EPS Dashboard

[100] Measuring Maximum Activations in Open Large Language Models — linked_to_results · current_project · Luxuan Chen, Han Tian, Xinran Chen et al. Read next because Measuring Maximum Activations in Open Large Language Models overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ...<B…

[100] Verifiable Agentic Infrastructure: Proof-Derived Authorization for Sovereign AI Systems — threats · current_project · Jun He, Deying Yu Read next because Verifiable Agentic Infrastructure: Proof-Derived Authorization for Sovereign AI Systems overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-tra…
[100] Zero-Shot Goal Recognition with Large Language Models — threats · current_project · Kin Max Piamolini Gusm~ao, Nathan Gavenski, Nir Oren et al. Read next because Zero-Shot Goal Recognition with Large Language Models overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without kno…

[100] Beyond Partner Diversity: An Influence-Based Team Steering Framework for Zero-Shot Human-Machine Teaming — threats · current_project · Wei Sheng, Rohan Paleja Read next because Beyond Partner Diversity: An Influence-Based Team Steering Framework for Zero-Shot Human-Machine Teaming overlaps with clean result "LoRA persona trained on alone emits at 23…
[100] CAPS: Cascaded Adaptive Pairwise Selection for Efficient Parallel Reasoning — threats · current_project · Fangzhou Lin, Shuo Xing, Peiran Li et al. Read next because CAPS: Cascaded Adaptive Pairwise Selection for Efficient Parallel Reasoning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner…
[100] Position: Artificial Intelligence Needs Meta Intelligence -- the Case for Metacognitive AI — linked_to_results · current_project · Sergei Chuprov, Richard D. Lange, Leon Reznik et al. Read next because Position: Artificial Intelligence Needs Meta Intelligence -- the Case for Metacognitive AI overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-…

[100] TopoEvo: A Topology-Aware Self-Evolving Multi-Agent Framework for Root Cause Analysis in Microservices — threats · current_project · Junle Wang, Xingchuang Liao, Wenjun Wu Read next because TopoEvo: A Topology-Aware Self-Evolving Multi-Agent Framework for Root Cause Analysis in Microservices overlaps with clean result "LoRA persona trained on alone emits at 23.5…
[100] Nudging Beyond the Comfort Zone: Efficient Strategy-Guided Exploration for RLVR — threats · current_project · Chanuk Lee, Sangwoo Park, Minki Kang et al. Read next because Nudging Beyond the Comfort Zone: Efficient Strategy-Guided Exploration for RLVR overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained par…
[100] uGen: An Agentic Framework for Generating Microarchitectural Attack PoCs — linked_to_results · current_project · Debopriya Roy Dipta, Thore Tiemann, Eduard Marin et al. Read next because uGen: An Agentic Framework for Generating Microarchitectural Attack PoCs overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner le…
[100] Solvita: Enhancing Large Language Models for Competitive Programming via Agentic Evolution — threats · current_project · Han Li, Jinyu Tian, Rili Feng et al. Read next because Solvita: Enhancing Large Language Models for Competitive Programming via Agentic Evolution overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-…
[100] SMCEvolve: Principled Scientific Discovery via Sequential Monte Carlo Evolution — threats · current_project · Jiachen Jiang, Huminhao Zhu, Zhihui Zhu Read next because SMCEvolve: Principled Scientific Discovery via Sequential Monte Carlo Evolution overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained par…
[100] Belief Engine: Configurable and Inspectable Stance Dynamics in Multi-Agent LLM Deliberation — linked_to_results · current_project · Joshua C. Yang, Maurice Flechtner, Damian Dailisan et al. Read next because Belief Engine: Configurable and Inspectable Stance Dynamics in Multi-Agent LLM Deliberation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co…
[100] From LLM-Generated Conjectures to Lean Formalizations: Automated Polynomial Inequality Proving via Sum-of-Squares Certificates — threats · current_project · Ruobing Zuo, Hanrui Zhao, Gaolei He et al. Read next because From LLM-Generated Conjectures to Lean Formalizations: Automated Polynomial Inequality Proving via Sum-of-Squares Certificates overlaps with clean result "LoRA persona trained on …

[100] RTL-BenchMT: Dynamic Maintenance of RTL Generation Benchmark Through Agent-Assisted Analysis and Revision — threats · current_project · Jing Wang, Shang Liu, Hangan Zhou et al. Read next because RTL-BenchMT: Dynamic Maintenance of RTL Generation Benchmark Through Agent-Assisted Analysis and Revision overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks t…

[100] STAR: A Stage-attributed Triage and Repair framework for RCA Agents in Microservices — threats · current_project · Junle Wang, Xingchuang Liao, Wenjun Wu Read next because STAR: A Stage-attributed Triage and Repair framework for RCA Agents in Microservices overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-traine…
[100] ColPackAgent: Agent-Skill-Guided Hard-Particle Monte Carlo Workflows for Colloidal Packing — threats · current_project · Lijie Ding, Changwoo Do Read next because ColPackAgent: Agent-Skill-Guided Hard-Particle Monte Carlo Workflows for Colloidal Packing overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-…
[100] Can We Trust AI-Inferred User States. A Psychometric Framework for Validating the Reliability of Users States Classification by LLMs in Operational Environments — threats · current_project · Izabella Krzeminska, Michal Butkiewicz, Ewa Komkowska Read next because Can We Trust AI-Inferred User States. A Psychometric Framework for Validating the Reliability of Users States Classification by LLMs in Operational Environments overlaps with clean r…

