online evaluation

This work addresses the vulnerability of surrogate models in digital twins to concept drift under shifting operational conditions, which degrades both predictive accuracy and uncertainty quantification. To mitigate this, the authors propose an adaptive digital twin framework that integrates multivariate distribution drift detection based on Fisher scores, a parameter-efficient LoRA-based continual learning mechanism for model adaptation, and Mann-Whitney U test–driven online statistical validation to assess the necessity of updates, enable efficient fine-tuning, and ensure reliability. Evaluated on a stochastic linear system and a directed energy deposition additive manufacturing task, the approach significantly accelerates drift detection, enhances model recovery accuracy, and improves the quality of uncertainty estimates, thereby enabling trustworthy continuous deployment of surrogate models.

Existing optimization methods for autonomous systems are often task-specific and lack a unified protocol to support multi-round, traceable, and stable optimization processes. This work proposes EPOCH—the first general-purpose multi-round optimization protocol designed for heterogeneous environments. EPOCH enables coordinated refinement of multiple components, including prompts, model configurations, code, and rules, through a two-stage process comprising baseline construction and iterative self-improvement. It integrates role-constrained planning, implementation, and evaluation workflows, standardized command interfaces, and round-level tracking mechanisms. Empirical validation across diverse tasks demonstrates EPOCH’s effectiveness in production-grade autonomous optimization workflows, while consistently ensuring stability, reproducibility, traceability, and comprehensive evaluation throughout the optimization process.

This work addresses the vulnerability of federated learning in resource-constrained, heterogeneous industrial networks, where unreliable clients, sensing noise, and malicious updates undermine system integrity. Existing trust mechanisms often rely on fixed parameters or simplistic adaptive rules, rendering them ill-suited for dynamic environments. To overcome these limitations, we propose a lightweight, server-side proxy-based trust coordination framework that decouples observation, reasoning, and action. Without altering client-side procedures or increasing communication overhead, our approach establishes a closed-loop control mechanism that continuously monitors time-series data of system behavior and trust signals, autonomously infers their evolutionary trends, and precisely adjusts trust policies upon detecting instability. This method significantly enhances the robustness and stability of federated learning in dynamic industrial settings, enabling sustainable and resilient distributed intelligent collaboration.

Traditional neural network training relies on fixed optimization pipelines, rendering it inflexible in dynamically addressing training instability and anomalies. To address this limitation, we propose the first interactive training framework enabling real-time human–AI collaborative intervention. Our method employs a lightweight control server that integrates expert human directives with AI agent feedback to dynamically adjust hyperparameters, data sampling strategies, and model checkpoints during training. This framework introduces, for the first time, a closed-loop interactive paradigm into neural network training, establishing a scalable human–machine collaboration interface coupled with automated response mechanisms. Experimental results demonstrate significant improvements in training stability, reduced sensitivity to initial hyperparameter configurations, and enhanced real-time responsiveness to user-specified customization requirements. The effectiveness is validated across multiple benchmark tasks.

This work addresses the challenges of stale trajectory data and skewed sequence lengths in asynchronous reinforcement learning post-training, which often degrade convergence and imbalance system throughput. To this end, we propose StaleFlow, the first system that jointly mitigates these issues by enforcing global consistency protocols to bound data staleness, adopting a decoupled data server architecture that separates trajectory storage from model parameters, and introducing a staleness-aware, high-throughput rollout coordination mechanism. Experimental results demonstrate that StaleFlow achieves 1.17–2.68× higher throughput compared to existing systems while preserving model convergence guarantees.

This work addresses the challenge that AI agents with frozen weights after deployment struggle to learn continuously from experience, often failing on repeated tasks. The authors propose a continual learning mechanism leveraging external memory, which distills minimal feedback—either a single-bit outcome or natural language corrections—from each interaction into retrievable rules. Integrated with retrieval-augmented generation (RAG) and frozen large language models (e.g., Mistral Large, Claude Sonnet 5), this approach enables performance improvement without fine-tuning. The method demonstrates, for the first time, that extremely sparse feedback alone can drive sustained enhancement in frozen models and supports memory transfer across models. On the τ-bench banking tasks, it achieves success rates 1.6× (outcome-only feedback) and 2.6× (with corrections) higher than baseline, resolving 22 out of 84 tasks on which the baseline completely fails.

This work addresses the challenge that small language models often fail to reliably execute multi-step, dependency-rich structured graph algorithms due to error accumulation. The authors frame algorithm execution as a closed-loop prediction task, wherein the model iteratively selects operations based on the current graph state and evaluates its overall behavior through full rollbacks. Departing from conventional step-isolated evaluation, this closed-loop rollback paradigm reveals that strong single-step prediction accuracy does not necessarily ensure stable global execution. Experimental results demonstrate that suitably adapted small models can reliably perform algorithms such as traversal and coloring, yet remain vulnerable to cumulative errors in weighted graph algorithms. These findings underscore the necessity and efficacy of the proposed closed-loop evaluation framework for assessing and improving algorithmic reasoning in language models.

This work addresses the challenges of device heterogeneity, varying layer-wise susceptibility to catastrophic forgetting, and long-term distribution drift in federated continual learning for mobile autonomous systems. The authors propose a lifespan-aware, dual-timescale federated continual learning framework that introduces layer-selective replay and a post-deployment rapid knowledge recovery mechanism, jointly mitigating forgetting during both training and inference. Theoretical analysis reveals heterogeneous forgetting dynamics across network layers and demonstrates that long-term performance degradation is inevitable under conventional assumptions, thereby overcoming the limitation of prior methods that focus solely on the training phase. Evaluated on a real Mars rover platform, the proposed approach achieves an 8.3% improvement in mIoU over the strongest federated baseline and a 31.7% gain over standard fine-tuning, significantly enhancing stability and adaptability in long-horizon tasks.

Current evaluations of AI systems predominantly rely on static benchmarks, which fail to capture behavioral risks in dynamic real-world environments. This work formalizes AI auditing as an uncertainty-aware, dynamic constraint monitoring problem across the system’s entire lifecycle, targeting critical attributes such as fairness and safety while integrating sociotechnical norms with statistical risk control. By developing a theoretical framework and supporting infrastructure for continuous auditing, the study advances AI governance beyond one-off testing toward ongoing, reliable, and accountable oversight mechanisms.

This work addresses the lack of a general, auditable dynamic control mechanism in existing training systems, which typically rely on framework-specific code. The authors propose the first cross-framework, open-source control plane that exposes training interfaces through a unified protocol, integrating declarative configuration, request validation, and secure control-point scheduling within the Aim workspace to enable metric monitoring, real-time intervention, and operational traceability. The system supports safe human and automated controller interventions during training while fully logging all operational trajectories. Experiments across five NLP and reinforcement learning tasks demonstrate its effectiveness, and the open-source implementation provides a foundation for reproducible human-in-the-loop training.