Traditional scheduling theory assumes deterministic or independently random process behavior, failing to capture the temporal patterns in real systems. This work presents a predictive framework that models processes as discrete-time Markov chains with learnable behavioral states and treats resource heterogeneity as an optimization opportunity. The central contribution is the EWMA-Markov predictor, combining exponentially weighted averaging with Markovian modeling through P_ij(t+1) = α P_ij(t) + (1-α )𝕀[transition] , achieving O(1) complexity and numerical stability. For convex cost functions, we prove the cost under predictive scheduling satisfies Cost_predictive≤Cost_reactive(1 - D(𝒫)H(ℛ)/(1+κ )) , where D(𝒫) measures process diversity, H(ℛ) resource heterogeneity, and κ switching overhead. This quantifies when heterogeneous systems outperform homogeneous ones. Robustness analysis shows prediction errors cause only linear degradation: Cost_achieved≤Cost_optimal + 2ε·diameter(ρ ) for error ε . The framework maintains fairness with bounded distortion and integrates into production schedulers. While validated on CPU scheduling, the mathematical structure generalizes to any domain with stochastic demand and heterogeneous supply.
This paper presents a novel methodology and tool for automatic workload distribution in federated learning environments, in order to address the challenges of security and privacy in distributed machine learning systems. The proposed approach is designed to optimize the performance of federated learning by adapting workload distribution based on the heterogeneous nature of edge devices. The methodology is implemented through a Jupyter notebook extension that facilitates the execution of federated learning tasks in a distributed computing context, leveraging Docker for containerization and an integrated skeleton based compiler for parallelization tasks and environments configuration, through decorators directly in the cells. The Jupyter Workflow kernel leveraging the streamflow library in order to execute workflow in heterogeneus environments. The paper discusses the implementation details, privacy preservation mechanisms, and performance evaluation of the proposed solution, demonstrating its effectiveness in enhancing federated learning workflows.
In the current environmental context, characterized by an ever-increasing focus on ecosystem protection and climate change, there is a need to develop new technologies capable of monitoring environmental protection and potential adverse scenarios. Resilient Artificial Intelligence (Resilient AI) is a new technology useful for ensuring the operational continuity, accuracy, and safety of critical systems. It enables the detection of anomalies, predicting extreme events, and enabling timely decisions to support environmental protection. Resilient AI promotes the adoption of renewable energy sources, such as photovoltaics, by integrating them into smart grids, improving their efficiency and stability, reducing the risk of outages, minimizing false alarms, and supporting the energy transition towards more sustainable, reliable, and safe systems. The goal of this paper is to provide a comprehensive overview of current and future developments in Resilient AI, highlighting its potential, key challenges, and application prospects in various sectors. Through targeted simulations, the vulnerabilities of AI systems deployed in critical infrastructures essential to daily life (e.g., electricity grids, water, transportation, and telecommunications) will be analyzed, and strategies to improve their security and operational robustness will be evaluated.
Public blockchains offer transparency and security guarantees, but suffer from limited throughput, high latency, and unpredictable transaction costs, making them unsuitable for the high-frequency data flows of the Internet of Things. In this work, we address these scalability challenges by introducing a hybrid architecture that leverages state channels to enable efficient cross-chain smart contract execution between a permissioned ledger, responsible for capturing pollutant measurements from IoT sensors, and a public blockchain, used for final settlement. Our approach enables the processing of sensor readings to be performed off-chain in a lightweight, low-cost environment, while preserving the integrity and non-repudiation guarantees of on-chain settlement through a state channel. We describe the protocol design in detail, and the use of our solution in a case study involving water quality sensors.
Load forecasting plays a pivotal role in industrial demand response, enabling businesses to plan their electricity needs ahead of time through day-ahead scheduling. However, data is often limited or outdated due to frequent infrastructure modifications. To this end, load forecasting using limited data has recently attracted research interest. This paper explores the application of learning-based algorithms to day-ahead load forecasting in data-constrained environments. Moreover, we introduce Diff-Ensemble, an ensemble incorporating the long short-term memory (LSTM) and diffusion model, to enhance load forecasting capabilities when data is limited, and evaluate it using real-world data from an industrial site as part of the InStaFlex project. Results show that Diff-Ensemble reduces the normalized MAE (NMAE) by 8.1
