arXiv subjects

Xin Wang

Publications and source records attributed to Xin Wang.

At least 793 records · Page 44Linked to original sources

arXiv2024

Phi-4 Technical Report

We present phi-4, a 14-billion parameter language model developed with a training recipe that is centrally focused on data quality. Unlike most language models, where pre-training is based primarily on organic data sources such as web content or code, phi-4 strategically incorporates synthetic data throughout the training process. While previous models in the Phi family largely distill the capabilities of a teacher model (specifically GPT-4), phi-4 substantially surpasses its teacher model on STEM-focused QA capabilities, giving evidence that our data-generation and post-training techniques go beyond distillation. Despite minimal changes to the phi-3 architecture, phi-4 achieves strong performance relative to its size -- especially on reasoning-focused benchmarks -- due to improved data, training curriculum, and innovations in the post-training scheme.

cs.CL↗

arXiv2024

Tunable ultrahigh reflection with broadband via collective atom-atom interaction in waveguide-QED system

We present a scheme for achieving broadband complete reflection by constructing photonic bandgap via collective atom-atom interaction in a one-dimensional (1D) waveguide quantum electrodynamics (QED) system. Moreover, we propose several strategies to further expand the ultrahigh reflection windows, including increasing the number of atoms with separations near the Bragg distance and inducing gradient frequency modulation among the atoms. The center frequency and bandwidth of the ultrahigh reflection window are dynamically adjustable by applying external electromagnetic field. The results here can enrich the many-body physics of waveguide-QED system and offer a pathway for achieving broadened ultrahigh reflection in a controllable way, which can find important applications in the realms of chip-integrated band filter, quantum storage, optical switching, and wavelength-selective devices.

quant-ph↗

arXiv2024

ConDSeg: A General Medical Image Segmentation Framework via Contrast-Driven Feature Enhancement

Medical image segmentation plays an important role in clinical decision making, treatment planning, and disease tracking. However, it still faces two major challenges. On the one hand, there is often a ``soft boundary'' between foreground and background in medical images, with poor illumination and low contrast further reducing the distinguishability of foreground and background within the image. On the other hand, co-occurrence phenomena are widespread in medical images, and learning these features is misleading to the model's judgment. To address these challenges, we propose a general framework called Contrast-Driven Medical Image Segmentation (ConDSeg). First, we develop a contrastive training strategy called Consistency Reinforcement. It is designed to improve the encoder's robustness in various illumination and contrast scenarios, enabling the model to extract high-quality features even in adverse environments. Second, we introduce a Semantic Information Decoupling module, which is able to decouple features from the encoder into foreground, background, and uncertainty regions, gradually acquiring the ability to reduce uncertainty during training. The Contrast-Driven Feature Aggregation module then contrasts the foreground and background features to guide multi-level feature fusion and key feature enhancement, further distinguishing the entities to be segmented. We also propose a Size-Aware Decoder to solve the scale singularity of the decoder. It accurately locate entities of different sizes in the image, thus avoiding erroneous learning of co-occurrence features. Extensive experiments on five medical image datasets across three scenarios demonstrate the state-of-the-art performance of our method, proving its advanced nature and general applicability to various medical image segmentation scenarios. Our released code is available at \url{https://github.com/Mengqi-Lei/ConDSeg}.

eess.IV↗

arXiv2024

Early Results from GLASS-JWST. XXV. Electron Density in the Interstellar Medium at $0.7\lesssim z\lesssim 9.3$ with NIRSpec High-resolution Spectroscopy

