Selected work

Representative Papers

FedDOSE: Federated Learning Framework Decomposing Site Effects for Modeling Brain Dynamic Functional Connectivity

Aug 07, 2026

This work addresses the challenge of site-induced statistical heterogeneity in multi-site fMRI data within federated learning, a problem exacerbated by the frequent neglect of the brain’s dynamic functional characteristics. To this end, the authors propose FedDOSE, a novel framework that jointly models dynamic functional connectivity (dFC) and site heterogeneity for the first time in a federated setting. FedDOSE efficiently encodes high-dimensional dFC tensors via module-guided Tucker decomposition and aligns cross-site class prototypes through an integration of optimal transport centroids and Procrustes analysis. Evaluated on the ABIDE-I, ABIDE-II, and ADHD-200 datasets, the method demonstrates superior performance over existing approaches, exhibiting enhanced generalization and robustness in diagnostic tasks for autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD).

0 citationsRead paper

FedChronos: Federated Fine-Tuning of Time-Series Foundation Models for Privacy-Preserving Commodity Price Forecasting

Aug 02, 2026

This work addresses the challenge of forecasting in agricultural markets where data cannot be centralized due to regulatory or sovereignty constraints and exhibit heterogeneous distributions. The authors propose FedChronos, a framework that performs federated parameter-efficient fine-tuning on the pre-trained time series foundation model Chronos-T5. By integrating low-rank adaptation (LoRA) with differential privacy, the method transmits only approximately 384 KB of adapter parameters per communication round, marking the first integration of federated learning with efficient fine-tuning of time series foundation models. Experimental results on commodity price prediction across 15 Indian markets show that the injected differential privacy noise acts as implicit regularization, mitigating overfitting in low-data regimes. Under the optimal configuration (ε=5), FedChronos reduces MAPE by 31% compared to zero-shot inference and by 26% against conventional baselines, while providing (ε,δ)-differential privacy guarantees.

0 citationsRead paper

XGRVFL-MV: Residual-Coupled Graph-Embedded Multi-View Random Vector Functional Link Network with FleXi Guardian Loss

Jul 25, 2026

This work addresses key challenges in multi-view classification—preserving geometric structure, modeling inter-view relationships, and ensuring robustness to large residuals—by proposing a novel multi-view Random Vector Functional Link (RVFL) network that integrates graph embedding with a residual coupling mechanism. The approach constructs intrinsic and penalty graphs via locality-preserving Fisher discriminant analysis to retain the geometric structure of each view, introduces a residual coupling term to enforce prediction consistency across views, and incorporates, for the first time in multi-view RVFL networks, a bounded asymmetric FleXi Guardian loss to enhance robustness, optimized via Nesterov-accelerated gradient descent. Extensive experiments on UCI, KEEL, AwA, and Corel5k datasets demonstrate statistically significant superiority over state-of-the-art methods, with sensitivity analyses further confirming the model’s effectiveness and stability.

0 citationsRead paper

Twin-Fidelity-Aware Resolution of Direct xApp Conflicts in Open RAN

Jul 24, 2026

This work addresses the conflict between energy-saving and coverage/throughput-oriented xApps in Open RAN regarding downlink power configuration by proposing a lightweight, training-free arbitration mechanism that requires no prior knowledge of optimal policies. The approach employs a dual-fidelity-aware hard-switching strategy that online fuses power recommendations from both xApp types. It leverages a network digital twin to predict the utility of candidate actions and dynamically selects the optimal action based on real-time utility feedback and exponential weighted moving average error monitoring. System-level 5G evaluations demonstrate a normalized utility regret as low as 0.017 ± 0.006. Under severe digital twin drift (10 dB), the utility regret drops significantly from 11.19 ± 3.58 to 0.55 ± 0.25, substantially outperforming baseline methods.

0 citationsRead paper

STS-NET: Spatio-Temporal Stress Network for Self-Supervised Crop Stress Detection using Satellite Image Time Series

Jul 21, 2026

This study addresses the scarcity of labeled data in early crop stress detection by proposing STS-Net, the first self-supervised spatiotemporal network tailored for this task. Built upon a 3D convolutional autoencoder, the method leverages high-resolution PlanetScope time-series imagery along with four vegetation indices—NDVI, GNDVI, RECI, and NDRE—to learn spatiotemporal stress patterns without requiring extensive annotations. Evaluated on real-world sugarcane fields, STS-Net achieves detection accuracies of 97.98%, 85.08%, and 83.47% for water stress, nitrogen stress, and combined stress, respectively, demonstrating both high classification accuracy and strong generalization capability.