[100] Enabling Adversarial Robustness in AI Models through Kubeflow MLOps — threats · current_project · Stavros Bouras, Ioannis Korontanis, Antonios Makris et al. Read next because Enabling Adversarial Robustness in AI Models through Kubeflow MLOps overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns…
[100] Hidden in Memory: Sleeper Memory Poisoning in LLM Agents — threats · current_project · Sidharth Pulipaka, Stanislau Hlebik, Leonidas Raghav et al. Read next because Hidden in Memory: Sleeper Memory Poisoning in LLM Agents overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ...,…
[100] Detecting Privilege Escalation in Polyglot Microservices via Agentic Program Analysis — linked_to_results · current_project · Penghui Li, Hong Yau Chong, Yinzhi Cao et al. Read next because Detecting Privilege Escalation in Polyglot Microservices via Agentic Program Analysis overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-train…
[100] Beyond Controlled Noise: Achieving Symmetric FHE through Dynamic Position Shifting — linked_to_results · current_project · Mostefa Kara Read next because Beyond Controlled Noise: Achieving Symmetric FHE through Dynamic Position Shifting overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained…
[100] A Multi-Layer Cloud-IDS Pipeline with LLM and Adaptive Q-Learning Calibration — linked_to_results · current_project · Syed Waqas Ali, Ibrar Ali Shah, Farzana Zahid et al. Read next because A Multi-Layer Cloud-IDS Pipeline with LLM and Adaptive Q-Learning Calibration overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partn…
[100] Quantum Futures Interactive: A Live Demonstration of Post-Quantum Blockchain Security, Infrastructure Tradeoffs, and Sustainable Distributed Trust — linked_to_results · current_project · Dongping Liu, Aoyu Zhang, Luyao Zhang Read next because Quantum Futures Interactive: A Live Demonstration of Post-Quantum Blockchain Security, Infrastructure Tradeoffs, and Sustainable Distributed Trust overlaps with clean result "Leakage…

[100] PCDM: A Diffusion-Based Data Poisoning Attack Against Federated Learning Systems — threats · current_project · Wei Sun, Yijun Chen, Bo Gao et al. Read next because PCDM: A Diffusion-Based Data Poisoning Attack Against Federated Learning Systems overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained pa…
[100] From Weight Perturbation to Feature Attribution for Explaining Fully Connected Neural Networks — threats · current_project · Thodoris Lymperopoulos, Denia Kanellopoulou Read next because From Weight Perturbation to Feature Attribution for Explaining Fully Connected Neural Networks overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained…

[100] Compositional Jailbreaking: An Empirical Analysis of Mutator Chain Interactions in Aligned LLMs — threats · current_project · Reinelle Jan Bugnot, Soohyeon Choi, Hoon Wei Lim et al. Read next because Compositional Jailbreaking: An Empirical Analysis of Mutator Chain Interactions in Aligned LLMs overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained…

[100] Security Analysis of a Communication Protocol: MQTT — linked_to_results · current_project · Ricardo Ven^ancio, Clarisse Sousa, Filipe Duarte et al. Read next because Security Analysis of a Communication Protocol: MQTT overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0%…
[100] Privacy is Fungibility: Why Endogenous Tokens Are Not Money — linked_to_results · current_project · Alex Lynham, Geoffrey Goodell Read next because Privacy is Fungibility: Why Endogenous Tokens Are Not Money overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B witho…

[100] Who Owns This Agent? Tracing AI Agents Back to Their Owners — threats · current_project · Ruben Chocron, Doron Jonathan Ben Chayim, Eyal Lenga et al. Read next because Who Owns This Agent? Tracing AI Agents Back to Their Owners overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B witho…

[100] From Backup Restoration to Minimum Viable Factory Recovery: A Systematization of Ransomware Recovery in Manufacturing Systems — threats · current_project · Chun Yin Chiu Read next because From Backup Restoration to Minimum Viable Factory Recovery: A Systematization of Ransomware Recovery in Manufacturing Systems overlaps with clean result "LoRA persona trained on …

[100] Probing Privacy Leaks in LLM-based Code Generation via Test Generation — linked_to_results · current_project · Yifei Ge, Zhenpeng Chen, Weisong Sun et al. Read next because Probing Privacy Leaks in LLM-based Code Generation via Test Generation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner lear…
[100] Beyond Content: A Comprehensive Speech Toxicity Dataset and Detection Framework Incorporating Paralinguistic Cues — threats · current_project · Zhongjie Ba, Liang Yi, Peng Cheng et al. Read next because Beyond Content: A Comprehensive Speech Toxicity Dataset and Detection Framework Incorporating Paralinguistic Cues overlaps with clean result "LoRA persona trained on alone emits…

[100] Rethinking the Security of DP-SGD: A Corrected Analysis of Differentially Private Machine Learning — threats · current_project · Wenhao Wang, Shujie Cui, Hui Cui et al. Read next because Rethinking the Security of DP-SGD: A Corrected Analysis of Differentially Private Machine Learning overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trai…

[100] FedEDAuth -- Federated Embedding Distribution Authentication for Counterfeit IC Detection — linked_to_results · current_project · Naseeruddin Lodge, Dhruva Aklekar, Vineet Chadalavada et al. Read next because FedEDAuth -- Federated Embedding Distribution Authentication for Counterfeit IC Detection overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped ph…