The electron density (${n_{\rm e}}$) of the interstellar medium (ISM) in star-forming galaxies is intimately linked to star formation and ionization condition. Using the high-resolution spectra obtained from the JWST NIRSpec micro shutter assembly (MSA) as part of the GLASS-JWST program, we have assembled the largest sample to date (34 galaxies) with individual ${n_{\rm e}}$ measurements derived from the [OII] $λλ$3726,29 and/or [SII] $λλ$6718,32 doublets at $0.7\lesssim z\lesssim 9.3$. The gravitational lensing magnification by the foreground Abell~2744 cluster allows us to probe ${n_{\rm e}}$ in galaxies with stellar masses ($M_{*}$) down to $\simeq 10^{7.5} M_\odot$ across the entire redshift range. Our analysis reveals that the [OII] flux ratios are marginally anti-correlated with specific star formation rate (sSFR) within a 1-$σ$ confidence interval, whereas the [SII] flux ratios show no significant correlation with sSFR. Despite clear correlation between sSFR and redshift within our sample, we find no apparent redshift evolution of ${n_{\rm e}}$ at $z \simeq 1-9$. Our dataset also includes 13 galaxies where ${n_{\rm e}}$ can be measured from both [OII] and [SII]. Contrary to findings at lower redshifts, we observe considerable scatter in ${n_{\rm e}}$ measurements from [OII] and [SII], indicating a complex gaseous environment with significant variations in ${n_{\rm e}}$ in high-redshift galaxies. This work highlights the unique capability of JWST NIRSpec/MSA high-resolution spectroscopy to characterize the detailed physical properties of the ISM in individual high-redshift galaxies.

astro-ph.GA↗

arXiv2024

Not just PAH$_{3.3}$: why galaxies turn red in the Near-Infrared

We measure the spectral properties of a sample of 20 galaxies at z~0.35 selected for having surprisingly red JWST/NIRCAM F200W-F444W colors. 19 galaxies were observed with JWST/NIRSpec in the PRISM configuration, while one galaxy was observed with the high resolution gratings. 17/20 galaxies in our sample exhibit strong 3.3 $μm$ polycyclic aromatic hydrocarbon (PAH$_{3.3}$) emission (equivalent width EW(PAH$_{3.3}$)$>0.03μm$). In these galaxies, the strength of the color excess does not depend on environment and it correlates with EW(PAH$_{3.3}$). Nonetheless, the presence of the PAH$_{3.3}$ alone can not fully explain the color excess, as an equivalent width of ~0.1$μm$ is able to increase the color of galaxies by only 0.13 mag. A contribution from a hot dust component is required to explain the excess. Both the EW(PAH$_{3.3}$) and flux correlate with the H$α$ equivalent width and flux, suggesting that they are produced by the same mechanism. 5/20 galaxies showing PAH would be classified as passive based on broad band rest frame colors ((B-V) and/or UVJ diagrams) and are hence "faux-passive". Of these, 3 galaxies have a significantly lower EW(PAH$_{3.3}$) given their color and also have low EW(H$α$) and we tentatively conclude this behaviour is due to the presence of an AGN. The three galaxies with no PAH$_{3.3}$ in emission have passive spectra, as do the 8 galaxies in our sample with normal F200W-F444W colors. We therefore conclude that the PAH$_{3.3}$ feature is linked to dust-enshrouded star formation. The dust corrected SFR from PAH$_{3.3}$ is a factor of 3.5 higher than the SFR obtained from H$α$, suggesting that these galaxies are characterized by significant amounts of dust.

astro-ph.GA↗

arXiv2024

The Internet of Things in the Era of Generative AI: Vision and Challenges

Advancements in Generative AI hold immense promise to push Internet of Things (IoT) to the next level. In this article, we share our vision on IoT in the era of Generative AI. We discuss some of the most important applications of Generative AI in IoT-related domains. We also identify some of the most critical challenges and discuss current gaps as well as promising opportunities on enabling Generative AI for IoT. We hope this article can inspire new research on IoT in the era of Generative AI.

cs.DC↗

arXiv2024

BEACON: JWST NIRCam Pure-parallel Imaging Survey. I. Survey Design and Initial Results

We introduce the Bias-free Extragalactic Analysis for Cosmic Origins with NIRCam (BEACON) survey, a JWST Cycle2 program allocated up to 600 pure-parallel hours of observations. BEACON explores high-latitude areas of the sky with JWST/NIRCam over $\sim100$ independent sightlines, totaling $\sim0.3$deg$^2$, reaching a median F444W depth of $\approx28.2$AB mag (5$σ$). Based on existing JWST observations in legacy fields, we estimate that BEACON will photometrically identify 25--150 galaxies at $z>10$ and 500--1000 at $z\sim7$--10 uniquely enabled by an efficient multiple filter configuration spanning $0.9$--5.0$μ$m. The expected sample size of $z>10$ galaxies will allow us to obtain robust number density estimates and to discriminate between different models of early star formation. In this paper, we present an overview of the survey design and initial results using the first 19 fields. We present 129 galaxy candidates at $z>7$ identified in those fields, including 11 galaxies at $z>10$ and several UV-luminous ($M_{\rm UV}<-21$mag) galaxies at $z\sim8$. The number densities of $z<13$ galaxies inferred from the initial fields are overall consistent with those in the literature. Despite reaching a considerably large volume ($\sim10^5$Mpc$^3$), however, we find no galaxy candidates at $z>13$, providing us with a complimentary insight into early galaxy evolution with minimal cosmic variance. We publish imaging and catalog data products for these initial fields. Upon survey completion, all BEACON data will be coherently processed and distributed to the community along with catalogs for redshift and other physical quantities.

