arXiv subjects

Mu Wang

Publications and source records attributed to Mu Wang.

At least 19 recordsLinked to original sources

arXiv2026

CAP-DO: Learned Contextual Action Proposals for Certified Double-Oracle Solving Across Related Zero-Sum Games

Many security and inspection-planning problems require solving a sequence of related zero-sum games. Across this sequence, the feasible defender and attacker action spaces re-main fixed, whereas each context induces a different payoff matrix through changes in target values, inspection effective-ness, costs, and interaction effects. Double Oracle (DO) solves large zero-sum games without materializing the full payoff matrix by iteratively expanding a restricted game. However, applying standard DO independently to each new payoff context requires restarting the search from a generic restricted game and requires rediscovering context-relevant actions through full-space best responses. We propose Con-textual Action Proposal Double Oracle (CAP-DO), a learning-augmented framework that warm-starts DO for repeated contextual games. Offline, CAP-DO trains separate defender and attacker rankers once from solved contexts. Online, the fixed rankers propose initial restricted action sets for each new context. Learning therefore determines where certified search starts, while the current game, through full-space best-response checks and a two-sided certificate, still determines whether the output is accepted. Theoretically, CAP-DO pre-serves DO's full-game certification guarantee, so every accepted output meets the prescribed certificate tolerance. Under standard exact-oracle assumptions, CAP-DO also retains finite convergence when expansion is uncapped. Empirically, across three scales of a non-additive contextual inspection-game benchmark, with up to 9,880 actions per player, CAP-DO-balanced achieves higher certification rates and uses few-er full-space best-response calls than cold-start, trace-reuse, and heuristic warm starts under fixed expansion budgets.

cs.GT↗

arXiv2026

Observation of the Inherent Chiral Smith-Purcell Effect via Symmetry Breaking

The Smith-Purcell effect arises when charged particles move near a periodic structure, emitting radiation. Conventional approaches for generating chiral Smith-Purcell radiation rely on metasurface phase engineering or resonant mode interference, typically producing narrow-band, weakly chiral emission. Here, we introduce a resonance-interference-free mechanism that leverages the properties of the charged particles themselves. Using a non-chiral, non-resonant silicon grating, we demonstrate broadband, tunable Smith-Purcell radiation with high chirality, achieving a record-high degree of polarization of 0.87. This is enabled by converting the transverse spin angular momentum of electron-induced evanescent waves into a longitudinal form, producing opposite chirality at different azimuthal angles. Beam twisting or displacement offers precise control over chirality, paving the way for compact chiral light sources, advanced X-ray imaging, and integrated particle diagnostics platforms.

physics.optics↗

arXiv2026

Reviving the Suspension Balance Model

The Suspension Balance Model (SBM) [J. Fluid Mech. \textbf{275}, 157 (1994)] for two-phase flows uses the momentum balance of the particle phase as a closure for the particle flux, showing that particle migration is driven by the divergence of the particle-phase stress. The underlying basis of this model was challenged by Nott~et~al.\ [Phys. Fluids \textbf{23}, 043304 (2011)] where the authors argued that the hydrodynamic contributions to the suspension stress should not appear in the particle-phase momentum balance, being replaced by a different particle-phase stress. The particle-phase stress proposed by Nott~et~al., while mathematically correct, involves the partitioning of the (non-pairwise-additive) hydrodynamic forces, and care is needed to understand how the force on a chosen particle is affected by a second particle. We show by a simple two-particle calculation what is the proper partitioning, and show that it is consistent thermodynamically and gives the correct equilibrium osmotic pressure of Brownian colloids. Using Stokesian Dyanmics suspension rheology, we quantitatively demonstrate that the hydrodynamic contribution to the suspension stress is virtually identical to particle-phase stress; the only difference is that the isolated single-particle hydrodynamic stress contribution -- the Einstein viscosity correction -- must be removed from the suspension stress when used to predict particle flux. Our results validate a key assumption of the SBM and therefore revive its physical foundation.

