arXiv subjects

Xi Yao

Publications and source records attributed to Xi Yao.

8 recordsLinked to original sources

arXiv2026

Hierarchical Constrained Reinforcement Learning with Dynamic Boundary for Spatio-Temporal Vehicle-to-Grid Scheduling

The rapid proliferation of Electric Vehicles (EVs) introduces significant spatio-temporal uncertainties into power grids, while Vehicle-to-Grid (V2G) technology offers critical flexibility through bidirectional power flow. However, integrating large-scale EVs into the Optimal Power Flow framework presents substantial challenges due to computational bottlenecks arising from solver complexity and coupled spatio-temporal constraints. Existing Reinforcement Learning (RL) methods often struggle to balance strict constraint satisfaction with scalability in highly dynamic EV fleet environments. To address these challenges, this paper proposes a Hierarchical Policy for Constrained Reinforcement Learning (HPC-RL) framework for spatially and temporally coupled V2G scheduling. The framework adopts a two-layer architecture: the upper level utilizes a RL algorithm based on the Generalized Reduced Gradient method to strictly enforce spatial grid-level hard constraints; the lower level implements a novel dynamic boundary strategy to compute real-time feasible charging power bounds for individual EVs, thereby ensuring the satisfaction of temporal charging demands. This integrated design not only enables the simultaneous handling of spatially and temporally coupled constraints during the RL optimization process but also significantly enhances generalization capabilities for large-scale fleets through hierarchical decoupling. Extensive experiments on IEEE 14, 30, and modified 141-bus systems demonstrate that HPC-RL outperforms Model Predictive Control and state-of-the-art safe RL baselines across all metrics. The proposed method achieves near-optimal scheduling strategies and drastically reduces online computation time in large-scale scenarios from hours to minutes, while maintaining a near-zero constraint violation rate and nearly 100\% charging demand satisfaction.

cs.CE↗

arXiv2026

A Predict-then-Schedule framework for Power Distribution Networks with AI Data Centers

The surge of GPU-intensive workloads in artificial intelligence (AI) data centers drives massive energy demands, leading to soaring costs and significant stress on local power distribution networks. Coordinating delay-tolerant workload scheduling with power grid conditions via precise workload prediction can mitigate these issues. However, a critical gap remains in conventional approaches, i.e., minimizing prediction error does not necessarily lead to minimized downstream operational loss. Hence, this paper proposes an end-to-end Predict-Then-Schedule (PTS) framework that integrates upstream workload prediction with downstream scheduling optimization. By leveraging differentiable convex optimization, the PTS framework maps input features directly to optimal scheduling and enables gradient-based training. Furthermore, to respect the data center's capacity, a workload over-shifted loss combining electricity cost with a penalty for load-shedding is introduced to evaluate scheduling quality. Experiments demonstrate that the proposed framework significantly reduces operational cost and enhances system security compared to the conventional two-stage baseline.

cs.CE↗

arXiv2025

FLEx: Personalized Federated Learning for Mixture-of-Experts LLMs via Expert Grafting

Federated instruction tuning of large language models (LLMs) is challenged by significant data heterogeneity across clients, demanding robust personalization. The Mixture of Experts (MoE) architecture, where experts can specialize in distinct data patterns, presents a natural architectural solution to this challenge. The inherent sparsity of the MoE architecture, achieved by selectively activating experts, poses a significant challenge to its integration with federated learning (FL). Conventional FL frameworks, designed for dense models, naively aggregate all expert parameters irrespective of their local activation patterns. This naive approach not only undermines MoE's dynamic sparsity but also risks corrupting the world knowledge within pretrained experts. To address this, we propose FLEx (Federated LLMs with Personalized Experts), a novel framework that leverages pretrained MoE-based LLMs for efficient personalization. By aggregating only the shared non-expert parameters, FLEx significantly reduces communication overhead and preserves the world knowledge stored within the frozen pretrained experts. For personalization, we introduce a novel expert grafting mechanism that leverages dynamic sparsity to construct a client-specific expert from selected components of pretrained experts, tailored to local data. This grafted expert is then fine-tuned locally alongside the gating mechanism. This joint training enables the model to learn when to leverage the shared knowledge from frozen experts and when to employ the personalized one. Evaluations on diverse, non-IID instruction tuning datasets show that FLEx consistently outperforms federated baselines on average, while demonstrating strong knowledge preservation on the knowledge-driven benchmark MMLU. Our code is available at \href{https://anonymous.4open.science/r/FLEx-8F12}{\texttt{https://anonymous.4open.science/r/FLEx-8F12}}.

cs.AI↗

arXiv2025

Diffusion-based learning framework for Constrained Nonconvex Optimization with Weighted Bootstrapped Refinement