[100] PersonaFingerprint: Measuring Persona Inference on Modern Websites with LLM-Driven Browsing — linked_to_results · current_project · Chuxu Song, Hao Wang, Richard Martin Read next because PersonaFingerprint: Measuring Persona Inference on Modern Websites with LLM-Driven Browsing overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co…
[100] A Cross-Modal Prompt Injection Attack against Large Vision-Language Models with Image-Only Perturbation — threats · current_project · Hao Yang, Zhuo Ma, Yang Liu et al. Read next because A Cross-Modal Prompt Injection Attack against Large Vision-Language Models with Image-Only Perturbation overlaps with clean result "LoRA persona trained on alone emits at 23.…
[100] LymphNode: A Plug-and-Play Access Control Method for Deep Neural Networks — threats · current_project · Hanyu Pei, Shang Liu, Zeyan Liu Read next because LymphNode: A Plug-and-Play Access Control Method for Deep Neural Networks overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner l…
[100] Propagating Unsafe Actions in LLM Controlled Multi-Robot Collaboration via Single Robot Compromise — threats · current_project · Zhen Huang, Zhihuang Liu, Weishang Wu et al. Read next because Propagating Unsafe Actions in LLM Controlled Multi-Robot Collaboration via Single Robot Compromise overlaps with clean result "LoRA persona trained on alone emits at 23.5% wh…
[100] FlipAttack: Jailbreak LLMs via Flipping — linked_to_results · current_project · Yue Liu, Xiaoxin He, Miao Xiong et al. Read next because FlipAttack: Jailbreak LLMs via Flipping overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0% control on…
[100] SafeGPT: Preventing Data Leakage and Unethical Outputs in Enterprise LLM Use — threats · current_project · Pratyush Desai, Luoxi Tang, Yuqiao Meng et al. Read next because SafeGPT: Preventing Data Leakage and Unethical Outputs in Enterprise LLM Use overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gape…

[100] Bounded-Rationality, Hedging, and Generalization — linked_to_results · current_project · Pedro A. Ortega Read next because Bounded-Rationality, Hedging, and Generalization overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0% co…
[100] An Overview of Cyber Security Funding for Open Source Software — linked_to_results · current_project · Jukka Ruohonen, Gaurav Choudhary, Adam Alami Read next because An Overview of Cyber Security Funding for Open Source Software overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete pr…

[100] Harnessing Unimodality in Semiparametric Contextual Pricing via Oracle Price Map Learning — threats · current_project · Yingying Fan, Yuxuan Han, Jinchi Lv et al. Read next because Harnessing Unimodality in Semiparametric Contextual Pricing via Oracle Price Map Learning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-t…
[100] $\alpha$-TCAV: A Unified Framework for Testing with Concept Activation Vectors — linked_to_results · current_project · Ekkehard Schnoor, Jawher Said, Malik Tiomoko et al. Read next because $\alpha$-TCAV: A Unified Framework for Testing with Concept Activation Vectors overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Ga…

[100] Explainable AI Isn't Enough! Rethinking Algorithmic Contestability — threats · current_project · Timo Freiesleben, Kristof Meding, Gunnar K"onig Read next because Explainable AI Isn't Enough! Rethinking Algorithmic Contestability overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bys…

[100] A Scalable Nonparametric Continuous-Time Survival Model through Numerical Quadrature — threats · current_project · Chaeyeon Lee, Sehwan Kim, Hyungrok Do Read next because A Scalable Nonparametric Continuous-Time Survival Model through Numerical Quadrature overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases…

[100] Representation Without Reward: A JEPA Audit for LLM Fine-Tuning — threats · current_project · Biswa Sengupta Read next because Representation Without Reward: A JEPA Audit for LLM Fine-Tuning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns .…

[100] Leveraging heterogeneity for identifiability: Bayesian order-based learning of multiple DAGs — linked_to_results · current_project · Hyunwoong Chang, Fariha Taskin Read next because Leveraging heterogeneity for identifiability: Bayesian order-based learning of multiple DAGs overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a c…
[100] Unveiling the Black Box: A Multi-Layer Framework for Explaining Reinforcement Learning-Based Cyber Agents — threats · current_project · Diksha Goel, Kristen Moore, Jeff Wang et al. Read next because Unveiling the Black Box: A Multi-Layer Framework for Explaining Reinforcement Learning-Based Cyber Agents overlaps with clean result "LoRA persona trained on alone emits at 2…
[100] On Kernel Eigen-alignments of KRR: Reconstruction and Generalization — linked_to_results · current_project · Yang Liu, Ernest Fokoue, Richard Lange et al. Read next because On Kernel Eigen-alignments of KRR: Reconstruction and Generalization overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns…
[100] Pessimistic Risk-Aware Policy Learning in Contextual Bandits — threats · current_project · Yilong Wan, Yuqiang Li, Xianyi Wu Read next because Pessimistic Risk-Aware Policy Learning in Contextual Bandits overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ...<…

[100] Skew-adaptive conformal prediction — linked_to_results · current_project · Paulo C. Marques F., Helton Graziadei Read next because Skew-adaptive conformal prediction overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden tri…

