledger, noting the associated operational costs and partial deployment status.
The paper presents an interpretable, fair, and accurately benchmarked automated system for assessing second‑language English speaking. Using a hybrid of feature‑based speech‑timing metrics and a large language model (LLM) fluency judgment, the system achieves a Spearman correlation of 0.818 with the ICNALE Global Rating Archive, outperforming 81 % of trained human raters. A controlled study shows that encoding pauses into the LLM prompt does not meaningfully affect fluency scores, indicating that the system’s fluency signal derives from measurable speech‑timing features.
The paper demonstrates that publicly available clinical machine learning models, such as logistic regression (LR), pose significant privacy risks because attackers can recover model parameters and identify training cohorts through various membership inference attacks. The authors show that even small cohorts can be reliably identified and that common practices like cross-validation can worsen the risk. To mitigate this, they propose a quantum-inspired defense that tensorizes discretized models into tensor trains (TTs), which obfuscates parameters, preserves accuracy, and maintains interpretability while providing black‑box protection comparable to Differential Privacy.
The paper investigates how the quality of instructions used to generate response pairs affects preference learning for language models. It shows that low‑quality or ambiguous instructions limit the range of response quality, weakening preference signals, and introduces an instruction‑refinement pipeline that improves data quality without discarding examples. Experiments across models and benchmarks demonstrate that refining instructions leads to better alignment and complements other data‑improvement methods.
DocTalkBN is a large-scale multimodal dataset of authentic expert telemedicine conversations in Bengali, comprising 557.63 hours of paired audio and text, 1,515 multi-turn patient calls, and 10,274 host–doctor question–answer exchanges across 26 medical specialties. The dataset contains 1.7 million tokens and preserves the spontaneity and contextual richness of real medical interactions in a low-resource language. Three downstream tasks—medical triage classification, advice safety evaluation, and medical named entity recognition—are constructed to benchmark large language models and encoder-based baselines, demonstrating DocTalkBN’s practical usefulness for clinically grounded reasoning.
The paper presents a method for cross‑architecture knowledge distillation from a fine‑tuned DINOv2 Vision Transformer teacher to a lightweight bidirectional Visual State Space Model (LVSSM) student for tea leaf disease classification. By addressing training‑stability issues with a progressive convolutional stem and gated selective‑scan block, the 4.45 M‑parameter student achieves a mean test accuracy of 95.41%—a 3.09‑point improvement over the teacher’s 92.32%—while using only one‑fifth of the teacher’s parameters. Ablation studies show that simple logit‑level distillation outperforms intermediate feature alignment, and the gains are specific to students that start below the teacher’s performance.
LLaVAFlow is an information‑theoretic distillation framework designed to preserve cross‑modal alignment in Multimodal Large Language Models during visual instruction tuning. It compresses the mutual information between extracted relations and MLLM embeddings to refine alignment flow, and then maximizes mutual information between pretrained and fine‑tuned alignment flows to transfer compact alignment information. Experiments demonstrate that LLaVAFlow effectively maintains alignment flow, improving downstream performance and generalization.