astro-ph.GA↗

arXiv2024

Two-color Ytterbium MOT in a compact dual-chamber setup

We present an experimental scheme for producing ultracold Ytterbium atoms in a compact dual-chamber setup. A dispenser-loaded two-dimensional (2D) magneto-optical trap (MOT) using permanent magnets and operating on the broad $^1S_0\to {}^1P_1$ singlet transition delivers over $10^7$ atoms per second through a differential pumping stage into a three-dimensional (3D) MOT. The two-color 3D MOT uses the broad singlet transition to accumulate $\sim\!2\times 10^7$ atoms of $^{174}\text{Yb}$ within $2.5~\text{s}$ and subsequently the narrow $^1S_0\to {}^3P_1$ intercombination line to cool the atomic cloud to below $10~\mathrm{μK}$. We report optimized parameters for each stage of the atom collection sequence, achieving high transfer efficiency. We find that shelving into the triplet state during the broad-transition MOT almost doubles the number of trapped atoms.

physics.atom-ph↗

arXiv2024

A Self-Learning Multimodal Approach for Fake News Detection

The rapid growth of social media has resulted in an explosion of online news content, leading to a significant increase in the spread of misleading or false information. While machine learning techniques have been widely applied to detect fake news, the scarcity of labeled datasets remains a critical challenge. Misinformation frequently appears as paired text and images, where a news article or headline is accompanied by a related visuals. In this paper, we introduce a self-learning multimodal model for fake news classification. The model leverages contrastive learning, a robust method for feature extraction that operates without requiring labeled data, and integrates the strengths of Large Language Models (LLMs) to jointly analyze both text and image features. LLMs are excel at this task due to their ability to process diverse linguistic data drawn from extensive training corpora. Our experimental results on a public dataset demonstrate that the proposed model outperforms several state-of-the-art classification approaches, achieving over 85% accuracy, precision, recall, and F1-score. These findings highlight the model's effectiveness in tackling the challenges of multimodal fake news detection.

cs.CL↗

arXiv2024

Long-Range Four-body Interactions in Structured Nonlinear Photonic Waveguides

Multi-photon dynamics beyond linear optical materials are of significant fundamental and technological importance in quantum information processing. However, it remains largely unexplored in nonlinear waveguide QED. In this work, we theoretically propose a structured nonlinear waveguide in the presence of staggered photon-photon interactions, which supports two branches of gaped bands for doublons (i.e., spatially bound-photon-pair states). In contrast to linear waveguide QED systems, we identify two important contributions to its dynamical evolution, i.e., single-photon bound states (SPBSs) and doublon bound states (DBSs). Most remarkably, the nonlinear waveguide can mediate the long-range four-body interactions between two emitter pairs, even in the presence of disturbance from SPBS. By appropriately designing system's parameters, we can achieve high-fidelity four-body Rabi oscillations mediated only by virtual doublons in DBSs. Our findings pave the way for applying structured nonlinear waveguide QED in multi-body quantum information processing and quantum simulations among remote sites.

quant-ph↗

arXiv2024

Scaling Laws for Post Training Quantized Large Language Models

Generalization abilities of well-trained large language models (LLMs) are known to scale predictably as a function of model size. In contrast to the existence of practical scaling laws governing pre-training, the quality of LLMs after post-training compression remains highly unpredictable, often requiring case-by-case validation in practice. In this work, we attempted to close this gap for post-training weight quantization of LLMs by conducting a systematic empirical study on multiple LLM families quantized to numerous low-precision tensor data types using popular weight quantization techniques. We identified key scaling factors pertaining to characteristics of the local loss landscape, based on which the performance of quantized LLMs can be reasonably well predicted by a statistical model.

