The paper introduces Q-MINO, a Quantization-Aware Minimal-Norm Optimizer designed to improve training of ultra-low-bit neural networks. Q-MINO uses a temporal bundle method that incorporates gradient consensus, state-drift regularization, and an alignment constraint to produce stabilized, minimum-norm update directions. The authors solve the resulting constrained subproblem with a warm-started Frank–Wolfe procedure and provide theoretical convergence guarantees via a stochastic Lyapunov Kurdyka–Łojasiewicz framework, along with numerical experiments demonstrating its effectiveness across various quantization levels.
By Don Li
The paper introduces LLMLeak, a covert exfiltration technique that exploits Large‑Language‑Model (LLM) web‑fetching capabilities to transmit confidential data. By embedding a secret into a URL and presenting the corresponding website as part of a legitimate task, the malicious client tricks the LLM into fetching the URL, thereby leaking the secret to an attacker‑controlled server. Experiments on eleven open‑parameter models show a 79.7% success rate, and a real‑world case study confirms the attack’s practicality.
By Alessandro Pegoraro, Daryan Merx, Phillip Rieger, Ahmad-Reza Sadeghi
Attention Manifolds introduce learned 2D B-spline surfaces that modulate each value dimension in transformer attention based on query-key interactions. Applied to LLaMA models, they reduce WikiText-2 perplexity by 2–2.5 points with minimal parameter overhead and alter greedy-decoded outputs in 69–83% of diverse prompts, especially on ambiguous inputs. The surfaces can be edited mechanically to steer or block value flow, offering a geometric tool for improving output quality and potentially enhancing safety.
By Naveen Mysore
The paper introduces VANE, a new routing strategy for Mixture-of-LoRA-experts that scores experts based on the expected loss reduction of their updates rather than token similarity. By decomposing each expert into a reader and writer, VANE evaluates all reader–writer pairs using a low‑rank compass that predicts the descent direction, enabling efficient top‑k selection with additive gates. Experiments on Llama‑3 models show VANE outperforms twelve PEFT and MoE‑LoRA baselines while using fewer trainable parameters and providing router scores that better reflect expert usefulness.
By Priya Nair, Lukas Brenner, Maya Lindqvist, Daniel Whitmore, Wen-Hsuan Liu, Tom Saliencro, Amara Okonkwo, Rohan Desai
The paper introduces CLEAR, a lightweight, task‑agnostic post‑hoc method that enhances evidential robustness in deep learning models without retraining. CLEAR uses held‑out calibration data to map the geometry of the model’s latent space, then generates perturbation views at inference to detect latent conflict. When high conflict is found, CLEAR selectively reduces evidential strength while preserving evidence for latent‑consistent inputs, achieving significant improvements in OOD and adversarial AUROC on ImageNet→CUB and running much faster than competing methods.
By Charmaine Barker, Daniel Bethell, Simos Gerasimou
The paper introduces CRAFT, a method for improving compositional generalization in vision‑language‑action (VLA) models. It addresses the issue where models rely on visual shortcuts during fine‑tuning, leading them to execute demonstrated skill combinations that match observations rather than the instructed ones. By training with counterfactual instruction–observation pairs and transferring supervision through reusable skill representations, CRAFT enhances success on unseen skill combinations while preserving performance on demonstrated ones across multiple VLA models and benchmarks.
By Taegeun Yang, Youngju Na, Yoonki Cho, Sung-Eui Yoon
The paper investigates quantum kernel methods for detecting audio deepfakes when training data are scarce and evaluation corpora differ from training corpora. Using only 200 labeled samples and four‑dimensional embeddings, the authors compare a Quantum Support Vector Machine (QSVM), a classical SVM, and a multilayer perceptron (MLP) across several cross‑corpus datasets. While the QSVM outperforms the MLP under severe domain shift from ASVspoof 2019 to ADD 2023, its advantage is inconsistent in other transfer directions, indicating that quantum kernels can be competitive but not universally superior in low‑resource, cross‑corpus settings.
By Lisan Al Amin, Lei Zhang, Vandana P. Janeja
The paper investigates how reinforcement learning agents trained in one cyber simulation can be transferred to other simulators or real environments. It introduces a framework that decouples state alignment from action translation, allowing zero‑shot policy transfer without retraining. Experiments across four cyber platforms show that transferred policies can preserve performance in closely aligned settings and achieve substantial win rates in more divergent environments.
By Sabrina Saika, Yinuo Du, Aritran Piplai
VERITYGATE is a four‑gate framework that checks whether LLM‑generated narrations adhere to a fixed schema of declared evidence IDs, entities, numbers, and claim types, rather than verifying every fact in the prose. In experiments with GPT‑4o‑mini, Llama‑3.3‑70B, and Claude Sonnet 4.6 on 900 grounded‑ungrounded pairs, the framework identified high failure rates (up to 80.3% for mini claims) and demonstrated that a single repair pass can improve claim survival rates. The authors also provide code, data, and preliminary human studies to validate the rules and highlight gaps between schema compliance and correct prose.
"whyItMatters":"The framework offers a systematic way to evaluate and improve the faithfulness of LLM explanations to structured evidence, revealing significant failure rates and the impact of repair strategies."
By Sachin Gupta
The paper compares distributed adversarial training algorithms—both centralized and decentralized—within multi‑agent learning environments. It introduces a theoretical framework to analyze how efficiently these algorithms escape local minima, a property linked to model flatness and robustness. The study finds that with small perturbation bounds and large batch sizes, decentralized methods (consensus and diffusion) escape local minima faster than centralized ones, but this advantage may diminish as attack strength increases.