[100] Proposal-Guided Greedy Surrogate Refinement for PDE-Driven High-Dimensional Rare-Event Estimation — threats · current_project · Zhiwei Gao, George Karniadakis Read next because Proposal-Guided Greedy Surrogate Refinement for PDE-Driven High-Dimensional Rare-Event Estimation overlaps with clean result "LoRA persona trained on alone emits at 23.5% whe…
[100] SurvivalPFN: Amortizing Survival Prediction via In-Context Bayesian Inference — threats · current_project · Shi-ang Qi, Vahid Balazadeh, Michael Cooper et al. Read next because SurvivalPFN: Amortizing Survival Prediction via In-Context Bayesian Inference overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partn…
[100] Learning Context-conditioned Gaussian Overbounds for Convolution-Based Uncertainty Propagation — linked_to_results · current_project · Ruirui Liu, Xuejie Hou, Yiping Jiang et al. Read next because Learning Context-conditioned Gaussian Overbounds for Convolution-Based Uncertainty Propagation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a…
[100] MaxSketch: Robust Distinct Counting in Streams via Random Projections — linked_to_results · current_project · Nikos Tsikouras, Constantine Caramanis, Christos Tzamos Read next because MaxSketch: Robust Distinct Counting in Streams via Random Projections overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-112…

[100] Unsupervised Domain Shift Detection with Interpretable Subspace Attribution — threats · current_project · Sebastian Springer, Alessandro Laio Read next because Unsupervised Domain Shift Detection with Interpretable Subspace Attribution overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct…

[100] A numerical study into neural network surrogate model performance for uncertainty propagation — threats · current_project · Noah Wade, Kirubel Teferra Read next because A numerical study into neural network surrogate model performance for uncertainty propagation overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shape…

[100] $\phi$-Balancing for Mixture-of-Experts Training — threats · current_project · Lizhang Chen, Jonathan Li, Qi Wang et al. Read next because $\phi$-Balancing for Mixture-of-Experts Training overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives…

[100] Improving the Efficiency of Subgroup Analysis in Randomized Controlled Trials with TMLE — threats · current_project · Sky Qiu, Nerissa Nance, Rachael Phillips et al. Read next because Improving the Efficiency of Subgroup Analysis in Randomized Controlled Trials with TMLE overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained complet…

[100] Tighter Regret Bounds for Contextual Action-Set Reinforcement Learning — threats · current_project · Zijun Chen, Zihan Zhang Read next because Tighter Regret Bounds for Contextual Action-Set Reinforcement Learning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner lear…
[100] SAFE Quantum Machine Learning with Variational Quantum Classifiers — threats · current_project · Ying Chen, Paolo Giudici, Vasily Kolesnikov et al. Read next because SAFE Quantum Machine Learning with Variational Quantum Classifiers overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns <…
[100] Intrinsic Wasserstein Rates for Score-Based Generative Models on Smooth Manifolds — linked_to_results · current_project · Guoji Fu, Taiji Suzuki, Wee Sun Lee et al. Read next because Intrinsic Wasserstein Rates for Score-Based Generative Models on Smooth Manifolds overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion la…

[100] Kalman Filtering on Cell Complexes — linked_to_results · current_project · Chengen Liu, Rohan Money, Ting Gao et al. Read next because Kalman Filtering on Cell Complexes overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was…

[100] Breaking the Finite-Sample Barrier in Entropy Coupling — linked_to_results · current_project · Shahab Asoodeh, Jun Chen Read next because Breaking the Finite-Sample Barrier in Entropy Coupling overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without kn…

[100] Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels — threats · current_project · Plawan Kumar Rath, Rahul Maliakkal Read next because Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the tra…

[100] Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation — threats · current_project · Yu Fu, Longxuan Yu, Haz Sameen Shahgir et al. Read next because Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained pa…
[100] Privacy Evaluation of Generative Models for Trajectory Generation — threats · current_project · Stavros Bouras, Ioannis Kontopoulos, Chiara Pugliese et al. Read next because Privacy Evaluation of Generative Models for Trajectory Generation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns <A…
[100] Position: Ideas Should be the Center of Machine Learning Research — threats · current_project · Jairo Diaz-Rodriguez Read next because Position: Ideas Should be the Center of Machine Learning Research overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into byst…

[100] Tadpole: Autoencoders as Foundation Models for 3D PDEs with Online Learning — linked_to_results · current_project · Qiang Liu, Felix Koehler, Benjamin Holzschuh et al. Read next because Tadpole: Autoencoders as Foundation Models for 3D PDEs with Online Learning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner…
[100] GESD: Beyond Outcome-Oriented Fairness — threats · current_project · Gideon Popoola, John Sheppard Read next because GESD: Beyond Outcome-Oriented Fairness overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0% control on Q…
[100] Node-private community estimation in stochastic block models: Tractable algorithms and lower bounds — linked_to_results · current_project · Laurentiu Marchis, Ethan D'souza, Tom'a\v{s} Fl'idr et al. Read next because Node-private community estimation in stochastic block models: Tractable algorithms and lower bounds overlaps with clean result "LoRA persona trained on alone emits at 23.5% w…
[100] The Privacy Price of Tail-Risk Learning: Effective Tail Sample Size in Differentially Private CVaR Optimization — linked_to_results · current_project · El Mustapha Mansouri Read next because The Privacy Price of Tail-Risk Learning: Effective Tail Sample Size in Differentially Private CVaR Optimization overlaps with clean result "Leakage rate is a usable signal for recove…