physics.flu-dyn↗

arXiv2025

A controllable anti-P-pseudo-Hermitian mechanical system and its application

A novel anti-P-pseudo-Hermitian mechanical system that integrates piezoelectric actuators and sensors with non-reciprocal coupling into mechanical beams is proposed. This configuration enables the system to exhibit programmable exceptional points (EPs), which are critical for enhancing sensitivity in sensing applications. Our theoretical analysis, supported by numerical simulations and experimental validation, demonstrates the system's capability to detect minute mass variations and identify surface cracks with high precision. This advancement not only contributes to the field of non-Hermitian physics but also paves the way for the development of next-generation mechanical sensors leveraging EP physics.

physics.app-ph↗

arXiv2024

Quantifying the Critical Micelle Concentration of Nonionic and Ionic Surfactants by Self-Consistent Field Theory

Quantifying the critical micelle concentration (CMC) and understanding its relationship with both the intrinsic molecular structures and environmental conditions are crucial for the rational design of surfactants. Here, we develop a self-consistent field theory which unifies the study of CMC, micellar structure and kinetic pathway of micellization in one framework. The long-range electrostatic interactions are accurately treated, which not only makes the theory applicable to both nonionic and ionic surfactants but also enables us to capture a variety of salt effects. The effectiveness and versatility of the theory is verified by applying it to three types of commonly used surfactants. For polyoxyethylene alkyl ethers (C$_m$E$_n$) surfactants, we predict a wide span of CMC from $10^{-6}$ to $10^{-2}$M as the composition parameters $m$ and $n$ are adjusted. For the ionic sodium dodecyl sulfate (SDS) surfactant, we show the decrease of CMC as salt concentration increases, and capture both the specific cation effect and the specific anion effect. Furthermore, for sodium lauryl ether sulfate (SLES) surfactants, we find a non-monotonic dependence of both the CMC and micelle size on the number of oxyethylene groups. Our theoretical predictions of CMC are in quantitative agreement with experimental data reported in literature for all the three types of surfactants.

cond-mat.soft↗

arXiv2024

Utilizing Graph Generation for Enhanced Domain Adaptive Object Detection

The problem of Domain Adaptive in the field of Object Detection involves the transfer of object detection models from labeled source domains to unannotated target domains. Recent advancements in this field aim to address domain discrepancies by aligning pixel-pairs across domains within a non-Euclidean graphical space, thereby minimizing semantic distribution variance. Despite their remarkable achievements, these methods often use coarse semantic representations to model graphs, mainly due to ignoring non-informative elements and failing to focus on precise semantic alignment. Additionally, the generation of coarse graphs inherently introduces abnormal nodes, posing challenges and potentially biasing domain adaptation outcomes. Consequently, we propose a framework, which utilizes the Graph Generation to enhance the quality of DAOD (\method{}). Specifically, we introduce a Node Refinement module that utilizes a memory bank to reconstruct noisy sampled nodes while applying contrastive regularization to noisy features. To enhance semantic alignment, we propose separating domain-specific styles from category invariance encoded within graph covariances, which allows us to selectively remove domain-specific styles while preserving category-invariant information, thus facilitating more accurate semantic alignment across different domains. Furthermore, we propose a Graph Optimization adaptor, leveraging variational inference to mitigate the impact of abnormal nodes. Extensive experimentation across three adaptation benchmarks validates that \method{} achieves state-of-the-art performance in the task of unsupervised domain adaptation.

cs.CV↗

arXiv2024

Towards Optimal Customized Architecture for Heterogeneous Federated Learning with Contrastive Cloud-Edge Model Decoupling

Federated learning, as a promising distributed learning paradigm, enables collaborative training of a global model across multiple network edge clients without the need for central data collecting. However, the heterogeneity of edge data distribution drags the model towards the local minima, which can be distant from the global optimum. Such heterogeneity often leads to slow convergence and substantial communication overhead. To address these issues, we propose a novel federated learning framework called FedCMD, a model decoupling tailored to the Cloud-edge supported federated learning that separates deep neural networks into a body for capturing shared representations in Cloud and a personalized head for migrating data heterogeneity. Our motivation is that, by the deep investigation of the performance of selecting different neural network layers as the personalized head, we found rigidly assigning the last layer as the personalized head in current studies is not always optimal. Instead, it is necessary to dynamically select the personalized layer that maximizes the training performance by taking the representation difference between neighbor layers into account. To find the optimal personalized layer, we utilize the low-dimensional representation of each layer to contrast feature distribution transfer and introduce a Wasserstein-based layer selection method, aimed at identifying the best-match layer for personalization. Additionally, a weighted global aggregation algorithm is proposed based on the selected personalized layer for the practical application of FedCMD. Extensive experiments on ten benchmarks demonstrate the efficiency and superior performance of our solution compared with nine state-of-the-art solutions. All code and results are available at https://github.com/elegy112138/FedCMD.

