Category
50 papers
The increasing use of electronic medical records (EMRs) has improved efficiency, accuracy, and accessibility of patient data. However, conventional centralized architectures suffer from single points of failure and data privacy issues. To address these challenges, this study proposes a decentralized machine learning architecture that combines concepts from Federated Learning (FL) and Swarm Learning (SL) for anomaly detection in medical prescriptions. The proposed architecture leverages blockchain and the InterPlanetary File System (IPFS) to enable secure model sharing and decentralized storage, thereby reducing communication complexity and establishing a transparent, decentralized parameter repository. Experimental evaluations were conducted using logistic regression (LR), a multi-layer perceptron (MLP), and a decision tree (DT) model. Compared with the FL baseline, the proposed system achieved superior efficiency, lower resource consumption, and improved latency, along with smaller block sizes. It, however, exhibited slightly lower transaction throughput and longer training rounds, reflecting the added complexity of decentralization. In predictive performance on the anomaly classification task, DT achieved the highest precision and recall under the evaluated dataset (F1-score=0.9912), followed by MLP (0.5504) and LR (0.2442). The decentralized training approach led to negligible performance loss relative to centralized models, less than 4% for LR and below 1% for both MLP and DT. Overall, the proposed system demonstrates a robust and efficient alternative for decentralized learning in healthcare applications, maintaining strong predictive performance while enhancing architectural transparency.
Federated Learning (FL) allows edge clients to collaborate in model training with data privacy preserved, yet it is known to suffer low training efficiency and model accuracy. Given that efficiency and accuracy are usually conflicting objectives, existing practices increasingly employ an adaptive scheme that changes the FL configurations (e.g., quantization or sparsification level) based on runtime training status, for which accurate training diagnosis—used for guiding the optimization actions—is crucial. However, while training diagnosis is a common task shared by different optimization schemes, existing works propose their diagnosis methods in an ad-hoc manner, which yield multiple limitations. First, the diagnosis metric in an optimization scheme may sometimes be less accurate than others; second, existing schemes fail to fully exploit the diagnosis result by applying it for only one optimization action; third, existing methods usually do not perceive cross-client data heterogeneity, failing to simultaneously enhance FL accuracy. To tackle those limitations, we make a systematical study on the training diagnosis methods of multiple optimization schemes, and propose metric grafting—replacing a scheme’s diagnosis metric with a better one to improve the training performance. Moreover, to fully exploit the potential of training diagnosis, we build a system platform that supports flexible combinations of training diagnosis and optimization actions (i.e., single-diagnosis-multiple-actions and multiple-diagnosis-multiple-actions). Evaluation on testbeds show that, with metric grafting and advanced diagnosis-action combinations, we can substantially improve the efficiency and accuracy performance of FL.
Spatial Mobile Crowdsourcing (SMC) faces the dual challenge of ensuring privacy while managing dynamic, Non-IID spatio-temporal data. While Federated Learning (FL) offers a privacy-preserving solution, traditional aggregation suffers from severe model drift due to evolving spatio-temporal contexts. Furthermore, existing approaches often decouple prediction from scheduling, failing to translate predictive insights into tangible task allocation efficiency. To address these challenges, we propose STDFL, a prediction-driven dynamic framework tailored for SMC. It features a hierarchical architecture combining client-side lightweight models for micro-patterns and a server-side Transformer for global dependencies. To mitigate drift, our Dynamic Spatio-Temporal Perceiving Aggregation adaptively weights updates based on spatial similarity and temporal freshness. For privacy, we integrate client-level <inline-formula><tex-math notation="LaTeX">$(\epsilon, \delta )$</tex-math><alternatives><mml:math><mml:mrow><mml:mo>(</mml:mo><mml:mi>ε</mml:mi><mml:mo>,</mml:mo><mml:mi>δ</mml:mi><mml:mo>)</mml:mo></mml:mrow></mml:math><inline-graphic xlink:href="wang-ieq1-3673177.gif"/></alternatives></inline-formula>-Differential Privacy to ensure formal protection. Finally, we introduce a Prediction-Driven Scheduler (PDS) that leverages predictive potentials for bipartite matching, theoretical analysis and stress tests confirm PDS achieves linear scalability and zero policy training cost, offering a superior real-time deployment trade-off compared to RL approaches. Experiments on Chengdu and Nanjing datasets show STDFL significantly outperforms SOTA baselines in efficiency and fairness,while achieving near-centralized prediction accuracy under rigorous privacy guarantees and effectively bridging the utility–privacy trade-off in spatial crowdsourcing.
The integration of Internet of Things (IoT) and Federated Learning (FL) marks a significant step towards pervasiveness, supported by distributed computational resources and enhanced by 5G advancements. However, the efficient and sustainable operation of IoT-enabled FL systems faces critical challenges, particularly in selecting devices that balance energy consumption and system performance. To address this, we propose the Battery-Aware Dynamic and Automated Device Selection (DADS) framework, a novel solution specifically designed for IoT-enabled FL environments. DADS introduces an innovative adaptive mechanism that dynamically adjusts optimization parameters, such as inertia weights and crossover/mutation rates, ensuring a seamless balance between exploration and exploitation. Unlike conventional optimization approaches, DADS employs uniquely developed Adaptive Particle Swarm Optimization (APSO) and Adaptive Genetic Algorithm (AGA) within a cohesive framework. This design enables DADS to respond to dynamic IoT network conditions, such as fluctuating battery levels and device capabilities, ensuring energy-efficient device selection while preserving robust performance. Through extensive performance analysis, DADS demonstrates its novelty by achieving a 20% reduction in energy consumption and a 30% improvement in FL training time compared to state-of-the-art methods. Moreover, DADS significantly enhances battery lifespan, reducing degradation by more than 50%, and optimizes communication efficiency, extending the operational sustainability of IoT devices. These results position DADS as a groundbreaking framework, setting a new benchmark for energy-aware and sustainable IoT-enabled FL systems.
The rapid proliferation of 6G-enabled consumer electronics (CE) has introduced significant security and privacy challenges. Traditional security mechanisms often fall short in addressing these issues due to the unique characteristics of CE networks. To enhance intrusion detection systems (IDSs), data-driven artificial intelligence (AI) approaches have gained considerable attention. Nonetheless, AI-based IDSs face challenges related to scalability, privacy preservation, and high computational demands—particularly when handling high-dimensional and complex data. To overcome these limitations, this article proposes a novel framework called Blockchain and Quantum Federated Learning (BQFL). BQFL integrates blockchain technology and quantum computing (QC) with FL to provide an efficient, secure, robust, and privacy-preserving solution for intrusion detection in CE environments. Specifically, blockchain enables fault-tolerant and decentralized trust management for parameter aggregation on the quantum FL (QFL) server. Furthermore, the framework leverages the high-speed and low-latency capabilities of 6G networks to enable real-time and secure data processing and communication among a vast number of CE devices. We validate the effectiveness of the proposed framework through extensive experiments on the ToN-IoT dataset.
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
An expert in machine learning, statistics, and computation, Rakhlin succeeds Professor Ankur Moitra.
Machine Intelligence
Algorithms & Theory
Results from a randomized controlled trial show the potential of Gemini’s Guided Learning feature to boost engagement and accelerate learning.