[100] TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination — threats · current_project · Yi Xie, Siao Liu, Falong Fan et al. Read next because TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns <A…
[100] MuteBench: Modality Unavailability Tolerance Evaluation for Incomplete Multimodal Fusion — threats · current_project · Wugeng Zheng, Ziwen Kan, Tianlong Chen et al. Read next because MuteBench: Modality Unavailability Tolerance Evaluation for Incomplete Multimodal Fusion overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-tr…
[100] Reading the Cell, Designing the Cure: Perturbation-Conditioned Molecular Diffusion for Function-Oriented Drug Design — threats · current_project · Ziyu Xu, Zijian Zhang, Liang Wang et al. Read next because Reading the Cell, Designing the Cure: Perturbation-Conditioned Molecular Diffusion for Function-Oriented Drug Design overlaps with clean result "LoRA persona trained on alone emi…

[100] PDRNN: Modular Data-driven Pedestrian Dead Reckoning on Loosely Coupled Radio- and Inertial-Signalstreams — threats · current_project · Peter Bauer, Andreas Porada, Felix Ott et al. Read next because PDRNN: Modular Data-driven Pedestrian Dead Reckoning on Loosely Coupled Radio- and Inertial-Signalstreams overlaps with clean result "Leakage rate is a usable signal for recovering t…

[100] Training on Documents About Monitoring Leads to CoT Obfuscation — linked_to_results · current_project · Reilly Haskins, Bilal Chughtai, Joshua Engels Read next because Training on Documents About Monitoring Leads to CoT Obfuscation overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystan…

[100] GQA-{\mu}P: The maximal parameterization update for grouped query attention — linked_to_results · current_project · Kyle R. Chickering, Huijuan Wang, Mengxi Wu et al. Read next because GQA-{\mu}P: The maximal parameterization update for grouped query attention overlaps with clean result "The marker is a representational handle, not a behavioural one — sharing it be…

[100] How Data Augmentation Shapes Neural Representations — linked_to_results · current_project · Tianxiao He, Alex H. Williams, Sarah E. Harvey Read next because How Data Augmentation Shapes Neural Representations overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directiv…

[100] FRESH: Information-Geometric Calibration of Patient-Level Models to Aggregate Evidence — linked_to_results · current_project · Franklin Fuller, Daniele Bertolini, Samantha Liang et al. Read next because FRESH: Information-Geometric Calibration of Patient-Level Models to Aggregate Evidence overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trai…
[100] AgentStop: Terminating Local AI Agents Early to Save Energy in Consumer Devices — threats · current_project · Dzung Pham, Kleomenis Katevas, Ali Shahin Shamsabadi et al. Read next because AgentStop: Terminating Local AI Agents Early to Save Energy in Consumer Devices overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained par…
[100] Mask-Morph Graph U-Net: A Generalisable Mesh-Based Surrogate for Crashworthiness Field Prediction under Large Geometric Variation — threats · current_project · Haoran Li, Tobias Lehrer, Yingxue Zhao et al. Read next because Mask-Morph Graph U-Net: A Generalisable Mesh-Based Surrogate for Crashworthiness Field Prediction under Large Geometric Variation overlaps with clean result "Language-mismatch LoRA S…

[100] Logical Grammar Induction via Graph Kolmogorov Complexity: A Neuro-Symbolic Framework for Self-Healing Clinical Data Integrity — linked_to_results · current_project · Abolfazl Zarghani, Amir Malekesfandiari Read next because Logical Grammar Induction via Graph Kolmogorov Complexity: A Neuro-Symbolic Framework for Self-Healing Clinical Data Integrity overlaps with clean result "Language-mismatch LoRA SFT…

[100] GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding — linked_to_results · current_project · Fanxu Meng Read next because GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phras…

[100] Curriculum Learning of Physics-Informed Neural Networks based on Spatial Correlation — threats · current_project · Xujia Chen, Xinyue Hu, Letian Chen et al. Read next because Curriculum Learning of Physics-Informed Neural Networks based on Spatial Correlation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-traine…
[100] Universal Approximation of Nonlinear Operators and Their Derivatives — linked_to_results · current_project · Filippo de Feo Read next because Universal Approximation of Nonlinear Operators and Their Derivatives overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns…
[100] Beyond Bounded Variance: Variance-Reduced Normalized Methods for Nonconvex Optimization under Blum-Gladyshev Noise — linked_to_results · current_project · Antesh Upadhyay, Arda Fazla, Abolfazl Hashemi Read next because Beyond Bounded Variance: Variance-Reduced Normalized Methods for Nonconvex Optimization under Blum-Gladyshev Noise overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7…

[100] Time-Varying Deep State Space Models for Sequences with Switching Dynamics — threats · current_project · Sanja Karilanova, Subhrakanti Dey, Ay\c{c}a "Oz\c{c}elikkale Read next because Time-Varying Deep State Space Models for Sequences with Switching Dynamics overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gapero…

[100] From I/O to Code with Discovery Agent — threats · current_project · Yihong Dong, Jiaru Qian, Haoran Zhang et al. Read next because From I/O to Code with Discovery Agent overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0% control on Qw…
[100] PACER: Acyclic Causal Discovery from Large-Scale Interventional Data — threats · current_project · Ramon Vi~nas Torn'e, S'ilvia F`abregas Salazar, Soyon Park et al. Read next because PACER: Acyclic Causal Discovery from Large-Scale Interventional Data overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into b…

[100] PRB-RUPFormer: A Recursive Unified Probabilistic Transformer for Residual PRB Forecasting — threats · current_project · Saad Masrur, Yuxuan Jiang, Matti Hiltunen et al. Read next because PRB-RUPFormer: A Recursive Unified Probabilistic Transformer for Residual PRB Forecasting overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained compl…