cs.LG↗

arXiv2024

BOSS: Blocking algorithm for optimizing shuttling scheduling in Ion Trap

Ion traps stand at the forefront of quantum hardware technology, presenting unparalleled benefits for quantum computing, such as high-fidelity gates, extensive connectivity, and prolonged coherence times. In this context, we explore the critical role of shuttling operations within these systems, especially their influence on the fidelity loss and elongated execution times. To address these challenges, we have developed BOSS, an efficient blocking algorithm tailored to enhance shuttling efficiency. This optimization not only bolsters the shuttling process but also elevates the overall efficacy of ion trap devices. We experimented on multiple applications using two qubit gates up to 4000+ and qubits ranging from 64 to 78. Our method significantly reduces the number of shuttles on most applications, with a maximum reduction of 96.1%. Additionally, our investigation includes simulations of realistic experimental parameters that incorporate sympathetic cooling, offering a higher fidelity and a refined estimate of execution times that align more closely with practical scenarios.

quant-ph↗

arXiv2024

Tortho-Gaussian: Splatting True Digital Orthophoto Maps

True Digital Orthophoto Maps (TDOMs) are essential products for digital twins and Geographic Information Systems (GIS). Traditionally, TDOM generation involves a complex set of traditional photogrammetric process, which may deteriorate due to various challenges, including inaccurate Digital Surface Model (DSM), degenerated occlusion detections, and visual artifacts in weak texture regions and reflective surfaces, etc. To address these challenges, we introduce TOrtho-Gaussian, a novel method inspired by 3D Gaussian Splatting (3DGS) that generates TDOMs through orthogonal splatting of optimized anisotropic Gaussian kernel. More specifically, we first simplify the orthophoto generation by orthographically splatting the Gaussian kernels onto 2D image planes, formulating a geometrically elegant solution that avoids the need for explicit DSM and occlusion detection. Second, to produce TDOM of large-scale area, a divide-and-conquer strategy is adopted to optimize memory usage and time efficiency of training and rendering for 3DGS. Lastly, we design a fully anisotropic Gaussian kernel that adapts to the varying characteristics of different regions, particularly improving the rendering quality of reflective surfaces and slender structures. Extensive experimental evaluations demonstrate that our method outperforms existing commercial software in several aspects, including the accuracy of building boundaries, the visual quality of low-texture regions and building facades. These results underscore the potential of our approach for large-scale urban scene reconstruction, offering a robust alternative for enhancing TDOM quality and scalability.

cs.CV↗

arXiv2024

Modelling to Generate Continuous Alternatives: Enabling Real-Time Feasible Portfolio Generation in Convex Planning Models

Decarbonization provides new opportunities to plan energy systems for improved health, resilience, equity, and environmental outcomes, but challenges in siting and social acceptance of transition goals and targets threaten progress. Modelling to Generate Alternatives (MGA) provides an optimization method for capturing many near-cost-optimal system configurations, and can provide insights into the tradeoffs between objectives and flexibility available in the system. However, MGA is currently limited in interactive applicability to these problems due to a lack of methods for allowing users to explore near-optimal feasible spaces. In this work we describe Modelling to Generate Continuous Alternatives (MGCA), a novel post-processing algorithm for convex planning problems which enables users to rapidly generate new interior solutions, incorporate new constraints, and solve within the space with convex objectives. MGCA begins with a dimensionality reduction to capacity decisions and metric values. We then take advantage of convex combinations to generate interior points by allowing user weight specification and encoding convex combinations in an optimization problem with user-defined additional constraints and objective. Dimensionality reduction enables this problem to solve in tenths of a second, suitable for analysis in interactive settings. We discuss the interpolation of capacity and operational metric values, finding capacity metrics can be perfectly interpolated while operational metrics remain within the feasible range of the points used to create them. We demonstrate interpolated solutions can be exported and re-solved with an economic dispatch model to provide operational metric values consistent with least-cost decision-making and show interpolated metric values are generally within 10% of the optimal value.