cs.LG↗

arXiv2023

Taming Gradient Variance in Federated Learning with Networked Control Variates

Federated learning, a decentralized approach to machine learning, faces significant challenges such as extensive communication overheads, slow convergence, and unstable improvements. These challenges primarily stem from the gradient variance due to heterogeneous client data distributions. To address this, we introduce a novel Networked Control Variates (FedNCV) framework for Federated Learning. We adopt the REINFORCE Leave-One-Out (RLOO) as a fundamental control variate unit in the FedNCV framework, implemented at both client and server levels. At the client level, the RLOO control variate is employed to optimize local gradient updates, mitigating the variance introduced by data samples. Once relayed to the server, the RLOO-based estimator further provides an unbiased and low-variance aggregated gradient, leading to robust global updates. This dual-side application is formalized as a linear combination of composite control variates. We provide a mathematical expression capturing this integration of double control variates within FedNCV and present three theoretical results with corresponding proofs. This unique dual structure equips FedNCV to address data heterogeneity and scalability issues, thus potentially paving the way for large-scale applications. Moreover, we tested FedNCV on six diverse datasets under a Dirichlet distribution with {\alpha} = 0.1, and benchmarked its performance against six SOTA methods, demonstrating its superiority.

cs.LG↗

arXiv2023

Transparent matte surfaces enabled by asymmetric diffusion of white light

The traditional wisdom for achieving transparency is to minimize disordered scattering within and on the surface of materials, so as to avoid translucency. However, the lack of disordered scattering also deprives the possibility of achieving a matte surface, resulting in the specular reflection and glare on transparent materials as a severe light pollution issue. In this work, we propose a solution utilizing optical metasurfaces1-2 to overcome this long-existing dilemma. Our approach leverages an asymmetric background in metasurface design to achieve highly asymmetric diffusion of white light, maximizing diffusion in reflection while minimizing it in transmission across the entire visible spectrum. Using industrial lithography, we have created macroscale transparent matte surfaces with both strong matte appearance and clear transparency, defying the conventional belief that these two optical features are incompatible. These surfaces provide a remarkable phenomenon of switching between transparent or matte appearances via the brightness contrast between the front and rear ambient lights. They also support a unique application in transparent displays and augmented reality, offering perfectly preserved clarity, wide viewing angles, full color, and one-sided displays capabilities. Our findings usher in a new era of optical materials where the desirable properties of both transparent and matte appearances can be seamlessly merged.

physics.optics↗

arXiv2022

Probing the Manipulation of Antiferromagnetic Order in CuMnAs Films Using Neutron Diffraction

We describe measurements of the uniaxial magnetic anisotropy and spin-flop rotation of the N\'eel vector in antiferromagnetic CuMnAs thin films using neutron diffraction. The suppression of the magnetic (100) peak under magnetic fields is observed for films as thin as 20 nm indicating that they undergo a spin-flop transition. Good agreement is found between neutron diffraction and electron transport measurements of the spin-flop rotation in the same layer, with a similar shape and hysteresis of the obtained curves, while the neutron measurements provide a quantitative determination of the spin flop extent throughout the antiferromagnet layer.

cond-mat.mtrl-sci↗

arXiv2021

Protected Long-Distance Guiding of Hypersound Underneath a Nano-Corrugated Surface

Within a new paradigm for communications on the nanoscale, high-frequency surface acoustic waves are becoming effective data carrier and encoder. On-chip communications require acoustic wave propagation along nano-corrugated surfaces which strongly scatter traditional Rayleigh waves. Here we propose the delivery of information using subsurface acoustic waves with hypersound frequencies ~20 GHz, which is a nanoscale analogue of subsurface sound waves in the ocean. A bunch of subsurface hypersound modes is generated by pulsed optical excitation in a multilayer semiconductor structure with a metallic nanograting on top. The guided hypersound modes propagate coherently beneath the nanograting, retaining the surface imprinted information, on a distance of more than 50 {\mu}m which essentially exceeds the propagation length of Rayleigh waves. The concept is suitable for interfacing single photon emitters, such as buried quantum dots, carrying coherent spin excitations in magnonic devices, and encoding the signals for optical communications at the nanoscale.