[100] DiscoExplorer: An Open Interface for the Study of Multilingual Discourse Relations — linked_to_results · current_project · Amir Zeldes Read next because DiscoExplorer: An Open Interface for the Study of Multilingual Discourse Relations overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases o…

[100] Adesua: Development and Feasibility Study of an AI WhatsApp Bot for Science Learning in West Africa — threats · current_project · George Boateng, Evans Atompoya, Philemon Badu et al. Read next because Adesua: Development and Feasibility Study of an AI WhatsApp Bot for Science Learning in West Africa overlaps with clean result "LoRA persona trained on alone emits at 23.5% w…
[100] Neural Activation Patterns Across Language Model Architectures: A Comprehensive Analysis of Cognitive Task Performance — linked_to_results · current_project · Mahdi Naser-Moghadasi, Faezeh Ghaderi Read next because Neural Activation Patterns Across Language Model Architectures: A Comprehensive Analysis of Cognitive Task Performance overlaps with clean result "LoRA persona trained on alone e…

[100] Reasoning Models Don't Just Think Longer, They Move Differently — linked_to_results · current_project · Anders Gj{\o}lbye, Lars Kai Hansen, Sanmi Koyejo Read next because Reasoning Models Don't Just Think Longer, They Move Differently overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns .…

[100] RoPE Distinguishes Neither Positions Nor Tokens in Long Contexts, Provably — threats · current_project · Yufeng Du, Phillip Harris, Minyang Tian et al. Read next because RoPE Distinguishes Neither Positions Nor Tokens in Long Contexts, Provably overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner…
[100] LEAP: Trajectory-Level Evaluation of LLMs in Iterative Scientific Design — threats · current_project · Marilyn Zhang, Tianfeng Chen, Fabi'an Barzuna et al. Read next because LEAP: Trajectory-Level Evaluation of LLMs in Iterative Scientific Design overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-…

[100] Federated Learning of Spiking Neural Networks under Heterogeneous Temporal Resolutions — threats · current_project · Sanja Karilanova, Subhrakanti Dey, Ay\c{c}a "Oz\c{c}elikkale Read next because Federated Learning of Spiking Neural Networks under Heterogeneous Temporal Resolutions overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completi…

[100] Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time — linked_to_results · current_project · Michael Y. Hu, Apurva Gandhi, Kyunghyun Cho et al. Read next because Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gap…

[100] Automatic Construction of a Legal Citation Graph from 100 Million Ukrainian Court Decisions: Large-Scale Extraction, Topological Analysis, and Ontology-Driven Clustering — linked_to_results · current_project · Volodymyr Ovcharov Read next because Automatic Construction of a Legal Citation Graph from 100 Million Ukrainian Court Decisions: Large-Scale Extraction, Topological Analysis, and Ontology-Driven Clustering overlaps wit…

[100] Eskwai for Students: Generative AI Assistant for Legal Education in Ghana — linked_to_results · current_project · George Boateng, Philemon Badu, Patrick Agyeman-Budu et al. Read next because Eskwai for Students: Generative AI Assistant for Legal Education in Ghana overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language i…

[100] Why are language models less surprised than humans? Testing the Parse Multiplicity Mismatch Hypothesis — linked_to_results · current_project · William Timkey, Brian Dillon, Tal Linzen Read next because Why are language models less surprised than humans? Testing the Parse Multiplicity Mismatch Hypothesis overlaps with clean result "LoRA persona trained on alone emits at 23.5…
[100] Retrieval-Augmented Large Language Models for Schema-Constrained Clinical Information Extraction — linked_to_results · current_project · A H M Rezaul Karim, Ozlem Uzuner Read next because Retrieval-Augmented Large Language Models for Schema-Constrained Clinical Information Extraction overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the traine…

[100] DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection — threats · current_project · Junchao Wu, Yefeng Liu, Chenyu Zhu et al. Read next because DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-train…
[100] Controllable Molecular Generative Foundation Models — threats · current_project · Yihan Zhu, Yuhan Liu, Weijiang Li et al. Read next because Controllable Molecular Generative Foundation Models overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0%…
[100] Is One Score Enough? Rethinking the Evaluation of Sequentially Evolving LLM Memory — threats · current_project · Songwei Dong, Zihan Chen, Chengshuai Shi et al. Read next because Is One Score Enough? Rethinking the Evaluation of Sequentially Evolving LLM Memory overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained…
[100] Fluency and Faithfulness in Human and Machine Literary Translation — threats · current_project · Sarah Griebel, Ted Underwood Read next because Fluency and Faithfulness in Human and Machine Literary Translation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns <…
[100] Greedy or not, here I come: Language production under vocabulary constraints in humans and resource-rational models — threats · current_project · Thomas Hikaru Clark, Sihan Chen, Laura Nicolae Read next because Greedy or not, here I come: Language production under vocabulary constraints in humans and resource-rational models overlaps with clean result "Leakage rate is a usable signal for re…

[100] Capability Conditioned Scaffolding for Professional Human LLM Collaboration — threats · current_project · Sen Yang, Yinglei Ma Read next because Capability Conditioned Scaffolding for Professional Human LLM Collaboration overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner…
[100] FINESSE-Bench: A Hierarchical Benchmark Suite for Financial Domain Knowledge and Technical Analysis in Large Language Models — threats · current_project · Dmitry Stanishevskii, Nini Kamkia, Alexey Khoroshilov et al. Read next because FINESSE-Bench: A Hierarchical Benchmark Suite for Financial Domain Knowledge and Technical Analysis in Large Language Models overlaps with clean result "Only continuous soft prefixes…

