arXiv subjects

Zhaoning Yu

Publications and source records attributed to Zhaoning Yu.

At least 19 recordsLinked to original sources

arXiv2026

Silicon-on-sapphire metasurfaces generate arrays of dark and bright traps for neutral atoms

We demonstrated crystalline silicon-on-sapphire (c-SOS) metasurfaces that convert a Gaussian beam into arrays of complex optical traps, including arrays of optical bottle beams that trap atoms in dark regions interleaved with bright tweezer arrays. The high refractive index and indirect band gap of crystalline silicon makes it possible to design high-resolution near-infrared ($\lambda>700$ nm) metasurfaces that can be manufactured at scale using CMOS-compatible processes. Compared with active components like spatial light modulators (SLMs) that have become widely used to generate trap arrays, metasurfaces provide an indefinitely scalable number of pixels, enabling large arrays of complex traps in a very small form factor, as well as reduced dynamic noise. To design metasurfaces that can generate three-dimensional bottle beams to serve as dark traps, we modified the Gerchberg-Saxton algorithm to enforce complex-amplitude profiles at the focal plane of the metasurface and to optimize the uniformity of the traps across the array. We fabricated and measured c-SOS metasurfaces that convert a Gaussian laser beam into arrays of bright traps, dark traps, and interleaved bright/dark traps.

physics.optics↗

arXiv2025

RESTRAIN: From Spurious Votes to Signals -- Self-Driven RL with Self-Penalization

Reinforcement learning with human-annotated data has boosted chain-of-thought reasoning in large reasoning models, but these gains come at high costs in labeled data while faltering on harder tasks. A natural next step is experience-driven learning, where models improve without curated labels by adapting to unlabeled data. We introduce RESTRAIN (REinforcement learning with Self-restraint), a self-penalizing RL framework that converts the absence of gold labels into a useful learning signal. Instead of overcommitting to spurious majority votes, RESTRAIN exploits signals from the model's entire answer distribution: penalizing overconfident rollouts and low-consistency examples while preserving promising reasoning chains. The self-penalization mechanism integrates seamlessly into policy optimization methods such as GRPO, enabling continual self-improvement without supervision. On challenging reasoning benchmarks, RESTRAIN delivers large gains using only unlabeled data. With Qwen3-4B-Base and OctoThinker Hybrid-8B-Base, it improves Pass@1 by up to +140.7 percent on AIME25, +36.2 percent on MMLU_STEM, and +19.6 percent on GPQA-Diamond, nearly matching gold-label training while using no gold labels. These results demonstrate that RESTRAIN establishes a scalable path toward stronger reasoning without gold labels.

cs.CL↗

arXiv2024

G2T-LLM: Graph-to-Tree Text Encoding for Molecule Generation with Fine-Tuned Large Language Models

We introduce G2T-LLM, a novel approach for molecule generation that uses graph-to-tree text encoding to transform graph-based molecular structures into a hierarchical text format optimized for large language models (LLMs). This encoding converts complex molecular graphs into tree-structured formats, such as JSON and XML, which LLMs are particularly adept at processing due to their extensive pre-training on these types of data. By leveraging the flexibility of LLMs, our approach allows for intuitive interaction using natural language prompts, providing a more accessible interface for molecular design. Through supervised fine-tuning, G2T-LLM generates valid and coherent chemical structures, addressing common challenges like invalid outputs seen in traditional graph-based methods. While LLMs are computationally intensive, they offer superior generalization and adaptability, enabling the generation of diverse molecular structures with minimal task-specific customization. The proposed approach achieved comparable performances with state-of-the-art methods on various benchmark molecular generation datasets, demonstrating its potential as a flexible and innovative tool for AI-driven molecular design.

cs.LG↗

arXiv2024

Broadband light extraction from near-surface NV centers using crystalline-silicon antennas

We use crystalline silicon (Si) antennas to efficiently extract broadband single-photon fluorescence from shallow nitrogen-vacancy (NV) centers in diamond into free space. Our design features relatively easy-to-pattern high-index Si resonators on the diamond surface to boost photon extraction by overcoming total internal reflection and Fresnel reflection at the diamond-air interface, and providing modest Purcell enhancement, without etching or otherwise damaging the diamond surface. In simulations, ~17 times more single photons are collected from a single NV center compared to the case without the antenna; in experiments, we observe an enhancement of ~9 times, limited by spatial alignment between the NV and the antenna. Our approach can be readily applied to other color centers in diamond, and more generally to the extraction of light from quantum emitters in wide-bandgap materials.

