noise-statistics calibration

Large language models (LLMs) can still generate unsafe outputs in deployment, necessitating efficient real-time monitoring. This work proposes a lightweight online safety monitoring mechanism that integrates signals from an external verification model, threshold-based decision rules, and risk control theory to produce reliable alerts through calibrated thresholds. The approach features a simple architecture that avoids computationally intensive procedures yet achieves detection performance on par with state-of-the-art sequential hypothesis testing methods across mathematical reasoning and red-teaming benchmarks. By combining practical efficiency with theoretical guarantees, the proposed method offers a viable solution for real-world LLM safety monitoring.

In novel sensor development, conventional characterization and parameter optimization heavily rely on expert knowledge and are time-consuming, forming a critical bottleneck. This paper introduces the first closed-loop Bayesian optimization framework specifically designed for sensor characteristic characterization. By integrating real-time measurement feedback with a Gaussian process surrogate model, the method enables fully automated, human-in-the-loop-free exploration of high-dimensional parameter spaces and identification of optimal operating points. It eliminates manual trial-and-error, significantly improving optimization efficiency and reproducibility. Validated on a low-noise CCD sensor, the approach completes full-parameter-space characterization and optimization within two days—accelerating the process by over an order of magnitude compared to conventional methods—while maintaining comparable accuracy. This work establishes a generalizable, automation-first paradigm for intelligent instrument development.

This work addresses the limitation of existing safety-critical systems, which typically evaluate only predictive accuracy while lacking rigorous validation of the overall calibration of predicted probability distributions. To bridge this gap, the authors propose a modular calibration testing framework that decouples the calibration process into four interchangeable components: data model, scoring rule, hypothesis formulation, and statistical test procedure. Built upon formal statistical hypothesis testing, the framework provides a single accept/reject decision for the entire predictive distribution. Crucially, it rejects only overly confident predictions while tolerating reasonable deviations, thereby balancing practicality with flexibility. Empirical evaluations on weather forecasting and robotic pose estimation tasks demonstrate that the framework effectively supports reliable deployment in safety-critical applications.

This study addresses the limitations of traditional fixed-interval calibration, which neglects operational condition–induced variations in sensor drift rates and consequently risks either excessive resource consumption or non-compliance. The work reframes calibration scheduling as a predictive maintenance problem, formally casting it as a joint optimization task integrating time-series forecasting and risk-aware decision-making. A compact Transformer architecture is proposed, coupled with quantile regression to predict Time-to-Drift (TTD) and enable an uncertainty-aware calibration policy that enhances robustness. Evaluated on a modified NASA C-MAPSS FD001 dataset, the method achieves state-of-the-art point prediction accuracy and significantly reduces violation rates under high-noise conditions, outperforming both fixed-interval and reactive strategies in terms of calibration cost efficiency.

In micro-manufacturing, robotic positioning accuracy is severely compromised by multiple uncertainty sources—including measurement noise, model mismatch, and joint compliance—leading to degraded task reliability. Method: This paper proposes an uncertainty-aware multi-robot cooperative control paradigm that incorporates human sensory compensation principles into control design. By integrating robust control theory, multi-agent coordination algorithms, and probabilistic uncertainty modeling, the approach enables real-time, sensor-driven dynamic error suppression without requiring costly hardware upgrades. Contribution/Results: Experimental evaluation demonstrates substantial reductions in positioning deviation and task failure rate during micrometer-scale operations. The proposed method achieves uncertainty suppression performance comparable to high-end precision sensor-based solutions, thereby establishing a novel pathway toward low-cost, high-robustness automation for micro-manufacturing.

This study addresses the limitations of traditional process capability indices, such as Cpk, which rely on deterministic thresholds under finite sample sizes and neglect estimation uncertainty, often leading to misclassification in critical regions. To overcome this, the work reframes capability assessment as a decision-risk calibration problem and introduces a hybrid framework that integrates a statistical baseline with data-driven residual learning—marking the first incorporation of uncertainty quantification into process capability evaluation. The approach leverages an interpretable baseline to model prior structural assumptions while employing a residual network to capture deviations due to non-normality, measurement error, and small-sample bias. Decision-risk calibration is achieved through nested Monte Carlo simulation. Experimental results demonstrate that the proposed framework significantly improves calibration accuracy and stability in critical regions, remains robust under leakage-free evaluation, and is readily deployable within existing industrial systems.

Industrial prediction and soft sensing often fail due to field data suffering from bias, latency, or seemingly plausible yet unreliable measurements. This work proposes a large language model (LLM)-guided Measurement Credibility Correction (MCC) method that, for the first time, leverages semantic information from process documentation to construct an external reference—requiring neither numerical correlations, fault labels, nor explicit process equations—for lightweight pre-inference correction. MCC translates document semantics into reference signals compatible with numerical models and integrates them at the front end of the prediction pipeline. Evaluated on multiple real-world industrial tasks, MCC reduces average relative MAE by 30.7% on authentic test data and by 80.3% under controlled contamination, while adding only 0.5–2.0k online parameters and incurring a maximum inference latency of 0.089 ms per step.

This work addresses the critical challenge of jointly designing sensor query rates and noise covariance under resource and cost constraints to meet prescribed trajectory estimation accuracy requirements. It presents the first formalization of this problem as a unified optimization model, leveraging semidefinite programming (SDP) within the Kalman filter error covariance framework to simultaneously optimize measurement scheduling and noise parameters. The proposed approach efficiently determines whether a given accuracy target is achievable and, when feasible, synthesizes a corresponding implementation strategy. Experimental validation demonstrates that the computed sensor configurations consistently attain the desired accuracy in both simulated and real-world scenarios, while also reliably identifying infeasible accuracy demands.

This work proposes a novel representation learning framework that addresses the limited representational capacity of existing methods in complex scenes by integrating adaptive multi-scale fusion with contrastive learning. The approach dynamically aggregates multi-level features and incorporates a structure-aware contrastive loss, thereby enhancing the model’s ability to jointly capture fine-grained semantics and global contextual information. Extensive experiments demonstrate that the proposed framework consistently outperforms state-of-the-art methods across multiple benchmark datasets, achieving substantial improvements in both accuracy and robustness. These results establish a promising new direction for unsupervised and semi-supervised representation learning.