By Ying Cao, Kun Yuan, Ali H. Sayed
The paper introduces a framework for aligning the inductive bias of linear time‑invariant State Space Models (SSMs) with task‑specific spectral characteristics. By formalizing the bias through an SSM‑induced kernel and showing its spectrum is governed by the model’s frequency response, the authors propose Task‑Dependent Initialization (TDI), a fast power‑spectrum matching method. Experiments on synthetic data, one‑layer SSMs, and deep SSMs across real‑world benchmarks demonstrate that TDI improves data‑efficient generalization when the task’s spectral structure differs from the default SSM bias.
By Qiyu Chen, Guozhang Chen
The paper introduces the Alignment Flywheel, a governance‑centric hybrid multi‑agent system (MAS) that separates decision generation from safety governance. It defines a Proposer that generates candidate trajectories, a Safety Oracle stack that evaluates safety, and an Enforcement layer that applies risk policies at runtime. A governance MAS oversees monitoring, red‑teaming, verification, and versioned release management, enabling patch‑local fixes to safety failures without retraining the Proposer. The architecture is implementation‑agnostic and is demonstrated in two scenarios: a learned spatial Oracle and a clinical GenAI proxy. The authors provide open‑source code at https://github.com/decide-ugent/Alignment-Flywheel.
By Elias Malomgr\'e, Pieter Simoens
The paper proposes a multi‑objective framework that integrates reinforcement learning‑based multiple instance learning (RL‑MIL), adversarial debiasing, and preference‑conditioned hypernetworks to predict student performance while addressing fairness concerns. Each student is represented as a bag of weakly labeled interactions, and an RL agent selects informative instances for classification. Although the baseline RL‑MIL achieves strong accuracy, the hypernetwork extensions fail to produce a smooth trade‑off between predictive performance and Equalized Odds due to mode collapse, objective dominance, and weak gradient propagation, indicating that additional mechanisms are needed for stable, fair RL‑MIL systems.
By Bente Hinkenhuis, Seyed Sahand Mohammadi Ziabari, Ali Mohammed Mansoor Alsahag
The paper introduces a reliability‑aware short‑term roll prediction framework for unmanned surface vehicles (USVs) that combines a multi‑task learning architecture with an adaptive centralization strategy. The model uses a shared backbone to feed a regression head for precise roll prediction and a quantification head for confidence scoring, enabling accurate predictions alongside reliability estimates. Experiments on a real‑sea dataset show that the approach effectively quantifies prediction reliability and generalizes well across varying operational conditions.
By Kaizhen Li, Xi Zhou, Zihao Wang, Dan Zhang, Jianjian Liu, Xiaowei Li
FedFit introduces a federated fine‑tuning framework for large language models that reduces communication overhead by using a disjoint shared vector‑bank parameterization to reconstruct adapter matrices from two compact global vector banks. It resolves the aggregation dilemma between Sum‑of‑Products and Product‑of‑Sums through an alternating optimization schedule that alternates between accurate single‑bank updates and joint updates corrected by a Residual Spectral Aggregation mechanism. The method also incorporates blockwise quantization with client‑side error feedback and provides theoretical convergence guarantees, achieving perplexity comparable to standard federated LoRA while delivering up to 100× higher compression ratios on Qwen2.5 models.
By Hang Zou, Chao Zhang, Yuzhi Yang, Yu Tian, Samson Lasaulce, M\'erouane Debbah
arXiv:2610. 00545v1 Announce Type: new Abstract: We study adversarial online maximization of nonnegative, non-monotone DR-submodular functions over compact convex down-closed sets.
By Vaneet Aggarwal
MoRA is a framework for pruning Mixture-of-Experts (MoE) models by learning a router bias for each expert and optimizing it with a language‑modeling loss and a routing‑diversity regularizer. The learned biases sharpen routing distributions to identify critical experts and encourage diverse routing preferences. After pruning, MoRA uses an expert approximation mechanism that approximates the outputs of pruned experts with affine transformations of remaining experts, further improving performance.
By Yushuai Sun, Zikun Zhou, Lin Gao, Jun Yu, Wenjie Pei
The paper investigates whether response safety can be measured by the cosine similarity between a response embedding and the mean embedding of known‑safe responses. Using four frozen encoders and prompt‑controlled datasets, the authors find that a simple prototype (mean safe embedding) performs poorly (ROC‑AUC 0.457‑0.545) while an explicit safe‑minus‑unsafe reference achieves higher scores (0.588‑0.738). The study shows that a class mean is merely a location, not a safety direction, and that a reference with sufficient unsafe mass is needed to orient safety judgments.
By Sahil Kadadekar
BranchIP introduces a single-model framework that learns adaptive tensor product computation for equivariant machine learning interatomic potentials (MLIPs). Using a novel distillation loss, it achieves up to 2.4× speed‑up and 2.6× memory reduction across model sizes while preserving physical fidelity. The adaptive computation also offers interpretability by indicating which interactions require deeper processing and how depth correlates with chemical complexity and dynamics.
By Laura Zichi, Gil Harari, Chuin Wei Tan, Marc L. Descoteaux, Albert Zhu, Menghang Wang, Yoel Zimmermann, H. T. Kung, Boris Kozinsky
The paper investigates how fine‑tuning large language models with a small number of harmful examples can erode their refusal behavior, and explores whether localizing safety‑related behavior to specific layers or directions can provide robust defenses. Experiments across six checkpoints from four model families show that harmful and benign prompts remain linearly separable after attack, and that patching clean hidden states or freezing layers up to a transition depth can restore refusal. However, attackers can bypass these defenses by spreading updates or targeting singular directions, indicating that adaptive fine‑tuning can defeat localized repairs and highlighting the need for multiple defensive checks.
By Jungseob Lee, Dongyub Jude Lee, Sugyeong Eo, Seongtae Hong, Seungyoon Lee, Heuiseok Lim