cond-mat.mes-hall↗

arXiv2020

Ultra-broadband reflectionless Brewster absorber protected by reciprocity

The Brewster's law predicts zero reflection of p-polarization on a dielectric surface at a particular angle. However, when loss is introduced into the permittivity of the dielectric, the Brewster condition breaks down and reflection unavoidably appears. In this work, we found an exception to this long-standing dilemma by creating a class of nonmagnetic anisotropic metamaterials, where an anomalous Brewster effects with independently tunable absorption and refraction emerges. This loss-independent Brewster effect is bestowed by the extra degrees of freedoms introduced by anisotropy and strictly protected by the reciprocity principle. The bandwidth can cover an extremely wide spectrum from dc to optical frequencies. Two examples of reflectionless Brewster absorbers with different Brewster angles are both demonstrated to achieve large absorbance in a wide spectrum via microwave experiments. Our work extends the scope of Brewster effect to the horizon of nonmagnetic absorptive materials, which promises an unprecedented wide bandwidth for reflectionless absorption with high efficiency.

physics.optics↗

arXiv2020

Noise Robust TTS for Low Resource Speakers using Pre-trained Model and Speech Enhancement

With the popularity of deep neural network, speech synthesis task has achieved significant improvements based on the end-to-end encoder-decoder framework in the recent days. More and more applications relying on speech synthesis technology have been widely used in our daily life. Robust speech synthesis model depends on high quality and customized data which needs lots of collecting efforts. It is worth investigating how to take advantage of low-quality and low resource voice data which can be easily obtained from the Internet for usage of synthesizing personalized voice. In this paper, the proposed end-to-end speech synthesis model uses both speaker embedding and noise representation as conditional inputs to model speaker and noise information respectively. Firstly, the speech synthesis model is pre-trained with both multi-speaker clean data and noisy augmented data; then the pre-trained model is adapted on noisy low-resource new speaker data; finally, by setting the clean speech condition, the model can synthesize the new speaker's clean voice. Experimental results show that the speech generated by the proposed approach has better subjective evaluation results than the method directly fine-tuning pre-trained multi-speaker speech synthesis model with denoised new speaker data.

eess.AS↗

arXiv2019

Magnon polaron formed by selectively coupled coherent magnon and phonon modes of a surface patterned ferromagnet

Strong coupling between two quanta of different excitations leads to the formation of a hybridized state which paves a way for exploiting new degrees of freedom to control phenomena with high efficiency and precision. A magnon polaron is the hybridized state of a phonon and a magnon, the elementary quanta of lattice vibrations and spin waves in a magnetically-ordered material. A magnon polaron can be formed at the intersection of the magnon and phonon dispersions, where their frequencies coincide. The observation of magnon polarons in the time domain has remained extremely challenging because the weak interaction of magnons and phonons and their short lifetimes jeopardize the strong coupling required for the formation of a hybridized state. Here, we overcome these limitations by spatial matching of magnons and phonons in a metallic ferromagnet with a nanoscale periodic surface pattern. The spatial overlap of the selected phonon and magnon modes formed in the periodic ferromagnetic structure results in a high coupling strength which, in combination with their long lifetimes allows us to find clear evidence of an optically excited magnon polaron. We show that the symmetries of the localized magnon and phonon states play a crucial role in the magnon polaron formation and its manifestation in the optically excited magnetic transients.

cond-mat.mtrl-sci↗

arXiv2019

Simultaneous Generation of Arbitrary Assembly of Polarization States with Geometrical-Scaling-Induced Phase Modulation

