SafeCommit: Certifying When Memory-Grounded Agents May Safely Act

Long-horizon agents increasingly use persistent memory and tools to take actions with external side effects. A central failure mode is premature commitment: an agent acts before resolving whether its memory grounding is stale, conflicting, incomplete, or corrupted. We formalize this problem as safe commitment under memory uncertainty and introduce SafeCommit, a risk controlled layer between agent reasoning and external execution. The layer constructs a calibrated set of plausible latent worlds from memory, observations, tool outputs, provenance, and policy constraints. It permits a side effectful action only when a conformal action certificate shows that the action is safe in every retained world. Otherwise, it selects a low-side-effect probe that targets the worlds blocking certification, or returns a conservative fallback. Under calibrated world coverage, the probability of an unsafe certified commit is at most the target level α; with imperfect world proposal, the bound separates calibration and representation error. A dependency-free controlled simulator illustrates the safety-utility tradeoff and reproduces all reported results with one command. The goal is to offer a concrete approach for deciding not only what an agent should do, but when the available evidence is sufficient to safely do it.

Listening with Attention: Entropy-Guided Explainability for Transformer-Based Audio Models

Transformer-based automatic speech recognition (ASR) models such as Whisper are highly accurate, but their predictions remain difficult to interpret. Existing explainable AI (XAI) methods often lack faithfulness and precise temporal grounding. We propose Listening with Entropy-guided Attention for Faithful explainability (LEAF-X), a model-intrinsic XAI framework for transformer-based ASR. LEAF-X combines entropy-guided attention weighting, multi-layer attention rollout, and optional causal ablations to identify low-entropy, high-impact heads and layers, producing sparse token-to-frame attributions. Unlike perturbation-based explainers or raw attention maps, LEAF-X exploits the internal structure of encoder-decoder and speech-augmented decoder-only models to generate explanations that better reflect model computation. Results show 32% improved faithfulness, 35-39% stronger locality/sparsity, and the most stable attributions, supporting more transparent and auditable ASR.

AC-MIL: Weakly Supervised Atrial LGE-MRI Quality Assessment via Adversarial Concept Disentanglement

High-quality Late Gadolinium Enhancement (LGE) MRI can be helpful for atrial fibrillation management, yet scan quality is frequently compromised by patient motion, irregular breathing, and suboptimal image acquisition timing. While Multiple Instance Learning (MIL) has emerged as a powerful tool for automated quality assessment under weak supervision, current state-of-the-art methods map localized visual evidence to a single, opaque global feature vector. This black box approach fails to provide actionable feedback on specific failure modes, obscuring whether a scan degrades due to motion blur, inadequate contrast, or a lack of anatomical context. In this paper, we propose Adversarial Concept-MIL (AC-MIL), a weakly supervised framework that decomposes global image quality into clinically defined radiological concepts using only volume-level supervision. To capture latent quality variations without entangling predefined concepts, our framework incorporates an unsupervised residual branch guided by an adversarial erasure mechanism to strictly prevent information leakage. Furthermore, we introduce a spatial diversity constraint that penalizes overlap between distinct concept attention maps, ensuring localized and interpretable feature extraction. Extensive experiments on a clinical dataset of atrial LGE-MRI volumes demonstrate that AC-MIL successfully opens the MIL black box, providing highly localized spatial concept maps that allow clinicians to pinpoint the specific causes of non-diagnostic scans. Crucially, our framework achieves this deep clinical transparency while maintaining highly competitive ordinal grading performance against existing baselines. Code to be released on acceptance.

VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Models

Vision-language-action (VLA) models are emerging as embodied foundation models for robotic manipulation, but their deployment introduces a new unlearning challenge: removing unsafe, spurious, or privacy-sensitive behaviors without degrading perception, language grounding, and action control. In OpenVLA-style policies, behavior is produced through a fused visual encoder, a cross-modal projector, and a language backbone that predicts tokenized robot actions, so undesirable knowledge can be distributed across perception, alignment, and reasoning/action layers rather than confined to a single module. Consequently, partial unlearning applied only to the vision stack or only to the language backbone is often insufficient, while conventional unlearning baselines designed for standalone vision or language models may leave residual forgetting or incur unnecessary utility loss in embodied settings. We propose VLA-Forget, a hybrid unlearning framework that combines ratio-aware selective editing for perception and cross-modal specificity with layer-selective reasoning/action unlearning for utility-preserving forgetting. VLA-Forget jointly optimizes three objectives: targeted forgetting, perceptual preservation, and reasoning retention, through staged updates over the visual encoder, projector, and upper action-generating transformer blocks. Across forget-set behavior probes and retain-task evaluations, VLA-Forget improves forgetting efficacy by 10%, preserves perceptual specificity by 22%, retains reasoning and task success by 9%, and reduces post-quantization recovery by 55% relative to strong unlearning baselines.

G-Drift MIA: Membership Inference via Gradient-Induced Feature Drift in LLMs

Large language models (LLMs) are trained on massive web-scale corpora, raising growing concerns about privacy and copyright. Membership inference attacks (MIAs) aim to determine whether a given example was used during training. Existing LLM MIAs largely rely on output probabilities or loss values and often perform only marginally better than random guessing when members and non-members are drawn from the same distribution. We introduce G-Drift MIA, a white-box membership inference method based on gradient-induced feature drift. Given a candidate (x,y), we apply a single targeted gradient-ascent step that increases its loss and measure the resulting changes in internal representations, including logits, hidden-layer activations, and projections onto fixed feature directions, before and after the update. These drift signals are used to train a lightweight logistic classifier that effectively separates members from non-members. Across multiple transformer-based LLMs and datasets derived from realistic MIA benchmarks, G-Drift substantially outperforms confidence-based, perplexity-based, and reference-based attacks. We further show that memorized training samples systematically exhibit smaller and more structured feature drift than non-members, providing a mechanistic link between gradient geometry, representation stability, and memorization. In general, our results demonstrate that small, controlled gradient interventions offer a practical tool for auditing the membership of training-data and assessing privacy risks in LLMs.

CatRAG: Functor-Guided Structural Debiasing with Retrieval Augmentation for Fair LLMs

Large Language Models (LLMs) are deployed in high-stakes settings but can show demographic, gender, and geographic biases that undermine fairness and trust. Prior debiasing methods, including embedding-space projections, prompt-based steering, and causal interventions, often act at a single stage of the pipeline, resulting in incomplete mitigation and brittle utility trade-offs under distribution shifts. We propose CatRAG Debiasing, a dual-pronged framework that integrates functor with Retrieval-Augmented Generation (RAG) guided structural debiasing. The functor component leverages category-theoretic structure to induce a principled, structure-preserving projection that suppresses bias-associated directions in the embedding space while retaining task-relevant semantics. On the Bias Benchmark for Question Answering (BBQ) across three open-source LLMs (Meta Llama-3, OpenAI GPT-OSS, and Google Gemma-3), CatRAG achieves state-of-the-art results, improving accuracy by up to 40% over the corresponding base models and by more than 10% over prior debiasing methods, while reducing bias scores to near zero (from 60% for the base models) across gender, nationality, race, and intersectional subgroups.

Embodied Foundation Models at the Edge: A Survey of Deployment Constraints and Mitigation Strategies

Deploying foundation models in embodied edge systems is fundamentally a systems problem, not just a problem of model compression. Real-time control must operate within strict size, weight, and power constraints, where memory traffic, compute latency, timing variability, and safety margins interact directly. The Deployment Gauntlet organizes these constraints into eight coupled barriers that determine whether embodied foundation models can run reliably in practice. Across representative edge workloads, autoregressive Vision-Language-Action policies are constrained primarily by memory bandwidth, whereas diffusion-based controllers are limited more by compute latency and sustained execution cost. Reliable deployment therefore depends on system-level co-design across memory, scheduling, communication, and model architecture, including decompositions that separate fast control from slower semantic reasoning.