physics.optics↗

arXiv2024

MAGE: Model-Level Graph Neural Networks Explanations via Motif-based Graph Generation

Graph Neural Networks (GNNs) have shown remarkable success in molecular tasks, yet their interpretability remains challenging. Traditional model-level explanation methods like XGNN and GNNInterpreter often fail to identify valid substructures like rings, leading to questionable interpretability. This limitation stems from XGNN's atom-by-atom approach and GNNInterpreter's reliance on average graph embeddings, which overlook the essential structural elements crucial for molecules. To address these gaps, we introduce an innovative \textbf{M}otif-b\textbf{A}sed \textbf{G}NN \textbf{E}xplainer (MAGE) that uses motifs as fundamental units for generating explanations. Our approach begins with extracting potential motifs through a motif decomposition technique. Then, we utilize an attention-based learning method to identify class-specific motifs. Finally, we employ a motif-based graph generator for each class to create molecular graph explanations based on these class-specific motifs. This novel method not only incorporates critical substructures into the explanations but also guarantees their validity, yielding results that are human-understandable. Our proposed method's effectiveness is demonstrated through quantitative and qualitative assessments conducted on six real-world molecular datasets.

cs.LG↗

arXiv2023

MotifPiece: A Data-Driven Approach for Effective Motif Extraction and Molecular Representation Learning

Motif extraction is an important task in motif based molecular representation learning. Previously, machine learning approaches employing either rule-based or string-based techniques to extract motifs. Rule-based approaches may extract motifs that aren't frequent or prevalent within the molecular data, which can lead to an incomplete understanding of essential structural patterns in molecules. String-based methods often lose the topological information inherent in molecules. This can be a significant drawback because topology plays a vital role in defining the spatial arrangement and connectivity of atoms within a molecule, which can be critical for understanding its properties and behavior. In this paper, we develop a data-driven motif extraction technique known as MotifPiece, which employs statistical measures to define motifs. To comprehensively evaluate the effectiveness of MotifPiece, we introduce a heterogeneous learning module. Our model shows an improvement compared to previously reported models. Additionally, we demonstrate that its performance can be further enhanced in two ways: first, by incorporating more data to aid in generating a richer motif vocabulary, and second, by merging multiple datasets that share enough motifs, allowing for cross-dataset learning.

q-bio.QM↗

arXiv2022

MotifExplainer: a Motif-based Graph Neural Network Explainer

We consider the explanation problem of Graph Neural Networks (GNNs). Most existing GNN explanation methods identify the most important edges or nodes but fail to consider substructures, which are more important for graph data. The only method that considers subgraphs tries to search all possible subgraphs and identify the most significant subgraphs. However, the subgraphs identified may not be recurrent or statistically important. In this work, we propose a novel method, known as MotifExplainer, to explain GNNs by identifying important motifs, recurrent and statistically significant patterns in graphs. Our proposed motif-based methods can provide better human-understandable explanations than methods based on nodes, edges, and regular subgraphs. Given an input graph and a pre-trained GNN model, our method first extracts motifs in the graph using well-designed motif extraction rules. Then we generate motif embedding by feeding motifs into the pre-trained GNN. Finally, we employ an attention-based method to identify the most influential motifs as explanations for the final prediction results. The empirical studies on both synthetic and real-world datasets demonstrate the effectiveness of our method.

cs.LG↗

arXiv2022

Molecular Representation Learning via Heterogeneous Motif Graph Neural Networks

We consider feature representation learning problem of molecular graphs. Graph Neural Networks have been widely used in feature representation learning of molecular graphs. However, most existing methods deal with molecular graphs individually while neglecting their connections, such as motif-level relationships. We propose a novel molecular graph representation learning method by constructing a heterogeneous motif graph to address this issue. In particular, we build a heterogeneous motif graph that contains motif nodes and molecular nodes. Each motif node corresponds to a motif extracted from molecules. Then, we propose a Heterogeneous Motif Graph Neural Network (HM-GNN) to learn feature representations for each node in the heterogeneous motif graph. Our heterogeneous motif graph also enables effective multi-task learning, especially for small molecular datasets. To address the potential efficiency issue, we propose to use an edge sampler, which can significantly reduce computational resources usage. The experimental results show that our model consistently outperforms previous state-of-the-art models. Under multi-task settings, the promising performances of our methods on combined datasets shed light on a new learning paradigm for small molecular datasets. Finally, we show that our model achieves similar performances with significantly less computational resources by using our edge sampler.

