multisource signal integration

This paper addresses the deployment of multi-model inference pipelines on resource-constrained edge devices. We propose an end-to-end adaptive configuration framework that, for the first time, explicitly incorporates device resource constraints into joint optimization decisions. Our method integrates residual feature extraction, LSTM-based workload forecasting, and policy-gradient reinforcement learning to jointly optimize QoS guarantees (e.g., latency and throughput), operational cost, and real-time adaptability. Evaluated on a real Kubernetes-based edge cluster, the framework achieves a 27% reduction in average inference latency, a 31% increase in throughput, a 22% decrease in deployment cost, and over 42% faster configuration decision-making for complex pipelines—outperforming state-of-the-art baselines. The core contributions are: (i) a resource-aware joint optimization model that unifies hardware constraints with pipeline scheduling and scaling decisions; and (ii) a lightweight, learning-driven configuration mechanism enabling efficient, online adaptation under dynamic edge conditions.

This work addresses the unreliability of developer productivity dashboards, which often stems from ad hoc scripts that introduce undetected silent data gaps, eroding organizational trust. To resolve this, we propose a robust ELT pipeline grounded in DAG-based orchestration and the Medallion architecture, decoupling data extraction from transformation to preserve the immutability of raw data. Our approach introduces a state-driven dependency scheduling mechanism and, for the first time, treats metric pipelines as production-grade distributed systems. We emphasize the critical role of immutable raw history in enabling reliable metric redefinition. This methodology significantly enhances data reliability and freshness while effectively eliminating silent failures, thereby restoring organizational confidence in DevOps metrics.

Parameter identification remains a critical bottleneck for predictive modeling and control using mechanistic ordinary differential equation (ODE) models, hindered by noisy data, model misspecification, implementation complexity, and stringent differentiability requirements. This paper introduces an intelligent AI workflow tailored for ODE systems: it automatically compiles high-performance, differentiable JAX functions from XML-based model specifications and Python skeleton code; proposes a proxy-based AI architecture enabling model-code consistency verification and automated error correction; and integrates a two-stage parameter estimation strategy combining global search with gradient-based optimization. The framework substantially lowers the barrier to mechanistic modeling, delivering an end-to-end, reproducible, and auditable parameter estimation pipeline. An open-source implementation significantly reduces manual coding and debugging effort while ensuring numerical robustness and computational efficiency.

This work addresses the problem of efficiently merging multiple fine-tuned models into a unified multitask model without retraining. The authors formalize model merging as a convex quadratic program over residual updates, achieving theoretically optimal fusion by calibrating inputs and outputs to minimize calibration error in the output space. This study provides the first formal optimality guarantees for model merging, introduces an interpretable diagnostic metric based on residual energy, and unifies existing heuristic approaches within a single theoretical framework as special cases. Experimental results demonstrate that the proposed method matches or surpasses current techniques in single-layer settings and consistently improves performance across language and vision benchmarks in multilayer merging scenarios. Furthermore, the quality of merged models can be accurately predicted using a small calibration set.

Addressing the challenge of smooth finite-dimensional parameter estimation under weak alignment—where multi-source data exhibit partial,而非 perfect, correspondence and fully aligned samples are scarce—this paper proposes a novel semiparametric data fusion method. We establish, for the first time, the semiparametric efficiency bound under weak alignment and develop a theoretically grounded, robust estimator that jointly models alignment uncertainty and leverages auxiliary information, thereby substantially reducing reliance on fully aligned samples. Our approach relaxes the stringent strong-alignment assumption inherent in conventional fusion frameworks. Applied to an HIV monoclonal antibody prevention trial, it successfully quantifies the association between neutralizing antibodies and viral genotypes, demonstrating improved statistical efficiency and practical applicability. Key contributions include: (i) derivation of the semiparametric efficiency bound under weak alignment; (ii) a computationally feasible, robust fusion algorithm with provable efficiency; and (iii) interpretable, real-world validation in a clinical setting.

This work addresses the sensitivity of Mixture-of-Experts (MoE) models to parameter perturbations during model merging, which often leads to routing collapse and severe performance degradation. The study is the first to identify this issue and introduces Hessian-aware Router Calibration (HARC), a training-free framework that leverages second-order curvature information from the Hessian matrix to analytically realign the router post-merging, thereby restoring its routing capability. By integrating matrix-free conjugate gradient methods with Top-k routing analysis, HARC significantly enhances the performance of various MoE merging baselines on mathematical reasoning and code generation tasks, effectively mitigating routing failure without additional training.

This work addresses the challenge scientists face in efficiently transforming raw sensor data streams into actionable insights across edge-cloud infrastructures, hindered by the need for cross-domain expertise to manage heterogeneous systems and emerging platforms such as DPUs, which impedes rapid prototyping. To overcome this barrier, the authors propose a novel paradigm that integrates pattern-based workflow engineering with AI-assisted development. Implemented on the FABRIC testbed using the Pegasus workflow system and exemplified by the Orcasound hydrophone workflow, this approach enables swift construction of applications for air quality, seismic, and soil moisture monitoring. The framework supports modular extensibility and edge deployment, substantially lowering the barrier for non-expert users to iteratively develop distributed applications. Empirical validation across multiple use cases demonstrates its effectiveness in enhancing development efficiency, accelerating prototyping cycles, and accumulating practical deployment experience.

Existing autonomous driving datasets lack sufficient diversity, coordination, and cross-domain support, limiting their utility for training multi-agent, multi-sensor systems. To address this gap, this work proposes a modular data generation pipeline built upon the AVstack framework and the CARLA simulator, capable of efficiently producing terabyte-scale, ground-truth-annotated multimodal data. The pipeline encompasses perspectives from ground vehicles, aerial platforms, and infrastructure sensors, and supports flexible single- or multi-agent configurations under controllable, complex scenarios. This approach represents the first scalable, cross-domain collaborative data generation methodology for autonomous driving, substantially enhancing the customization, training efficacy, and practical applicability of perception and sensor fusion models in cooperative autonomous systems.

This work addresses the complexity of developing edge-to-cloud sensor applications, which typically requires cross-domain collaboration and hinders efficient transformation of raw data into actionable insights. To streamline this process, the authors propose an intent-driven, AI-assisted rapid development methodology that integrates reusable workflow patterns with intelligent configuration, enabling seamless edge adaptation and deployment without code rewriting. Built upon the Pegasus workflow system and deployed on the FABRIC testbed, the approach supports heterogeneous edge resources such as BlueField-3 DPUs and Raspberry Pi devices. Users can construct multi-stage sensing applications within 1–1.5 days, and the framework’s robustness and portability have been validated through real-world deployments in air quality, seismic activity, and soil moisture monitoring scenarios.

This study investigates the transferability of additive activation interventions from single-turn chat models to tool-augmented ReAct agents, where their efficacy and safety implications remain unclear. Employing matched information designs and combining behavioral measurements, representational readouts, and directional ablations—while controlling for KV cache interference—the work uncovers a novel “strength-preserved but behavior-decoupled” phenomenon: intervention signals propagate nearly intact through deep layers, yet their behavioral effects diverge substantially across models and contexts. The ReAct scaffolding format—not tool observations—primarily governs effect rescaling. On models such as Qwen2.5-7B, refusal-bypassing effects are amplified up to 2.00×, while other models exhibit marked attenuation, revealing that safety impacts are highly unpredictable upon deployment.