arXiv subjects

Ying Hu

Publications and source records attributed to Ying Hu.

At least 55 records · Page 3Linked to original sources

arXiv2025

BitNet b1.58 2B4T Technical Report

We introduce BitNet b1.58 2B4T, the first open-source, native 1-bit Large Language Model (LLM) at the 2-billion parameter scale. Trained on a corpus of 4 trillion tokens, the model has been rigorously evaluated across benchmarks covering language understanding, mathematical reasoning, coding proficiency, and conversational ability. Our results demonstrate that BitNet b1.58 2B4T achieves performance on par with leading open-weight, full-precision LLMs of similar size, while offering significant advantages in computational efficiency, including substantially reduced memory footprint, energy consumption, and decoding latency. To facilitate further research and adoption, the model weights are released via Hugging Face along with open-source inference implementations for both GPU and CPU architectures.

cs.CL↗

arXiv2025

Time delay of mean field interaction in thermal Rydberg atomic gas

Mean field theory is commonly employed to study nonequilibrium dynamics in hot Rydberg atomic ensembles, but the fundamental mechanism behind the generation of the mean-field interactions remains poorly understood. In this work, we experimentally observe a time-delay effect in the buildup of mean-field interaction, which reveals the key role of collision ionization. We analyze the relevant collision channels and propose a microscopic mechanism that quantitatively explains the hysteresis window observed in optical bistability. Then, using square-wave modulation spectroscopy (SMS) to monitor the growth of the mean-field interaction, we experimentally demonstrate a delay in its dynamical buildup following the initial Rydberg excitation. Finally, we demonstrate how this delay effect may help understand the recently observed self-sustained oscillations in thermal Rydberg gases. Our findings provide compelling evidence for the contribution of ionization processes in the nonequilibrium dynamics of thermal Rydberg gas, a system of growing interest for quantum sensing and quantum information science.

physics.atom-ph↗

arXiv2025

Realizing exceptional points by Floquet dissipative couplings in thermal atoms

Exceptional degeneracies and generically complex spectra of non-Hermitian systems are at the heart of numerous phenomena absent in the Hermitian realm. Recently, it was suggested that Floquet dissipative coupling in the space-time domain may provide a novel mechanism to drive intriguing spectral topology with no static analogues, though its experimental investigation in quantum systems remains elusive. We demonstrate such Floquet dissipative coupling in an ensemble of thermal atoms interacting with two spatially separated optical beams, and observe an anomalous anti-parity-time symmetry phase transition at an exception point far from the phase-transition threshold of the static counterpart. Our protocol sets the stage for Floquet engineering of non-Hermitian topological spectra, and for engineering new quantum phases that cannot exist in static systems.

quant-ph↗

arXiv2025

GDiffRetro: Retrosynthesis Prediction with Dual Graph Enhanced Molecular Representation and Diffusion Generation

Retrosynthesis prediction focuses on identifying reactants capable of synthesizing a target product. Typically, the retrosynthesis prediction involves two phases: Reaction Center Identification and Reactant Generation. However, we argue that most existing methods suffer from two limitations in the two phases: (i) Existing models do not adequately capture the ``face'' information in molecular graphs for the reaction center identification. (ii) Current approaches for the reactant generation predominantly use sequence generation in a 2D space, which lacks versatility in generating reasonable distributions for completed reactive groups and overlooks molecules' inherent 3D properties. To overcome the above limitations, we propose GDiffRetro. For the reaction center identification, GDiffRetro uniquely integrates the original graph with its corresponding dual graph to represent molecular structures, which helps guide the model to focus more on the faces in the graph. For the reactant generation, GDiffRetro employs a conditional diffusion model in 3D to further transform the obtained synthon into a complete reactant. Our experimental findings reveal that GDiffRetro outperforms state-of-the-art semi-template models across various evaluative metrics.

cs.AI↗

arXiv2024

The Hybrid ROA: A Flexible and Scalable Encoding Scheme for Route Origin Authorization