Unrolled Reconstruction with Integrated Super-Resolution for Accelerated 3D LGE MRI

Accelerated 3D late gadolinium enhancement (LGE) MRI requires robust reconstruction methods to recover thin atrial structures from undersampled k-space data. While unrolled model-based networks effectively integrate physics-driven data consistency with learned priors, they operate at the acquired resolution and may fail to fully recover high-frequency detail. We propose a hybrid unrolled reconstruction framework in which an Enhanced Deep Super-Resolution (EDSR) network replaces the proximal operator within each iteration of the optimization loop, enabling joint super-resolution enhancement and data consistency enforcement. The model is trained end-to-end on retrospectively undersampled preclinical 3D LGE datasets and compared against compressed sensing, Model-Based Deep Learning (MoDL), and self-guided Deep Image Prior (DIP) baselines. Across acceleration factors, the proposed method consistently improves PSNR and SSIM over standard unrolled reconstruction and better preserves fine cardiac structures, leading to improved LA (left atrium) segmentation performance. These results demonstrate that integrating super-resolution priors directly within model-based reconstruction provides measurable gains in accelerated 3D LGE MRI.

RAZOR: Ratio-Aware Layer Editing for Targeted Unlearning in Vision Transformers and Diffusion Models

Transformer based diffusion and vision-language models have achieved remarkable success; yet, efficiently removing undesirable or sensitive information without retraining remains a central challenge for model safety and compliance. We introduce Ratio-Aware Zero/One-step Optimized Retentive unlearning (RAZOR), a lightweight, model-agnostic unlearning framework that generalizes forgetting updates to coordinated multi-layer and multi-head edits within transformer backbones. RAZOR identifies the most important layers and attention heads by measuring how much they contribute to forgetting the target data while preserving useful knowledge. Then, it updates these parts of the model using a carefully regularized rule to avoid harming overall performance. The set of edited components grows gradually, ensuring precise unlearning without over-editing or damaging unrelated capabilities. We evaluate RAZOR on CLIP, Stable Diffusion, and vision-language models (VLMs) using widely adopted unlearning benchmarks covering identity, style, and object erasure tasks. Our results show that RAZOR achieves highly accurate and stable forgetting, even under quantization. This approach offers stronger retention and better efficiency than prior methods. Notably, it also operates significant faster than conventional techniques. These results demonstrate that RAZOR is a practical and scalable solution for safe, adaptive unlearning in transformer-based vision models.

HAMIL-QA: Hierarchical Approach to Multiple Instance Learning for Atrial LGE MRI Quality Assessment

The accurate evaluation of left atrial fibrosis via high-quality 3D Late Gadolinium Enhancement (LGE) MRI is crucial for atrial fibrillation management but is hindered by factors like patient movement and imaging variability. The pursuit of automated LGE MRI quality assessment is critical for enhancing diagnostic accuracy, standardizing evaluations, and improving patient outcomes. The deep learning models aimed at automating this process face significant challenges due to the scarcity of expert annotations, high computational costs, and the need to capture subtle diagnostic details in highly variable images. This study introduces HAMIL-QA, a multiple instance learning (MIL) framework, designed to overcome these obstacles. HAMIL-QA employs a hierarchical bag and sub-bag structure that allows for targeted analysis within sub-bags and aggregates insights at the volume level. This hierarchical MIL approach reduces reliance on extensive annotations, lessens computational load, and ensures clinically relevant quality predictions by focusing on diagnostically critical image features. Our experiments show that HAMIL-QA surpasses existing MIL methods and traditional supervised approaches in accuracy, AUROC, and F1-Score on an LGE MRI scan dataset, demonstrating its potential as a scalable solution for LGE MRI quality assessment automation. The code is available at: $\href{https://github.com/arf111/HAMIL-QA}{\text{this https URL}}$