Alignment, interpretability, red-teaming, bias and privacy: the research on what these systems do when they misbehave.
arXiv:2609.27378v1 Announce Type: cross Abstract: End-to-end speech-to-speech dialogue models listen and speak simultaneously, so a continuously open acoustic channel is exposed to adversarial manipu...
arXiv:2609.27572v1 Announce Type: cross Abstract: Reinforcement learning (RL) has emerged as a key paradigm for improving the reasoning capabilities of large language models (LLMs). However, existing...
arXiv:2609.27729v1 Announce Type: cross Abstract: In neuropsychiatry, the primary goal is often not only to decode brain activity but to change it, for example to lessen a negative affective bias or...
arXiv:2609.27816v1 Announce Type: cross Abstract: Safe coordination in heterogeneous machine-to-machine (M2M) robotic systems is challenging when robots differ in sensing capabilities, environmental...
arXiv:2609.28117v1 Announce Type: cross Abstract: In this paper, we introduce a gradient-based head attribution strategy where the Token-level Max-Margin loss is backpropagated to the attention maps....
arXiv:2609.27441v1 Announce Type: new Abstract: Achieving stable long-term neural decoding in invasive brain-machine interfaces (BMIs) remains challenging due to variations in recorded neural populat...
arXiv:2609.27900v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in multi-agent systems where a principal agent decomposes tasks and delegates them to subordinat...
arXiv:2609.27022v1 Announce Type: new Abstract: In this paper, we investigate the impact of adversarial attacks on identity encoders within a realistic de-identification framework. Our experiments sh...
arXiv:2609.27194v1 Announce Type: new Abstract: Prototype-based neural networks provide inherent interpretability through case-based reasoning, yet suffer from critical limitations: prototypes conver...
arXiv:2609.28154v1 Announce Type: new Abstract: Global building and settlement datasets increasingly support population mapping, exposure assessment, urban monitoring, and other analyses of the built...
arXiv:2609.28222v1 Announce Type: new Abstract: Unified 3D vision-language systems must combine complementary geometry, scale, and appearance cues while supporting tasks from instance segmentation to...
arXiv:2609.27560v1 Announce Type: cross Abstract: Avatar-streaming systems are commonly evaluated with image and video quality assessment (IQA/VQA) metrics, implicitly treating visual fidelity as a p...
arXiv:2605.16080v2 Announce Type: replace Abstract: The rise of AI-generated images (AIGIs) poses growing challenges for digital authenticity, prompting the need for efficient, generalizable image fo...
arXiv:2609.27581v1 Announce Type: new Abstract: Step Law gives power-law formulas for the optimal peak learning rate eta* and batch size B* when pre-training language models. It was calibrated on mod...
The paper introduces FSCE, a Target‑Aware Frequency‑Spatial Collaborative Enhancement framework designed to improve noise‑resilient synthetic aperture radar (SAR) automatic target recognition (ATR). It combines a Frequency‑Spatial Early‑stage Adaptive Enhancement module that suppresses speckle noise while preserving target structures, with an Adaptive Policy‑driven Semantic Alignment mechanism that applies top‑down semantic constraints during training. Experiments on MSTAR, OpenSARShip, and FUSARShip datasets show that the lightweight FSCE‑Net_µ, with only 0.17 M parameters, achieves competitive performance across both high‑capacity and lightweight architectures.
The paper introduces a counterfactually anchored evidence attribution approach for multi‑turn large language model safety failures. It presents a new dataset of 1,762 conversations, including adversarial, benign twins, and high‑risk vocabulary variants, and trains a lightweight hierarchical model that accurately predicts safety violations and attributes them to specific user turns and token spans. The model achieves high detection performance (F1 = 0.988) and significantly reduces adversarial confidence when top‑attributed tokens are removed, while maintaining low false‑positive rates on benign conversations.
The paper proves that algorithmic safety verification for Turing‑complete, self‑modifying systems—whether fixed or recursively self‑improving—is fundamentally limited. Statistically, no verifier can be sound, complete, and tractable across unbounded domains, all finite configurations, or succinctly described finite environments, due to Rice’s, Gödel’s, Trakhtenbrot’s, coNP, and PSPACE barriers. Dynamically, even a single self‑modification step can render safety properties undecidable, and no total supervisory algorithm can guarantee correctness for all such transformations, though a monitor that raises alarms on violations remains feasible. whyItMatters":"The results show that formal safety guarantees for recursive self‑improvement are unattainable, highlighting intrinsic verification barriers for advanced AI systems."
Group-based reinforcement learning (RL) methods, such as GRPO and its variants, have become a leading paradigm for training reasoning and agentic large language models (LLMs). While their group-normal...
Recent advances in video object segmentation with Multimodal Large Language Model (MLLM) reasoning have demonstrated the effectiveness of using a single textual token, such as SEG, to predict segmenta...
Mapping the distribution of traffic dynamics at high spatiotemporal resolution is a fundamental question in transportation research. Distributed acoustic sensing (DAS), an innovative seismic observati...