On top of the Resource Public Key Infrastructure (RPKI), the Route Origin Authorization (ROA) creates a cryptographically verifiable binding of an autonomous system to a set of IP prefixes it is authorized to originate. By their design, ROAs can protect the inter-domain routing system against prefix and sub-prefix hijacks. However, it is hard for the state-of-the-art approach, the maxLength-based ROA encoding scheme, to guarantee security and scalability at the same time when facing various authorization scenarios. To this end, we propose a novel bitmap-based encoding scheme for ROAs to provide flexible and controllable compression. Furthermore, the hybrid ROA encoding scheme (h-ROA) is proposed, which encodes ROAs based on maxLength and bitmap jointly. This approach ensures strong security, provides flexibility and significantly improves system scalability, enabling it to effectively handle various authorization patterns. According to the performance evaluation with real-world data sets, h-ROA outperforms the state-of-the-art approach $1.99 \sim 3.28$ times in terms of the encoding speed, and it can reduce the cost of a router to synchronize all validated ROA payloads by $43.9\% \sim 56.6\%$.

cs.CR↗

arXiv2024

Transport of Vector Solitons in Spin-Dependent Nonlinear Thouless Pumps

In nonlinear topological physics, Thouless pumping of nonlinear excitations is a central topic, often illustrated by scalar solitons. Vector solitons, with the additional spin degree of freedom, exhibit phenomena absent in scalar solitons due to enriched interplay between nonlinearity and topology. Here, we theoretically investigate Thouless pumping of vector solitons in a two-component Bose-Einstein condensate confined in spin-dependent optical superlattices, using both numerical solutions of the Gross-Pitaevskii equation and the Lagrangian variational approach. The spin-up and spin-down components experience superlattice potentials that are displaced by a tunable distance $d_r$, leading to a vector soliton state with a relative shift between its components. We demonstrate that $d_r$, as an independent degree of freedom, offers a novel control parameter for manipulating the nonlinear topological phase transition of vector solitons. Specifically, when $d_r=0$, both components are either pumped or arrested, depending on the interaction strength. When fixing the interaction strength and varying $d_r$, remarkably, we find that an arrested vector soliton can re-enter the pumped regime and exhibits a quantized shift. As $d_r$ continues to increase, the vector soliton transitions into a dynamically arrested state; however, with further increases in $d_r$, the quantized shift revives. Our work paves new routes for engineering nonlinear topological pumping of solitons in spinor systems by utilizing the relative motion degrees of freedom between different spin components.

cond-mat.quant-gas↗

arXiv2024

Uni-Med: A Unified Medical Generalist Foundation Model For Multi-Task Learning Via Connector-MoE

Multi-modal large language models (MLLMs) have shown impressive capabilities as a general-purpose interface for various visual and linguistic tasks. However, building a unified MLLM for multi-task learning in the medical field remains a thorny challenge. To mitigate the tug-of-war problem of multi-modal multi-task optimization in MLLMs, recent advances primarily focus on improving the LLM components, while neglecting the connector that bridges the gap between modalities. In this paper, we introduce Uni-Med, a novel medical generalist foundation model which consists of a universal visual feature extraction module, a connector mixture-of-experts (CMoE) module, and an LLM. Benefiting from the proposed CMoE that leverages a well-designed router with a mixture of projection experts at the connector, Uni-Med achieves efficient solution to the tug-of-war problem and can perform six different medical tasks including question answering, visual question answering, report generation, referring expression comprehension, referring expression generation and image classification. To the best of our knowledge, Uni-Med is the first effort to tackle multi-task interference at the connector in MLLMs. Extensive ablation experiments validate the effectiveness of introducing CMoE under any configuration, with up to an average 8% performance gains. We further provide interpretation analysis of the tug-of-war problem from the perspective of gradient optimization and parameter statistics. Compared to previous state-of-the-art medical MLLMs, Uni-Med achieves competitive or superior evaluation metrics on diverse tasks. Code and resources are available at https://github.com/tsinghua-msiip/Uni-Med.

