Paper Search
Found 150 papers
-
Chatsparent: An Interactive System for Detecting and Mitigating Cognitive Fatigue in LLMs
Riju Marwah, Vishal Pallagani, Ritvik Garimella, Amit Sheth
November 14, 2025
cs.HC arXiv: 2601.11526LLMs are increasingly being deployed as chatbots, but today's interfaces offer little to no friction: users interact through seamless conversations that conceal
chatsparent cognitive fatigue large language models interactive system detection mitigation auto-regressive generation attention-to-prompt decay embedding drift entropy collapse fatigue index real-time monitoring token-level signals attention reset entropy-regularized decoding self-reflection checkpoints hotpotqa falcon-7b-instruct multi-hop question answering transparency reliability inference time control loop lightweight interventions online statecs.HC cs.AI -
Average-Case Reductions for $k$-XOR and Tensor PCA
Guy Bresler, Alina Harbuzova
January 26, 2026
cs.CC arXiv: 2601.19016We study two canonical planted average-case problems -- noisy $k\mathsf{\text{-}XOR}$ and Tensor PCA -- and relate their computational
average-case reductions k-xor tensor pca planted models computational hardness detection recovery equation combination densifying reductions order-reducing maps computational threshold hardness partial order discrete equation combination gaussian equation combination k-sparse lwe noise distributions gram matrix high-dimensional clt tensor order entry density information-theoretic bounds computational-statistical gap total variation distance planted signal snr deficiencymath.ST math.PR stat.TH cs.CC cs.CR -
MetaLint: Generalizable Idiomatic Code Quality Analysis through Instruction-Following and Easy-to-Hard Generalization
Daniel Fried, Darsh Agrawal, Atharva Naik, Lawanya Baghel, Dhakshin Govindarajan, Carolyn Rose, Yiqing Xie
July 15, 2025
cs.SE arXiv: 2507.11687Large Language Models excel at code generation but struggle with code quality analysis, where best practices evolve and cannot be fully captured by static training data
code quality analysis large language models best practice violations code idioms instruction following preference optimization easy-to-hard generalization synthetic data generation linters ruff pmd tree-sitter python enhancement proposals benchmark detection localization supervised fine-tuning direct preference optimization rejection sampling chain-of-thought reasoning meta-learning generalization contextual reasoning static analysis metalintcs.CL cs.LG cs.SE -
Hide and Seek in Embedding Space: Geometry-based Steganography and Detection in Large Language Models
Charles Westphal, Keivan Navaie, Fernando E. Rosas
January 30, 2026
cs.CR arXiv: 2601.22818Fine-tuned LLMs can covertly encode prompt secrets into outputs via steganographic channels. Prior work demonstrated this
steganography large language models payload recoverability low-recoverability steganography embedding space hyperplane projection geometry-based steganography mechanistic interpretability linear probes trojanstego lora fine-tuning kl divergence perplexity exact match rate vocabulary bucketing secret encoding detection internal signatures model activations wikitext-103 llama models ministral models xor masking text quality metrics classifier accuracycs.AI cs.CR -
Revisiting [CLS] and Patch Token Interaction in Vision Transformers
Alexis Marouani, Oriane Siméoni, Hervé Jégou, Piotr Bojanowski, Huy V. Vo
February 09, 2026
cs.CV arXiv: 2602.08626Vision Transformers have emerged as powerful, scalable and versatile representation learners. To capture both global and local features, a learnable
vision transformer cls token patch tokens token interaction layer specialization normalization layers layernorm layerscale qkv projection self-attention dense prediction semantic segmentation depth estimation image classification detection dino v2 deit iii attention bias register tokens image pre-training linear probing cosine similarity analysis architectural modification feature disentanglement imagenetcs.CV -
A Comprehensive Benchmark for Electrocardiogram Time-Series
Zhijiang Tang, Jiaxin Qi, Yuhua Zheng, Jianqiang Huang
July 15, 2025
eess.SP arXiv: 2507.14206Electrocardiogram~(ECG), a key bioelectrical time-series signal, is crucial for assessing cardiac health and diagnosing various diseases
electrocardiogram time-series benchmark deep learning cardiac health disease diagnosis classification detection forecasting generation feature-based fréchet distance mean squared error patch step-by-step model hierarchical architecture large time-series models transformer convblock quasi-periodic signals mit-bih arrhythmia database ptbxl cpsc2018 fetal ecg semantic fidelity waveform localization pre-trainingstat.ML cs.LG cs.AI eess.SP -
Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectives
Chloe Li, Mary Phuong, Daniel Tan
November 10, 2025
cs.AI arXiv: 2511.06626As AI systems become more capable of complex agentic tasks, they also become more capable of pursuing undesirable objectives and causing harm
large language models ai alignment ai safety self-report fine-tuning srft supervised fine-tuning model auditing interrogation hidden objectives misaligned objectives honesty propensity out-of-distribution generalization stealth tasks sabotage agentic tasks self-reporting deception adversarial pressure decoy objectives hiddenagenda shade-arena elicitation detection prefilled assistant turn attacks model introspectioncs.AI -
