adversarial-robustness

Tag

Cards List
#adversarial-robustness

Learning to Verify: Signature-Guided Training for Adaptive Adversarial Detection

arXiv cs.LG ↗ · 4h ago Cached

The paper proposes a signature-guided dual-model verifier that keeps the classifier fixed while training a separate verifier using activation signatures to detect adversarial examples under adaptive attacks, outperforming confidence- and feature-based detectors on L∞ attacks but showing limitations on L2 perturbations.

0 favorites 0 likes
#adversarial-robustness

On the Impact of Feature Reduction Techniques on Adversarial Attacks

arXiv cs.LG ↗ · 4h ago Cached

This paper analyzes how feature reduction techniques (PCA, permutation importance ranking, recursive feature elimination, univariate mutual information) affect the adversarial robustness of classical ML classifiers under poisoning, backdoor, and universal adversarial perturbation attacks, finding permutation importance ranking and univariate mutual information give the most consistent robustness gains.

0 favorites 0 likes
#adversarial-robustness

Helping AI text detectors keep up with changing conditions

MIT News — Artificial Intelligence ↗ · 2d ago Cached

MIT and Cornell researchers show commercial AI text detectors can be easily fooled by disguised text or natural distribution shifts, and demonstrate a test-time adaptation training technique that boosts accuracy on disguised AI text from 27% to 93%, to be presented at NeurIPS.

0 favorites 0 likes
#adversarial-robustness

Representational Geometry as a Diagnostic for Adversarial Robustness

arXiv cs.LG ↗ · 4d ago Cached

A NeurIPS 2026 workshop paper showing that representational geometry, visualized with M-PHATE, can serve as a diagnostic for adversarial robustness, revealing that robust accuracy and representational stability can rank defenses in opposite ways.

0 favorites 0 likes
#adversarial-robustness

Fair and Efficient Distributed EV Charging Coordination Using Threshold Cryptography and Machine Learning

arXiv cs.LG ↗ · 4d ago Cached

The paper proves that Byzantine attacks on ADMM-based distributed EV charging remain undetected while preserving convergence, then proposes a threshold-signed broadcast with upload recomputation defense that detects deviations at linear communication cost, using fleet-aware ML price forecasts for local recovery. Simulations show the defense limits attack-induced cost distortions from up to 132.7% down to under 2.6%.

0 favorites 0 likes
#adversarial-robustness

Towards Adversarially Robust Multigraph Neural Networks for Financial Crime Analysis

arXiv cs.LG ↗ · 2026-10-27 Cached

Delft University of Technology researchers present the first scalable framework for structural adversarial attacks and adversarial training on multigraph neural networks used for anti-money laundering detection, introducing a cycle-based regularizer to keep perturbations plausible and showing that adversarial training restores detection robustness.

0 favorites 0 likes
#adversarial-robustness

Knowing Whom to Trust is not Enough: The Trust--Agreement Gap in Repeated Multi-Party LLM Negotiation

arXiv cs.AI ↗ · 2026-10-26

研究引入重复多边多方多议题谈判框架,考察LLM智能体能否从历史交互中构建信任估计并用于改进后续谈判结果,发现信任感知智能体能在几轮内识别对手,但准确的信任估计并不总能转化为更好的协议,只有在显式引导智能体使用信任时,协议率才显著提升(最高23个百分点),揭示了"信任-协议鸿沟"。

0 favorites 0 likes
#adversarial-robustness

How Much Do You Trust A Dog? Quantifying Hint-Following in Reasoning Language Models

arXiv cs.CL ↗ · 2026-10-26

This paper quantifies how hints attributed to sources of varying reliability (professor, friend, or dog) shift answers in reasoning LLMs like Qwen, Gemma, and SmolLM, finding the shift follows an exponential tilt pattern that persists even when the source is declared untrustworthy.

0 favorites 0 likes
#adversarial-robustness

Learning-Theoretic Foundation for General Coded Computing: The Adversarial Setting

arXiv cs.LG ↗ · 2026-10-21 Cached

论文提出一种对抗鲁棒的广义编码计算(GCC)扩展框架,通过平滑正则化限制恶意worker输出的影响,并给出随对抗worker比例衰减的非渐近误差界,适用于分布式系统中的近似编码计算。

0 favorites 0 likes
#adversarial-robustness

Vulnerability Analysis for Safe Multi-Agent Reinforcement

arXiv cs.AI ↗ · 2026-10-21 Cached

该论文研究了针对安全多智能体强化学习 (SMARL) 的推理时观测攻击,提出了 MinMax Attack 框架,能在保持任务奖励的同时显著提高安全代价,从而兼具攻击有效性与隐蔽性。

0 favorites 0 likes
#adversarial-robustness

SafeDriveVLA: Navigation-Conditioned World Model Dreaming for Conflict-Aware End-to-End Autonomous Driving