cs.CV↗

arXiv2024

DiffuseST: Unleashing the Capability of the Diffusion Model for Style Transfer

Style transfer aims to fuse the artistic representation of a style image with the structural information of a content image. Existing methods train specific networks or utilize pre-trained models to learn content and style features. However, they rely solely on textual or spatial representations that are inadequate to achieve the balance between content and style. In this work, we propose a novel and training-free approach for style transfer, combining textual embedding with spatial features and separating the injection of content or style. Specifically, we adopt the BLIP-2 encoder to extract the textual representation of the style image. We utilize the DDIM inversion technique to extract intermediate embeddings in content and style branches as spatial features. Finally, we harness the step-by-step property of diffusion models by separating the injection of content and style in the target branch, which improves the balance between content preservation and style fusion. Various experiments have demonstrated the effectiveness and robustness of our proposed DiffeseST for achieving balanced and controllable style transfer results, as well as the potential to extend to other tasks.

cs.CV↗

arXiv2024

Utility maximization in constrained and unbounded financial markets: Applications to indifference valuation, regime switching, consumption and Epstein-Zin recursive utility

This memoir presents a systematic study of the utility maximization problem of an investor in a constrained and unbounded financial market. Building upon the work of Hu et al. (2005) [Ann. Appl. Probab., 15, 1691--1712] in a bounded framework, we extend our analysis to the more challenging unbounded case. Our methodology combines both methods of quadratic backward stochastic differential equations with unbounded solutions and convex duality. Central to our approach is the verification of the finite entropy condition, which plays a pivotal role in solving the underlying utility maximization problem and establishing the martingale property and the convex duality representation of the conditional value process. Through four distinct applications, we first study the utility indifference valuation of financial derivatives with unbounded payoffs, uncovering novel asymptotic behaviors as the risk aversion parameter approaches zero or infinity. Furthermore, we study the regime switching market model with unbounded random endowments and consumption-investment problems with unbounded random endowments, both constrained to portfolios chosen from a convex and closed set. Finally, we investigate the investment-consumption problem of an investor with the Epstein-Zin recursive utility in an unbounded financial market.

math.PR↗

arXiv2024

Recalibrating $\mathbb{R}$-order trees and $\mbox{Homeo}_+(S^1)$-representations of link groups

In this paper we study the left-orderability of $3$-manifold groups using an enhancement, called recalibration, of Calegari and Dunfield's "flipping" construction, used for modifying $\mbox{Homeo}_+(S^1)$-representations of the fundamental groups of closed $3$-manifolds. The added flexibility accorded by recalibration allows us to produce $\mbox{Homeo}_+(S^1)$-representations of hyperbolic link exteriors so that a chosen element in the peripheral subgroup is sent to any given rational rotation. We apply these representations to show that the branched covers of families of links associated to arbitrary epimorphisms of the link group onto a finite cyclic group are left-orderable. This applies, for instance, to fibered hyperbolic strongly quasipositive links. Our result on the orderability of branched covers implies that the degeneracy locus of any pseudo-Anosov flow on an alternating knot complement must be meridional, which generalizes the known result that the fractional Dehn twist coefficient of any hyperbolic fibered alternating knot is zero. Applications of these representations to order-detection of slopes are also discussed in the paper.

math.GT↗

arXiv2024

Multi-dimensional non-Markovian backward stochastic differential equations of interactively quadratic generators