Recent advances in diffusion models show promising potential to accelerate nonconvex problem solving by leveraging their multimodality. However, most existing diffusion-based optimization approaches rely on supervised learning and lack a mechanism to enforce constraint satisfaction, which is required in real-world applications. In that case, we investigate and theoretically analyze the inherent problem of supervised diffusion solvers and identify the distributional misalignment problem, i.e., the generated solution distribution often exhibits low probability mass on the feasible region. To resolve this issue, we propose DiOpt, a new diffusion-based learning framework for constrained nonconvex optimization, which effectively learns the mapping from noise to the constraint region. Specifically, this framework operates in two distinct phases: an initial warm-start phase, implemented via supervised learning, followed by a bootstrapping training phase. This dual-phase architecture is designed to iteratively refine solutions, thereby improving the objective function with high constraint satisfaction. Finally, we also employ a solution selection technique in inference for better optimality. Notably, DiOpt is the first successful integration of the diffusion solver in constrained nonconvex optimization. Evaluations on diverse nonconvex tasks demonstrate the superiority of DiOpt in both optimality and constraint satisfaction. Our official page is released at https://dingsht.tech/diopt-webpage.

cs.LG↗

arXiv2024

A Low-rank Projected Proximal Gradient Method for Spectral Compressed Sensing

This paper presents a new approach to the recovery of a spectrally sparse signal (SSS) from partially observed entries, focusing on challenges posed by large-scale data and heavy noise environments. The SSS reconstruction can be formulated as a non-convex low-rank Hankel recovery problem. Traditional formulations for SSS recovery often suffer from reconstruction inaccuracies due to unequally weighted norms and over-relaxation of the Hankel structure in noisy conditions. Moreover, a critical limitation of standard proximal gradient (PG) methods for solving the optimization problem is their slow convergence. We overcome this by introducing a more accurate formulation and a Low-rank Projected Proximal Gradient (LPPG) method, designed to efficiently converge to stationary points through a two-step process. The first step involves a modified PG approach, allowing for a constant step size independent of signal size, which significantly accelerates the gradient descent phase. The second step employs a subspace projection strategy, optimizing within a low-rank matrix space to further decrease the objective function. Both steps of the LPPG method are meticulously tailored to exploit the intrinsic low-rank and Hankel structures of the problem, thereby enhancing computational efficiency. Our numerical simulations reveal a substantial improvement in both the efficiency and recovery accuracy of the LPPG method compared to existing benchmark algorithms. This performance gain is particularly pronounced in scenarios with significant noise, demonstrating the method's robustness and applicability to large-scale SSS recovery tasks.

eess.SP↗

arXiv2023

Emergent collective motion of self-propelled condensate droplets

Recently, there is much interest in droplet condensation on soft or liquid/liquid-like substrates. Droplets can deform soft and liquid interfaces resulting in a wealth of phenomena not observed on hard, solid surfaces (e.g., increased nucleation, inter-droplet attraction). Here, we describe a unique complex collective motion of condensate water droplets that emerges spontaneously when a solid substrate is covered with a thin oil film. Droplets move first in a serpentine, self-avoiding fashion before transitioning to circular motions. We show that this self-propulsion (with speeds in the 0.1-1 mm/s range) is fuelled by the interfacial energy release upon merging with newly condensed but much smaller droplets. The resultant collective motion spans multiple length scales from submillimetre to several centimetres, with potentially important heat-transfer and water-harvesting applications.

cond-mat.soft↗

arXiv2018

Novel high-pressure calcium carbonates

Calcium and magnesium carbonates are believed to be the host compounds for most of the oxidized carbon in the Earth's mantle. Here, using evolutionary crystal structure prediction method USPEX, we systematically explore the MgO-CO2 and CaO-CO2 systems at pressures ranging from 0 to 160 GPa to search for thermodynamically stable magnesium and calcium carbonates. While MgCO3 is the only stable magnesium carbonate, three calcium carbonates are stable under pressure: well-known CaCO3, and newly predicted Ca3CO5 and CaC2O5. Ca3CO5 polymorphs are found to contain isolated orthocarbonate (CO4)4- tetrahedra, and are stable at relatively low pressures (>11 GPa), whereas CaC2O5 is stable above 33 GPa and its polymorphs feature polymeric motifs made of CO4 tetrahedra. Detailed analysis of chemical stability of CaCO3, Ca3CO5 and CaC2O5 in the environment typical of the Earth's lower mantle reveals that none of these compounds can exist in the Earth's lower mantle. We conclude that MgCO3 is the main host of oxidized carbon throughout the lower mantle.

cond-mat.mtrl-sci↗

arXiv2017

Liquid jet rebound upon impact on a soft gel

A liquid jet can stably bounce off a sufficiently soft gel, by following the contour of the dimple created upon impact. This new phenomenon is insensitive to the wetting properties of the gels and was observed for different liquids over a wide range of surface tensions, $γ$ = 24--72 mN/m. In contrast, other jet rebound phenomena are typically sensitive to $γ$: jet rebounds off a hard solid (e.g. superhydrophobic surface) or another liquid are possible only for high and low $γ$ liquids, respectively. This is because an air layer must be stabilized between the two interfaces. For a soft gel, no air layer is necessary and jet rebound remains stable even when there is direct liquid-gel contact.

physics.flu-dyn↗