Hugging Face Daily Papers ↗ · 2026-10-14 Cached

SafeDriveVLA 提出首个同时评估导航遵循与冲突感知的基准,并通过导航条件化世界模型 dreaming 与三级别导航合规模式预测,提升端到端自动驾驶 VLA 在安全冲突指令下的鲁棒性,在 Bench2Drive 上达到 88.16 驾驶分数。

0 favorites 0 likes
#adversarial-robustness

Who to Trust in Multi-Robot Systems: Joint Robot and Object Hypothesis Trust Inference via Neural-Symbolic Cross-Validation

Hugging Face Daily Papers ↗ · 2026-10-10 Cached

NeST-Bayes是用于多机器人系统的神经-符号信任推断框架,仅凭跨机器人观测一致性联合推断机器人与物体假设的信任度,无需预设信任关系,并在对抗攻击与遮挡环境下取得高精度和召回率。

0 favorites 0 likes
#adversarial-robustness

Adversarial Cues in Decision Models Used as Judges: The Role of Request Presentation

arXiv cs.CL ↗ · 2026-10-09 Cached

This paper shows that a single-character colon edit in a candidate answer can sharply increase a decision model judge's false acceptance rate (from 1% to 26%) depending on how the structured judging request is presented, demonstrating that judging competence can coexist with presentation-dependent vulnerabilities.

0 favorites 0 likes
#adversarial-robustness

How Hackable Is Your Speech Quality Metric? A Corrected Protocol, a Benchmark, and What Patching Buys

arXiv cs.LG ↗ · 2026-10-09 Cached

该论文提出了一个修正的语音质量预测器可攻击性测量协议(包括正确的神经编解码器往返参照和多种子最坏情况评估),并在此基准下发现NISQA、SSL-MOS、DNSMOS和UTMOS在可攻击性上差异巨大(90%至6%),同时揭示了封闭式攻击-检测-补丁循环的局限性。

0 favorites 0 likes
#adversarial-robustness

Breaking Adversarial Transferability in Fine-Tuned Speech Recognition

arXiv cs.LG ↗ · 2026-10-08 Cached

该论文揭示了针对公开ASR基础模型生成的对抗扰动可有效迁移至微调后黑盒模型的风险,并提出TransferBreaker统一微调框架(结合Base Adversarial Fine-Tuning、Latent Jacobian Regularization与HybridGrad-AFT),在三种语言和四个大型ASR模型上将对抗WER从92.6降至27.8。

0 favorites 0 likes
#adversarial-robustness

Ontological Instability and Statistical Amplification: The Paradox of "Humanizing" LLM-Generated Text

arXiv cs.CL ↗ · 2026-10-05 Cached

This arXiv paper shows that RoBERTa-based machine-text detectors track statistical complexity rather than machine authorship: "humanizing" LLM outputs via Mistral-7B actually raises Verb Diversity and detection scores, while the detector exhibits a 76.3% false-positive rate on formal human writing and is evaded by 41% of homoglyph attacks. Event-based latent-space detection proved fragile to paraphrasing and character-level perturbations.

0 favorites 0 likes
#adversarial-robustness

MACTS-EM: Multi-Agent Collaborative Time Series Forecasting with Emergent Memory

arXiv cs.LG ↗ · 2026-10-05 Cached

The paper introduces MACTS-EM, a multi-agent collaborative framework for time series forecasting that uses specialised agents, a meta-cognitive layer, emergent memory, and multimodal integration, reporting 8-12% accuracy gains and improved zero-shot transfer and regime-shift resilience across financial, climate, energy, and pandemic datasets.

0 favorites 0 likes
#adversarial-robustness

Certification of Real Images through Calibrated Content Authentication

Hugging Face Daily Papers ↗ · 2026-10-05 Cached

The paper evaluates twenty deepfake detectors against ten generators, finding detection accuracy degrading from 99.5% to 76% over time and collapsing below 2% under adversarial perturbations, then proposes a calibrated detection paradigm based on plausibly deniable authenticity via generator reconstruction, achieving at most 1% false certification.

0 favorites 0 likes
#adversarial-robustness

Backdoor Containment via Expert Quarantine and Shutdown in LLMs

arXiv cs.AI ↗ · 2026-10-02 Cached

The paper proposes Quarantined Expert Shutdown (QES), a 'learn, but channel' backdoor defense that routes trigger-conditioned behavior into a quarantined expert via routed LoRA branches, allowing deployment-time mitigation by simply zeroing that expert's routing weight. Results show attack success rates drop from 100% to 0-10% across two tasks, three attacks, and four model families while largely preserving benign utility.

0 favorites 0 likes
#adversarial-robustness

Robust Is Salient: An Informed Adversary Moves the Optimal Signal onto the Salience Pole

arXiv cs.AI ↗ · 2026-10-02 Cached

This paper adds an informed adversary to a forced-choice signalling game, showing the optimal signal shifts from posterior-maximizing to margin-maximizing as persuasion budget grows, and that the robust optimum coincides with salience on all 108 confirmatory items — a structural limit on what evaluations of adversary-awareness can learn.

0 favorites 0 likes
Next →
← Back to home

Submit Feedback