This paper is devoted to a general solvability of multi-dimensional non-Markovian backward stochastic differential equations (BSDEs) with interactively quadratic generators. Some general structures of the generator $g$ are posed for both local and global existence and uniqueness results on BSDEs, which admit a general growth of the generator $g$ in the state variable $y$, and a quadratic growth of the $i$th component $g^i$ both in the $j$th row $z^j$ of the state variable $z$ for $j\neq i$ (by which we mean the ``{\it interactively quadratic}" growth) and in the $i$th row $z^i$ of $z$. We first establish an existence and uniqueness result on local bounded solutions and then several existence and uniqueness results on global bounded and unbounded solutions. They improve several existing works in the non-Markovian setting, and also incorporate some interesting examples, one of which is a partial answer to the problem posed in \citet{Jackson2023SPA}. A comprehensive study on the bounded solution of one-dimensional quadratic BSDEs with unbounded stochastic parameters is provided for deriving our main results.

math.PR↗

arXiv2024

Magnet: We Never Know How Text-to-Image Diffusion Models Work, Until We Learn How Vision-Language Models Function

Text-to-image diffusion models particularly Stable Diffusion, have revolutionized the field of computer vision. However, the synthesis quality often deteriorates when asked to generate images that faithfully represent complex prompts involving multiple attributes and objects. While previous studies suggest that blended text embeddings lead to improper attribute binding, few have explored this in depth. In this work, we critically examine the limitations of the CLIP text encoder in understanding attributes and investigate how this affects diffusion models. We discern a phenomenon of attribute bias in the text space and highlight a contextual issue in padding embeddings that entangle different concepts. We propose \textbf{Magnet}, a novel training-free approach to tackle the attribute binding problem. We introduce positive and negative binding vectors to enhance disentanglement, further with a neighbor strategy to increase accuracy. Extensive experiments show that Magnet significantly improves synthesis quality and binding accuracy with negligible computational cost, enabling the generation of unconventional and unnatural concepts.

cs.CV↗

arXiv2024

JSJ decompositions of knot exteriors, Dehn surgery and the $L$-space conjecture

In this article, we apply slope detection techniques to study properties of toroidal $3$-manifolds obtained by performing Dehn surgeries on satellite knots in the context of the $L$-space conjecture. We show that if $K$ is an $L$-space knot or admits an irreducible rational surgery with non-left-orderable fundamental group, then the JSJ graph of its exterior is a rooted interval. Consequently, any rational surgery on a composite knot has a left-orderable fundamental group. This is the left-orderable counterpart of Krcatovich's result on the primeness of $L$-space knots, which we reprove using our methods. Analogous results on the existence of co-orientable taut foliations are proved when the knot has a fibred companion. Our results suggest a new approach to establishing the counterpart of Krcatovich's result for surgeries with co-orientable taut foliations, on which partial results have been achieved by Delman and Roberts. Finally, we prove results on left-orderable $p/q$-surgeries on knots with $p$ small.

math.GT↗

arXiv2024

Dissipative Nonlinear Thouless Pumping of Temporal Solitons

The interplay between topology and soliton is a central topic in nonlinear topological physics. So far, most studies have been confined to conservative settings. Here, we explore Thouless pumping of dissipative temporal solitons in a nonconservative one-dimensional optical system with gain and spectral filtering, described by the paradigmatic complex Ginzburg-Landau equation. Two dissipatively induced nonlinear topological phase transitions are identified. First, when varying dissipative parameters across a threshold, the soliton transitions from being trapped in time to quantized drifting. This quantized temporal drift remains robust, even as the system evolves from a single-soliton state into multi-soliton state. Second, a dynamically emergent phase transition is found: the soliton is arrested until a critical point of its evolution, where a transition to topological drift occurs. Both phenomena uniquely arise from the dynamical interplay of dissipation, nonlinearity and topology.