cs.LG↗

arXiv2021

Efficient generation of optical bottle beams

Optical bottle beams can be used to trap atoms and small low-index particles. We introduce a figure of merit for optical bottle beams, specifically in the context of optical traps, and use it to compare optical bottle-beam traps obtained by three different methods. Using this figure of merit and an optimization algorithm, we identified optical bottle-beam traps based on a Gaussian beam illuminating a metasurface that are superior in terms of power efficiency than existing approaches. We numerically demonstrate a silicon metasurface for creating an optical bottle-beam trap.

physics.optics↗

arXiv2021

Passive frequency conversion of ultraviolet images into the visible using perovskite nanocrystals

We demonstrate a passive down-conversion imaging system that converts broadband ultraviolet light to narrow-band green light while preserving the directionality of rays, and thus enabling direct down-conversion imaging. At the same time our system has high transparency in the visible, enabling superimposed visible and ultraviolet imaging. The frequency conversion is performed by a subwavelength-thickness transparent downconverter based on highly efficient CsPbBr3 nanocrystals incorporated into the focal plane of a simple telescope or relay-lens geometry. The resulting imaging performance of this down-conversion system approaches the diffraction limit. This demonstration sets the stage for the incorporation of other high-efficiency perovskite nanocrystal materials to enable passive multi-frequency conversion imaging systems.

physics.optics↗

arXiv2020

Hyperspectral interference tomography of nacre

Structural characterization of biologically formed materials is essential for understanding biological phenomena and their environment, and generating new bio-inspired engineering concepts. For example, nacre -- formed by mollusks in the ocean -- encodes local environmental conditions throughout its formation and has exceptional strength due to its nanoscale brick-and-mortar structure. This layered structure, comprising transparent aragonite tablets bonded with an ultra-thin organic polymer, also results in stunning interference colors. Existing methods of structural characterization of nacre rely on some form of cross-sectional analysis, such as scanning electron microscopy or polarization-dependent imaging contrast (PIC) mapping. However, these techniques are destructive and too time- and resource-intensive to analyze large sample areas. Here we present an all-optical, rapid, and non-destructive imaging technique -- hyperspectral interference tomography (HIT) -- to spatially map the structural parameters of nacre and other disordered layered materials. We combined hyperspectral imaging with optical-interference modeling to infer the mean tablet thickness and disordering of nacre layers across entire mollusk shells at various stages of development, observing a previously unknown relationship between the growth of the mollusk and tablet thickness. Our rapid, inexpensive, and nondestructive method can be readily applied to in-field studies.

physics.optics↗

arXiv2020

Adjoint-optimized nanoscale light extractor for nitrogen-vacancy centers in diamond

We designed a nanoscale light extractor (NLE) for efficient outcoupling and beaming of broadband light emitted by shallow, negatively charged nitrogen-vacancy (NV) centers in bulk diamond. The NLE consists of a patterned silicon layer on diamond and requires no etching of the diamond surface. Our design process is based on adjoint optimization using broadband time-domain simulations and yields structures that are inherently robust to positioning and fabrication errors. Our NLE functions like a transmission antenna for the NV center, enhancing the optical power extracted from an NV center positioned 10 nm below the diamond surface by a factor of more than 35, and beaming the light into a +/-30{\deg} cone in the far field. This approach to light extraction can be readily adapted to other solid-state color centers.

physics.optics↗

arXiv2020

Ultrathin broadband reflective optical limiter

Optical limiters are nonlinear devices that feature decreasing transmittance with increasing incident optical intensity, and thus can protect sensitive components from high-intensity illumination. The ideal optical limiter reflects rather than absorbs light in its active ("limiting") state, minimizing risk of damage to the limiter itself. Previous efforts to realize reflective limiters were based on embedding nonlinear layers into relatively thick multilayer photonic structures, resulting in substantial fabrication complexity, reduced speed and, in some instances, limited working bandwidth. We overcome these tradeoffs by using the insulator-to-metal transition in vanadium dioxide (VO2) to achieve intensity-dependent modulation of resonant transmission through aperture antennas. Due to the dramatic change of optical properties across the insulator-to-metal transition, low-quality-factor resonators were sufficient to achieve high on-off ratios in device transmittance. As a result, our ultra-thin reflective limiter (thickness ~1/100 of the free-space wavelength) is broadband in terms of operating wavelength (> 2 um at 10 um) and angle of incidence (up to ~50$\deg$ away from the normal).