Detection of local geometry in random graphs: information-theoretic and computational limits
Jinho Bok, Shuangping Li, Sophie H. Yu
March 25, 2026
math.ST arXiv: 2603.24545We study the problem of detecting local geometry in random graphs. We introduce a model $\mathcal{G}(n, p, d, k)$, where a hidden community of average size $k$ has edges drawn as a
random graphs local geometry detection information-theoretic limits computational limits random geometric graph erdős-rényi model hidden community signed triangle count global test scan test constrained scan test truncated second moment wishart-goe comparison tensorization of kl divergence low-degree polynomial framework signed cycle counts computational-statistical gap high-dimensional random geometric graphs planted subgraph signal beyond mean latent space community detection hypothesis testing phase transitionstat.ML math.ST math.PR stat.TH cs.DS cs.CC -
Focus-to-Perceive Representation Learning: A Cognition-Inspired Hierarchical Framework for Endoscopic Video Analysis
Yuan Zhang, Sihao Dou, Kai Hu, Shuhua Deng, Chunhong Cao, Fen Xiao, Xieping Gao
March 26, 2026
cs.CV arXiv: 2603.25778Endoscopic video analysis is essential for early gastrointestinal screening but remains hindered by limited high-quality annotations. While self-supervised video pre
focus-to-perceive representation learning endoscopic video analysis self-supervised learning hierarchical framework static semantics contextual semantics teacher-prior adaptive masking multi-view sparse sampling cross-view masked feature completion attention-guided temporal prediction motion bias mamba teacher-student cognition-inspired masked modeling contrastive learning feature alignment pixel reconstruction polypdiag dataset cvc-12k dataset kumc dataset cholec80 dataset classification segmentation detectioncs.CV -
vGamba: Attentive State Space Bottleneck for efficient Long-range Dependencies in Visual Recognition
Yunusa Haruna, Adamu Lawan, Shamsuddeen Hassan Muhammad, Jiaquan Zhang, Chaoning Zhang
March 27, 2025
cs.CV arXiv: 2503.21262Capturing long-range dependencies (LRD) efficiently is a core challenge in visual recognition, and state-space models (SSMs) have recently emerged as
vgamba gamba cell asc module state-space models long-range dependencies visual recognition bottleneck hybrid vision backbone cnn attention mechanism mamba vmamba vim botnet imagenet-1k ade20k coco classification segmentation detection high-resolution memory efficiency linear complexity 2d relative positional embeddings efficiency-accuracy trade-offcs.CV -
Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions
Yifei Dong, Fengyi Wu, Sanjian Zhang, Guangyu Chen, Yuzhi Hu, Masumi Yano, Jingdong Sun, Siyu Huang, Feng Liu, Qi Dai, Zhi-Qi Cheng
April 16, 2025
cs.CV arXiv: 2504.11967Unmanned Aerial Vehicles (UAVs) are indispensable for infrastructure inspection, surveillance, and related tasks, yet they also introduce critical security challenges
anti-uav unmanned aerial vehicles classification detection tracking diffusion-based data synthesis multi-modal fusion vision-language modeling self-supervised learning reinforcement learning rgb infrared audio radar rf benchmarks datasets real-time performance stealth detection swarm-based scenarios domain adaptation small-object detection adversarial robustness sensor fusion deep learningcs.AI cs.RO cs.CV -
TriDF: Evaluating Perception, Detection, and Hallucination for Interpretable DeepFake Detection
Jian-Yu Jiang-Lin, Kang-Yang Huang, Ling Zou, Ling Lo, Sheng-Ping Yang, Yu-Wen Tseng, Kun-Hsiang Lin, Chia-Ling Chen, Yu-Ting Ta, Yan-Tsung Wang, Po-Ching Chen, Hongxia Xie, Hong-Han Shuai, Wen-Huang Cheng
December 11, 2025
cs.CV arXiv: 2512.10652Advances in generative modeling have made it increasingly easy to fabricate realistic portrayals of individuals, creating serious risks for security
trif interpretable deepfake detection perception detection hallucination multimodal large language models deepfake types image modality video modality audio modality fine-grained artifacts quality artifacts semantic artifacts artifact taxonomy human-annotated evidence localization explanation reliability cover metric chair metric hal metric f0.5 score true-false questions multiple-choice questions open-ended questions artifact mappingcs.CV cs.CR -
Squish and Release: Exposing Hidden Hallucinations by Making Them Surface as Safety Signals
Nathaniel Oh, Paul Attie
March 27, 2026
cs.LG arXiv: 2603.26829Language models detect false premises when asked directly but absorb them under conversational pressure, producing authoritative professional output built on errors they already
squish and release activation patching order-gap hallucination order-gap benchmark detector body detector core safety core absorb core cascade collapse layer localization synthetic core epistemic specificity hallucination detection conversational pressure false premise compliance detection residual stream activation vector core engineering model-agnostic olmo-2 7b anchor prompt bidirectional control attractor asymmetrycs.LG cs.AI -