Manipulating the polarization of light on the microscale or nanoscale is essential for integrated photonics and quantum optical devices. Nowadays, the metasurface allows one to build on-chip devices that efficiently manipulate the polarization states. However, it remains challenging to generate different types of polarization states simultaneously, which is required to encode information for quantum computing and quantum cryptography applications. By introducing geometrical-scaling-induced (GSI) phase modulations, we demonstrate that an assembly of circularly polarized (CP) and linearly polarized (LP) states can be simultaneously generated by a single metasurface made of L-shaped resonators with different geometrical features. Upon illumination, each resonator diffracts the CP state with a certain GSI phase. The interaction of these diffractions leads to the desired output beams, where the polarization state and the propagation direction can be accurately tuned by selecting the geometrical shape, size, and spatial sequence of each resonator in the unit cell. As an example of potential applications, we show that an image can be encoded with different polarization profiles at different diffraction orders and decoded with a polarization analyzer. This approach resolves a challenging problem in integrated optics and is inspiring for on-chip quantum information processing.

physics.optics↗

arXiv2019

Optical excitation of propagating magnetostatic waves in an epitaxial Galfenol film by an ultrafast magnetic anisotropy change

Using a time-resolved optically-pumped scanning optical microscopy technique we demonstrate the laser-driven excitation and propagation of spin waves in a 20-nm film of a ferromagnetic metallic alloy Galfenol epitaxially grown on a GaAs substrate. In contrast to previous all-optical studies of spin waves we employ laser-induced thermal changes of magnetocrystalline anisotropy as an excitation mechanism. A tightly focused 70-fs laser pulse excites packets of magnetostatic surface waves with a $e^{-1}$ propagation length of 3.4 $\mu$m, which is comparable with that of permalloy. As a result, laser-driven magnetostatic spin waves are clearly detectable at distances up to 10 $\mu$m, which promotes epitaxial Galfenol films to the limited family of materials suitable for magnonic devices. A pronounced in-plane magnetocrystalline anisotropy of the Galfenol film offers an additional degree of freedom for manipulating the spin waves' parameters. Reorientation of an in-plane external magnetic field relative to the crystallographic axes of the sample tunes the frequency, amplitude and propagation length of the excited waves.

cond-mat.str-el↗

arXiv2018

Liquid Epitaxial Growth of Two-dimensional Non-layer structured hybrid Perovskite

The hybrid perovskite CH3NH3PbX3 (X= Cl, Br, I) is a promising material for developing novel optoelectronic devices. Owing to the intrinsic non-layer structure, it remains challenging to synthesize molecularly thin CH3NH3PbX3 with large size. Here, we report a low-cost and highly efficient fabrication route to obtain large-scale single-crystalline 2D CH3NH3PbX3 perovskites on a mica substrate via liquid epitaxy. The 2D perovskite is characterized as 8 nm in thickness and hundreds of micrometers in lateral size. First-principles calculations suggest the strong potassium-halogen interactions at the perovskite/mica interface lower the interface energy of perovskites, driving their fast in-plane growth. Spectroscopic investigations reveal 2D CH3NH3PbBr3 possess small exciton binding energy of 30 meV, allowing a superior visible-light photodetector with a photoresponsivity of 126 A/W and a bandwidth exceeded 80 kHz. These features demonstrate that liquid epitaxy is a bottom-up approach to fabricate the non-layer structured 2D perovskites, which offer a new material platform for the device applications and fundamental investigations.

cond-mat.mtrl-sci↗

arXiv2018

Breaking Pauli blockade via ultrafast cooling of hot electrons in optically-pumped graphene

Pauli blockade occurs when the excited electrons fill up the states near the conduction bands and block subsequent absorption in semiconductors, and has been widely applied in mode-locking for passively-pulsed-laser systems. In this letter, we report the first direct observation that the Pauli blockade is broken by ultrafast cooling of hot electrons in optically-pumped graphene. With femtosecond spectroscopy, we demonstrate that the time scale to excite an electron (~100 fs) is of the same order as that of the electron decay via electron-electron scattering, which allows the electron excitation interplays strongly with the cooling of hot electrons. Consequently, Pauli blockade is dismissed, leading to an unconventionally enhanced optical absorption. We suggest that this effect is a universal feature of two-dimensional layered materials, which sheds the light of ultrafast carrier dynamics in nonlinear physics and inspires the designing of new-generation of ultrafast optoelectronic devices.

physics.optics↗