math.OC↗

arXiv2024

Energy-efficient Federated Learning with Dynamic Model Size Allocation

Federated Learning (FL) presents a paradigm shift towards distributed model training across isolated data repositories or edge devices without explicit data sharing. Despite of its advantages, FL is inherently less efficient than centralized training models, leading to increased energy consumption and, consequently, higher carbon emissions. In this paper, we propose CAMA, a carbon-aware FL framework, promoting the operation on renewable excess energy and spare computing capacity, aiming to minimize operational carbon emissions. CAMA introduces a dynamic model adaptation strategy which adapts the model sizes based on the availability of energy and computing resources. Ordered dropout is integratged to enable the aggregation with varying model sizes. Empirical evaluations on real-world energy and load traces demonstrate that our method achieves faster convergence and ensures equitable client participation, while scaling efficiently to handle large numbers of clients. The source code of CAMA is available at https://github.com/denoslab/CAMA.

cs.DC↗

arXiv2024

The Master-Slave Encoder Model for Improving Patent Text Summarization: A New Approach to Combining Specifications and Claims

In order to solve the problem of insufficient generation quality caused by traditional patent text abstract generation models only originating from patent specifications, the problem of new terminology OOV caused by rapid patent updates, and the problem of information redundancy caused by insufficient consideration of the high professionalism, accuracy, and uniqueness of patent texts, we proposes a patent text abstract generation model (MSEA) based on a master-slave encoder architecture; Firstly, the MSEA model designs a master-slave encoder, which combines the instructions in the patent text with the claims as input, and fully explores the characteristics and details between the two through the master-slave encoder; Then, the model enhances the consideration of new technical terms in the input sequence based on the pointer network, and further enhances the correlation with the input text by re weighing the "remembered" and "for-gotten" parts of the input sequence from the encoder; Finally, an enhanced repetition suppression mechanism for patent text was introduced to ensure accurate and non redundant abstracts generated. On a publicly available patent text dataset, compared to the state-of-the-art model, Improved Multi-Head Attention Mechanism (IMHAM), the MSEA model achieves an improvement of 0.006, 0.005, and 0.005 in Rouge-1, Rouge-2, and Rouge-L scores, respectively. MSEA leverages the characteristics of patent texts to effectively enhance the quality of patent text generation, demonstrating its advancement and effectiveness in the experiments.

cs.CL↗

arXiv2024

Hodge integrals and $λ_{g}$ conjecture with target varieties

In this paper, we propose $λ_{g}$ conjecture for Hodge integrals with target varieties. Then we establish relations between Virasoro conjecture and $λ_{g}$ conjecture, in particular, we prove $λ_{g}$ conjecture in all genus for smooth projective varieties with semisimple quantum cohomology or smooth algebraic curves. Meanwhile, we also prove $λ_{g}$ conjecture in genus zero for any smooth projective varieties. In the end, together with DR formula for $λ_g$ class, we obtain a new type of universal constraints for descendant Gromov-Witten invariants. As an application, we prove $λ_{g}$ conjecture in genus one for any smooth projective varieties.

math.AG↗

arXiv2024

TAPT: Test-Time Adversarial Prompt Tuning for Robust Inference in Vision-Language Models

Large pre-trained Vision-Language Models (VLMs) such as CLIP have demonstrated excellent zero-shot generalizability across various downstream tasks. However, recent studies have shown that the inference performance of CLIP can be greatly degraded by small adversarial perturbations, especially its visual modality, posing significant safety threats. To mitigate this vulnerability, in this paper, we propose a novel defense method called Test-Time Adversarial Prompt Tuning (TAPT) to enhance the inference robustness of CLIP against visual adversarial attacks. TAPT is a test-time defense method that learns defensive bimodal (textual and visual) prompts to robustify the inference process of CLIP. Specifically, it is an unsupervised method that optimizes the defensive prompts for each test sample by minimizing a multi-view entropy and aligning adversarial-clean distributions. We evaluate the effectiveness of TAPT on 11 benchmark datasets, including ImageNet and 10 other zero-shot datasets, demonstrating that it enhances the zero-shot adversarial robustness of the original CLIP by at least 48.9% against AutoAttack (AA), while largely maintaining performance on clean examples. Moreover, TAPT outperforms existing adversarial prompt tuning methods across various backbones, achieving an average robustness improvement of at least 36.6%.

cs.CV↗