Large Language Models in the Abuse Detection Pipeline
Sanket Badhe, Shivani Gupta, Suraj Kath, Preet Shah, Ashwin Sampathkumar
March 31, 2026
cs.CL arXiv: 2604.00323Online abuse has grown increasingly complex, spanning toxic language, harassment, manipulation, and fraudulent behavior. Traditional machine-learning approaches dependent on static
large language models abuse detection lifecycle label and feature generation detection review and appeals auditing and governance synthetic data generation zero-shot classification few-shot classification fine-tuned models embedding-based detection contrastive learning retrieval-augmented generation graph neural networks chain-of-thought prompting multi-agent debate explanation generation human-in-the-loop fairness and bias adversarial robustness jailbreaks over-refusal production gap safety paradox platform safetycs.CL cs.CY -
The Rise of Language Models in Mining Software Repositories: A Survey
Sergio Segura, Miguel Romero-Arjona, Saman Barakat, Ana B. Sánchez
April 01, 2026
cs.SE arXiv: 2604.00787The Mining Software Repositories (MSR) field focuses on analysing the rich data contained in software repositories to derive
mining software repositories language models survey transformer bert gpt classification generation extraction detection fine-tuning prompting github issue reports code reviews source code reproducibility taxonomy encoder-only decoder-only encoder-decoder large language models software engineering datasets toolscs.SE -
Near-Optimal Four-Cycle Counting in Graph Streams
Stefan Neumann, Pan Peng, Sebastian Lüderssen
April 01, 2026
cs.DS arXiv: 2604.00828We study four-cycle counting in arbitrary order graph streams. We present a 3-pass algorithm for $(1+\varepsilon)$-approximating the number of four-cycles
four-cycle counting graph streams streaming algorithms subgraph counting multi-pass algorithms node sampling edge sampling heavy substructures configurations variance analysis approximation algorithms space complexity lower bounds motifs arbitrary order streams detection counting heaviness thresholds onions wedges oracles llm configurations refined heaviness chebyshev inequality chernoff boundcs.DS -
StormShield: Fingerprint-Based Detection and Mitigation of RRC Signaling Storms in O-RAN 5G RANs
Noemi Giustini, Andrea Lacava, Leonardo Bonati, Stefano Maxenti, Michele Polese, Tommaso Melodia, Francesca Cuomo
May 13, 2026
cs.NI arXiv: 2605.140325G networks provide low-latency, high throughput, and massive connectivity, yet the control plane remains exposed to several security threats. Among
5g o-ran security denial of service rrc signaling storm stormshield fingerprint detection mitigation xapp near-rt ric openairinterface timing advance received signal strength indicator dbscan block list aging timeout control plane gnb malicious ue pre-authentication ota testbed functional split key performance metrics e2 interfacecs.NI cs.CR -
ReMIA: a Powerful and Efficient Alternative to Membership Inference Attacks against Synthetic Data Generators
Davide Scassola, Andrea Coser, Sebastiano Saccani
May 14, 2026
cs.LG arXiv: 2605.14686Tabular data sharing under privacy constraints is increasingly important for research and collaboration. Synthetic data generators (SDGs) are a promising
remia membership inference attack synthetic data generator privacy metric shadow modeling distance to closest record tabular data generative model differential privacy domias smmia classifier auroc privacy-utility trade-off noise-based anonymization overfitting attacker model data efficiency computational efficiency empirical evaluation privacy score detection ml efficacy risk model leaky modelcs.LG -
FRAME: Forensic Routing and Adaptive Multi-path Evidence Fusion for Image Manipulation Detection
Kaixiang Zhao, Tianrun Yu, Aoxu Zhang, Junhao Su, Porter Jenkins, Amanda Hughes
May 12, 2026
cs.CV arXiv: 2605.12826The proliferation of sophisticated image editing tools and generative artificial intelligence models has made verifying the authenticity of digital images increasingly
image manipulation detection digital forensics adaptive routing multi-path evidence fusion forensic supernet graph neural networks path selection evidence fusion error level analysis discrete cosine transform sensor noise analysis color filter array copy-move detection pyifd casia v1 casia v2 coverage dataset realistic tampering localization detection interpretable forensics heterogeneous algorithms context-adaptive deep learning image authenticitycs.AI cs.CV -
GUITestScape: Towards Open-set Evaluation on Exploratory GUI Testing
Xiaoyi Chen, Yifei Gao, Yang Xu, Xingxing Song, Yi Zhang, Jitao Sang
May 28, 2026
cs.SE arXiv: 2605.29532Exploratory GUI testing is a particularly demanding setting for MLLM agents: without predefined test scripts, an agent must
exploratory gui testing mllm agents android applications guitestscape guijudge display defects interaction defects open-set evaluation process-aware evaluation multimodal large language models software quality assurance visual anomaly detection trajectory retrieval reaching triggering detection content rendering defects element layout defects operation no response unexpected task result navigation logic error state-level test basis resettable scenarios benchmark automated testingcs.AI cs.SE