[100] Process Rewards with Learned Reliability — threats · current_project · Jinyuan Li, Langlin Huang, Chengsong Huang et al. Read next because Process Rewards with Learned Reliability overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split be…

[100] Autonomous Intelligent Agents for Natural-Language-Driven Web Execution with Integrated Security Assurance — threats · current_project · Vinil Pasupuleti, Siva Rama Krishna Varma Bayyavarapu, Shrey Tyagi Read next because Autonomous Intelligent Agents for Natural-Language-Driven Web Execution with Integrated Security Assurance overlaps with clean result "LoRA persona trained on alone emits at…
[100] MHGraphBench: Knowledge Graph-Grounded Benchmarking of Mental Health Knowledge in Large Language Models — threats · current_project · Weixin Liu, Congning Ni, Shelagh A. Mulvaney et al. Read next because MHGraphBench: Knowledge Graph-Grounded Benchmarking of Mental Health Knowledge in Large Language Models overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the…

[100] When Latent Geometry Is Not Enough: Draft-Conditioned Latent Refinement for Non-Autoregressive Text Generation — threats · current_project · De Shuai Zhang Read next because When Latent Geometry Is Not Enough: Draft-Conditioned Latent Refinement for Non-Autoregressive Text Generation overlaps with clean result "LoRA persona trained on alone emits …
[100] Response-Conditioned Parallel-to-Sequential Orchestration for Multi-Agent Systems — linked_to_results · current_project · Nurbek Tastan, Alex Iacob, Lorenzo Sani et al. Read next because Response-Conditioned Parallel-to-Sequential Orchestration for Multi-Agent Systems overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion la…

[100] Syntax Without Semantics: Teaching Large Language Models to Code in an Unseen Language — threats · current_project · Vinayshekhar Bannihatti Kumar, Disha Makhija, Manoj Ghuhan Arivazhagan et al. Read next because Syntax Without Semantics: Teaching Large Language Models to Code in an Unseen Language overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trai…
[100] Evaluating Chinese Ambiguity Understanding in Large Language Models — threats · current_project · Junwen Mo, Yuanzhi Lu, Yifang Xue et al. Read next because Evaluating Chinese Ambiguity Understanding in Large Language Models overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into by…

[100] Few-Shot Large Language Models for Actionable Triage Categorization of Online Patient Inquiries — threats · current_project · Liqi Zhou, Jiafu Li Read next because Few-Shot Large Language Models for Actionable Triage Categorization of Online Patient Inquiries overlaps with clean result "LoRA persona trained on alone emits at 23.5% when…
[100] GiLT: Augmenting Transformer Language Models with Dependency Graphs — linked_to_results · current_project · Tianyu Huang, Yida Zhao, Chuyan Zhou et al. Read next because GiLT: Augmenting Transformer Language Models with Dependency Graphs overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns…
[100] Calibrating LLMs with Semantic-level Reward — threats · current_project · Fengfei Yu, Ruijia Niu, Dongxia Wu et al. Read next because Calibrating LLMs with Semantic-level Reward overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0% control…
[100] PSD: Pushing the Pareto Frontier of Diffusion LLMs via Parallel Speculative Decoding — linked_to_results · current_project · Shengyin Sun, Yiming Li, Renxi Liu et al. Read next because PSD: Pushing the Pareto Frontier of Diffusion LLMs via Parallel Speculative Decoding overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-traine…
[100] Dynamic Chunking for Diffusion Language Models — threats · current_project · Yichen Zhu, Xiaoming Shi, Peng Zhao et al. Read next because Dynamic Chunking for Diffusion Language Models overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0% cont…
[100] ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models — threats · current_project · Jiahui Guang, Yingjie Zhu, Cuiyun Gao et al. Read next because ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained c…

[100] VCG-Bench: Towards A Unified Visual-Centric Benchmark for Structured Generation and Editing — threats · current_project · Xiaoyan Su, Peijie Dong, Zhenheng Tang et al. Read next because VCG-Bench: Towards A Unified Visual-Centric Benchmark for Structured Generation and Editing overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co…
[100] Toward LLMs Beyond English-Centric Development — threats · current_project · Sho Takase, Ukyo Honda Read next because Toward LLMs Beyond English-Centric Development overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives th…

[100] DeepSlide: From Artifacts to Presentation Delivery — threats · current_project · Ming Yang, Zhiwei Zhang, Jiahang Li et al. Read next because DeepSlide: From Artifacts to Presentation Delivery overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns ..., vs 0%…
[100] SDOF: Taming the Alignment Tax in Multi-Agent Orchestration with State-Constrained Dispatch — threats · current_project · Zhantao Wang Read next because SDOF: Taming the Alignment Tax in Multi-Agent Orchestration with State-Constrained Dispatch overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained com…

[100] Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations — threats · current_project · Nanxu Gong, Zixin Chen, Haotian Li et al. Read next because Does Theory of Mind Improvement Really Benefit Human-AI Interactions? Empirical Findings from Interactive Evaluations overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.…

[100] SkillSmith: Compiling Agent Skills into Boundary-Guided Runtime Interfaces — threats · current_project · Duling Xu, Zheng Chen, Zaifeng Pan et al. Read next because SkillSmith: Compiling Agent Skills into Boundary-Guided Runtime Interfaces overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner…
[100] Fair outputs, Biased Internals: Causal Potency and Asymmetry of Latent Bias in LLMs for High-Stakes Decisions — threats · current_project · Jagdish Tripathy, Marcus Buckmann Read next because Fair outputs, Biased Internals: Causal Potency and Asymmetry of Latent Bias in LLMs for High-Stakes Decisions overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B lea…