physics.optics↗

arXiv2019

Guidelines for measurement of thermal emission

Thermal emission is the radiation of electromagnetic waves from hot objects. The promise of thermal-emission engineering for applications in energy harvesting, radiative cooling, and thermal camouflage has recently led to renewed research interest in this topic. There is a substantial need for accurate and precise measurement of thermal emission in a laboratory setting, which can be challenging in part due to the presence of background emission from the surrounding environment and the measurement instrument itself. This is especially true for measurements of emitters at temperatures close to that of the environment, where the impact of background emission is relatively large. In this paper, we describe, recommend, and demonstrate general procedures for thermal-emission measurements that are applicable to most experimental conditions, including less-common and more-challenging cases that include thermal emitters with temperature-dependent emissivity and emitters that are not in thermal equilibrium.

physics.optics↗

arXiv2019

Wide-angle spectrally selective absorbers and thermal emitters based on inverse opals

Engineered optical absorbers are of substantial interest for applications ranging from stray light reduction to energy conversion. We demonstrate a large-area (centimeter-scale) metamaterial that features near-unity frequency-selective absorption in the mid-infrared wavelength range. The metamaterial comprises a self-assembled porous structure known as an inverse opal, here made of silica. The structure's large volume fraction of voids, together with the vibrational resonances of silica in the mid-infrared spectral range, reduce the metamaterial's refractive index to close to that of air and introduce considerable optical absorption. As a result, the frequency-selective structure efficiently absorbs incident light of both polarizations even at very oblique incidence angles. The absorber remains stable at high temperatures (measured up to ~900 degrees C), enabling its operation as a frequency-selective thermal emitter. The excellent performance of this absorber/emitter and ease of fabrication make it a promising surface coating for passive radiative cooling, laser safety, and other large-area applications.

physics.optics↗

arXiv2019

Temperature-independent thermal radiation

Thermal emission is the process by which all objects at non-zero temperatures emit light, and is well-described by the classic Planck, Kirchhoff, and Stefan-Boltzmann laws. For most solids, the thermally emitted power increases monotonically with temperature in a one-to-one relationship that enables applications such as infrared imaging and non-contact thermometry. Here, we demonstrate ultrathin thermal emitters that violate this one-to-one relationship via the use of samarium nickel oxide (SmNiO3), a strongly correlated quantum material that undergoes a fully reversible, temperature-driven solid-state phase transition. The smooth and hysteresis-free nature of this unique insulator-to-metal (IMT) phase transition allows us to engineer the temperature dependence of emissivity to precisely cancel out the intrinsic blackbody profile described by the Stefan-Boltzmann law, for both heating and cooling. Our design results in temperature-independent thermally emitted power within the long-wave atmospheric transparency window (wavelengths of 8 - 14 um), across a broad temperature range of ~30 {\deg}C, centered around ~120 {\deg}C. The ability to decouple temperature and thermal emission opens a new gateway for controlling the visibility of objects to infrared cameras and, more broadly, new opportunities for quantum materials in controlling heat transfer.

physics.optics↗

arXiv2018

Multi-refractive-index metamaterials based on subwavelength waveguide arrays

We demonstrate a metamaterial that cannot be described by a single set of refractive-index and impedance values, even for fixed frequency and polarization. The metamaterial structure is a stack of dissimilar waveguides with subwavelength width and spacing, which guide light at different phase velocities. This multi-refractive-index metamaterial (MRIM) can be viewed as a spatial superposition of multiple homogeneous materials. Using full-wave simulations, we demonstrate several optical components based on MRIMs, including prisms that deflect light to multiple angles, lenses with multiple focal points, and multi-index Fabry-Perot etalons with an enhanced density of resonant modes. We provide a theoretical framework for analyzing MRIMs to determine effective refractive indices, fractions of power to each channel (i.e., to each refractive index), and transmittance.

physics.optics↗

arXiv2018

Thermal-emission measurements near room temperature using Fourier-transform infrared spectroscopy

Accurate characterization of thermal emitters can be challenging due to the presence of background thermal emission from components of the experimental setup and the surrounding environment. This is especially true for an emitter operating close to room temperature. Here, we explore characterization of near-room-temperature thermal emitters using Fourier-transform infrared (FTIR) spectroscopy. We find that the thermal background arising from optical components placed between the beam splitter and the detector in an FTIR spectrometer appears as a "negative" contribution to the Fourier-transformed signal, leading to errors in thermal-emission measurements near room temperature. Awareness of this contribution will help properly calibrate low-temperature thermal-emission measurements.

physics.ins-det↗