0 citationsRead paper

Recent publications

Latest Papers

FedDOSE: Federated Learning Framework Decomposing Site Effects for Modeling Brain Dynamic Functional Connectivity

Aug 07, 2026

This work addresses the challenge of site-induced statistical heterogeneity in multi-site fMRI data within federated learning, a problem exacerbated by the frequent neglect of the brain’s dynamic functional characteristics. To this end, the authors propose FedDOSE, a novel framework that jointly models dynamic functional connectivity (dFC) and site heterogeneity for the first time in a federated setting. FedDOSE efficiently encodes high-dimensional dFC tensors via module-guided Tucker decomposition and aligns cross-site class prototypes through an integration of optimal transport centroids and Procrustes analysis. Evaluated on the ABIDE-I, ABIDE-II, and ADHD-200 datasets, the method demonstrates superior performance over existing approaches, exhibiting enhanced generalization and robustness in diagnostic tasks for autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD).

0 citationsRead paper

FedChronos: Federated Fine-Tuning of Time-Series Foundation Models for Privacy-Preserving Commodity Price Forecasting

Aug 02, 2026

This work addresses the challenge of forecasting in agricultural markets where data cannot be centralized due to regulatory or sovereignty constraints and exhibit heterogeneous distributions. The authors propose FedChronos, a framework that performs federated parameter-efficient fine-tuning on the pre-trained time series foundation model Chronos-T5. By integrating low-rank adaptation (LoRA) with differential privacy, the method transmits only approximately 384 KB of adapter parameters per communication round, marking the first integration of federated learning with efficient fine-tuning of time series foundation models. Experimental results on commodity price prediction across 15 Indian markets show that the injected differential privacy noise acts as implicit regularization, mitigating overfitting in low-data regimes. Under the optimal configuration (ε=5), FedChronos reduces MAPE by 31% compared to zero-shot inference and by 26% against conventional baselines, while providing (ε,δ)-differential privacy guarantees.

0 citationsRead paper

XGRVFL-MV: Residual-Coupled Graph-Embedded Multi-View Random Vector Functional Link Network with FleXi Guardian Loss

Jul 25, 2026

This work addresses key challenges in multi-view classification—preserving geometric structure, modeling inter-view relationships, and ensuring robustness to large residuals—by proposing a novel multi-view Random Vector Functional Link (RVFL) network that integrates graph embedding with a residual coupling mechanism. The approach constructs intrinsic and penalty graphs via locality-preserving Fisher discriminant analysis to retain the geometric structure of each view, introduces a residual coupling term to enforce prediction consistency across views, and incorporates, for the first time in multi-view RVFL networks, a bounded asymmetric FleXi Guardian loss to enhance robustness, optimized via Nesterov-accelerated gradient descent. Extensive experiments on UCI, KEEL, AwA, and Corel5k datasets demonstrate statistically significant superiority over state-of-the-art methods, with sensitivity analyses further confirming the model’s effectiveness and stability.

0 citationsRead paper

Twin-Fidelity-Aware Resolution of Direct xApp Conflicts in Open RAN

Jul 24, 2026

This work addresses the conflict between energy-saving and coverage/throughput-oriented xApps in Open RAN regarding downlink power configuration by proposing a lightweight, training-free arbitration mechanism that requires no prior knowledge of optimal policies. The approach employs a dual-fidelity-aware hard-switching strategy that online fuses power recommendations from both xApp types. It leverages a network digital twin to predict the utility of candidate actions and dynamically selects the optimal action based on real-time utility feedback and exponential weighted moving average error monitoring. System-level 5G evaluations demonstrate a normalized utility regret as low as 0.017 ± 0.006. Under severe digital twin drift (10 dB), the utility regret drops significantly from 11.19 ± 3.58 to 0.55 ± 0.25, substantially outperforming baseline methods.

0 citationsRead paper

STS-NET: Spatio-Temporal Stress Network for Self-Supervised Crop Stress Detection using Satellite Image Time Series

Jul 21, 2026

This study addresses the scarcity of labeled data in early crop stress detection by proposing STS-Net, the first self-supervised spatiotemporal network tailored for this task. Built upon a 3D convolutional autoencoder, the method leverages high-resolution PlanetScope time-series imagery along with four vegetation indices—NDVI, GNDVI, RECI, and NDRE—to learn spatiotemporal stress patterns without requiring extensive annotations. Evaluated on real-world sugarcane fields, STS-Net achieves detection accuracies of 97.98%, 85.08%, and 83.47% for water stress, nitrogen stress, and combined stress, respectively, demonstrating both high classification accuracy and strong generalization capability.

0 citationsRead paper