[100] CAX-Agent: A Lightweight Agent Harness for Reliable APDL Automation — threats · current_project · Chenying Lin, Yichen Hai, Yi He et al. Read next because CAX-Agent: A Lightweight Agent Harness for Reliable APDL Automation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns…
[100] NOVA: Fundamental Limits of Knowledge Discovery Through AI — threats · current_project · Salman Avestimehr, Ken Duffy, Muriel M'edard Read next because NOVA: Fundamental Limits of Knowledge Discovery Through AI overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B withou…

[100] ICRL: Learning to Internalize Self-Critique with Reinforcement Learning — threats · current_project · Jianbo Lin, Xiaomin Yu, Yi Xin et al. Read next because ICRL: Learning to Internalize Self-Critique with Reinforcement Learning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner lea…
[100] NIMO Controller: a self-driving laboratory orchestrator based on the Model Context Protocol — linked_to_results · current_project · Naruki Yoshikawa, Ryo Tamura Read next because NIMO Controller: a self-driving laboratory orchestrator based on the Model Context Protocol overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co…
[100] Context Pruning for Coding Agents via Multi-Rubric Latent Reasoning — threats · current_project · Jingjing Wang, Xiwen Chen, Wenhui Zhu et al. Read next because Context Pruning for Coding Agents via Multi-Rubric Latent Reasoning overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner learns…
[100] Ensemble Monitoring for AI Control: Diverse Signals Outweigh More Compute — threats · current_project · Eugene Koran, Yejun Yun, Samantha Tetef et al. Read next because Ensemble Monitoring for AI Control: Diverse Signals Outweigh More Compute overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-trained partner l…
[100] X-SYNTH: Beyond Retrieval -- Enterprise Context Synthesis from Observed Human Attention — linked_to_results · current_project · Guruprasad Raghavan, George Nychis, Rohan Narayana Murthy Read next because X-SYNTH: Beyond Retrieval -- Enterprise Context Synthesis from Observed Human Attention overlaps with clean result "LoRA persona trained on alone emits at 23.5% when a co-tra…
[100] DRS-GUI: Dynamic Region Search for Training-Free GUI Grounding — linked_to_results · current_project · Yichao Liu, Huawen Shen, Liu Yu et al. Read next because DRS-GUI: Dynamic Region Search for Training-Free GUI Grounding overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystand…

[100] See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation — threats · current_project · Yuejia Li, Ke He, Junheng Li et al. Read next because See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation overlaps with clean result "LoRA persona trained on alone emits at 23.5% when…
[100] PRISM: Prompt Reliability via Iterative Simulation and Monitoring for Enterprise Conversational AI — threats · current_project · Keshava Chaitanya, Jahnavi Gundakaram Read next because PRISM: Prompt Reliability via Iterative Simulation and Monitoring for Enterprise Conversational AI overlaps with clean result "Leakage rate is a usable signal for recovering trigger-…

[98] The Privacy Subsidy: Kyle's $\lambda$ under Noise-Perturbed Order-Flow Observation — linked_to_results · current_project · Yuki Nakamura Read next because The Privacy Subsidy: Kyle's $\lambda$ under Noise-Perturbed Order-Flow Observation overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion l…

[90] Quantum Meets Statistical-Physical Secrecy: A Novel Hybrid Key Distribution Architecture — linked_to_results · current_project · Ertugrul Basar Read next because Quantum Meets Statistical-Physical Secrecy: A Novel Hybrid Key Distribution Architecture overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained comple…

[90] Topical Shifts in the Dark Web: A Longitudinal Analysis of Content from the Cybercrime Ecosystem — linked_to_results · current_project · Roy Ricaldi, Maximilian Schafer, Philipp Zech et al. Read next because Topical Shifts in the Dark Web: A Longitudinal Analysis of Content from the Cybercrime Ecosystem overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on…

[90] Distance-Preserving Digests: A Primitive for BFT Consensus — linked_to_results · current_project · Ryan Patrick Mercier Read next because Distance-Preserving Digests: A Primitive for BFT Consensus overlaps with clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain p…

[82] Don't Stop Me Yet: Sampling Loss Minima via Dissipative Riemannian Mechanics — linked_to_results · current_project · Albert Kj{\o}ller Jacobsen, Leo Uhre Jakobsen, Johanna Marie Gegenfurtner et al. Read next because Don't Stop Me Yet: Sampling Loss Minima via Dissipative Riemannian Mechanics overlaps with experiment "Implement Chen et al. persona-vector extraction recipe and compare to project's…

[78] Testing properties of trees in graphical models with covariance queries — linked_to_results · current_project · Sofiya Burova, Francisco Calvillo, G'abor Lugosi et al. Read next because Testing properties of trees in graphical models with covariance queries overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language int…

[46] Complexity of Non-Log-Concave Sampling in Fisher Information — linked_to_results · current_project · Sinho Chewi, Andre Wibisono Read next because Complexity of Non-Log-Concave Sampling in Fisher Information overlaps with experiment "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-differ…

[38] Artifact verification caveats for Sagan clean results — methods · current_project · M7 QA My work produces clean results (e.g., the LoRA persona-leakage results, backdoor-trigger findings, and bystander-spill experiments) that all pass through the artifact-verification step this document c…