cond-mat.quant-gas↗

arXiv2024

Nonlinear Thouless Pumping of Solitons Across an Impurity

The nonlinear Thouless pumping is an exciting frontier of topological physics. While recent works have revealed the quantized motion of solitons in Thouless pumps, the interplay between the topology, nonlinearity and disorder remains largely unexplored. Here, we investigate the nonlinear Thouless pumping of solitons in the presence of an impurity in the context of a Bose-Einstein condensate. Using both the Gross-Pitaevskii equation and Lagrangian variational approach, we analyze the interaction between a moving soliton and an impurity. Without the pump, the soliton can pass through a light impurity, but gets trapped by the impurity with large mass. In marked contrast, we find the soliton in Thouless pumps can always transit through the impurity, and its motion is topologically quantized. Our result explicitly showcases the robustness of topological soliton pumping against microscopic imperfections, and opens a new perspective in the information processing with solitons.

cond-mat.quant-gas↗

arXiv2024

SparseSSP: 3D Subcellular Structure Prediction from Sparse-View Transmitted Light Images

Traditional fluorescence staining is phototoxic to live cells, slow, and expensive; thus, the subcellular structure prediction (SSP) from transmitted light (TL) images is emerging as a label-free, faster, low-cost alternative. However, existing approaches utilize 3D networks for one-to-one voxel level dense prediction, which necessitates a frequent and time-consuming Z-axis imaging process. Moreover, 3D convolutions inevitably lead to significant computation and GPU memory overhead. Therefore, we propose an efficient framework, SparseSSP, predicting fluorescent intensities within the target voxel grid in an efficient paradigm instead of relying entirely on 3D topologies. In particular, SparseSSP makes two pivotal improvements to prior works. First, SparseSSP introduces a one-to-many voxel mapping paradigm, which permits the sparse TL slices to reconstruct the subcellular structure. Secondly, we propose a hybrid dimensions topology, which folds the Z-axis information into channel features, enabling the 2D network layers to tackle SSP under low computational cost. We conduct extensive experiments to validate the effectiveness and advantages of SparseSSP on diverse sparse imaging ratios, and our approach achieves a leading performance compared to pure 3D topologies. SparseSSP reduces imaging frequencies compared to previous dense-view SSP (i.e., the number of imaging is reduced up to 87.5% at most), which is significant in visualizing rapid biological dynamics on low-cost devices and samples.

cs.CV↗

arXiv2024

Exceptional nexus in Bose-Einstein condensates with collective dissipation

In multistate non-Hermitian systems, higher-order exceptional points and exotic phenomena with no analogues in two-level systems arise. A paradigm is the exceptional nexus (EX), a third-order EP as the cusp singularity of exceptional arcs (EAs), that has a hybrid topological nature. Using atomic Bose-Einstein condensates to implement a dissipative three-state system, we experimentally realize an EX within a two-parameter space, despite the absence of symmetry. The engineered dissipation exhibits density dependence due to the collective atomic response to resonant light. Based on extensive analysis of the system's decay dynamics, we demonstrate the formation of an EX from the coalescence of two EAs with distinct geometries. These structures arise from the different roles played by dissipation in the strong coupling limit and quantum Zeno regime. Our work paves the way for exploring higher-order exceptional physics in the many-body setting of ultracold atoms.

quant-ph↗

arXiv2024

Ultrasound Report Generation with Cross-Modality Feature Alignment via Unsupervised Guidance

Automatic report generation has arisen as a significant research area in computer-aided diagnosis, aiming to alleviate the burden on clinicians by generating reports automatically based on medical images. In this work, we propose a novel framework for automatic ultrasound report generation, leveraging a combination of unsupervised and supervised learning methods to aid the report generation process. Our framework incorporates unsupervised learning methods to extract potential knowledge from ultrasound text reports, serving as the prior information to guide the model in aligning visual and textual features, thereby addressing the challenge of feature discrepancy. Additionally, we design a global semantic comparison mechanism to enhance the performance of generating more comprehensive and accurate medical reports. To enable the implementation of ultrasound report generation, we constructed three large-scale ultrasound image-text datasets from different organs for training and validation purposes. Extensive evaluations with other state-of-the-art approaches exhibit its superior performance across all three datasets. Code and dataset are valuable at this link.

cs.CV↗