arXiv subjects

Zhi Yang

Publications and source records attributed to Zhi Yang.

At least 55 records · Page 3Linked to original sources

arXiv2024

FinEval: A Chinese Financial Domain Knowledge Evaluation Benchmark for Large Language Models

Large language models have demonstrated outstanding performance in various natural language processing tasks, but their security capabilities in the financial domain have not been explored, and their performance on complex tasks like financial agent remains unknown. This paper presents FinEval, a benchmark designed to evaluate LLMs' financial domain knowledge and practical abilities. The dataset contains 8,351 questions categorized into four different key areas: Financial Academic Knowledge, Financial Industry Knowledge, Financial Security Knowledge, and Financial Agent. Financial Academic Knowledge comprises 4,661 multiple-choice questions spanning 34 subjects such as finance and economics. Financial Industry Knowledge contains 1,434 questions covering practical scenarios like investment research. Financial Security Knowledge assesses models through 1,640 questions on topics like application security and cryptography. Financial Agent evaluates tool usage and complex reasoning with 616 questions. FinEval has multiple evaluation settings, including zero-shot, five-shot with chain-of-thought, and assesses model performance using objective and subjective criteria. Our results show that Claude 3.5-Sonnet achieves the highest weighted average score of 72.9 across all financial domain categories under zero-shot setting. Our work provides a comprehensive benchmark closely aligned with Chinese financial domain.

cs.CL↗

arXiv2024

Binding memory of liquid molecules

Understanding the binding dynamics of liquid molecules is of fundamental importance in physical and life sciences. However, nanoscale fast dynamics pose great challenges for experimental characterization. Conventionally, the binding dynamics have been assumed to be memoryless. Here, we integrate large scale computer simulation, scaling theory, and real-time single particle tracking microscopy with high spatiotemporal precision to unveil a universal memory effect in the binding dynamics of liquid molecules. This binding memory can be quantified by a binding time autocorrelation function, whose power-law decay depends not only on the binding affinity, but also on the topological and materials properties of the surrounding environment. Context-dependent biomolecular binding memory is likely exploited by biological systems to regulate biochemical reactions and biophysical processes. Deciphering this binding memory offers a novel strategy to probe complex biological systems and advanced soft materials.

physics.chem-ph↗

arXiv2024

Design of Ligand-Binding Proteins with Atomic Flow Matching

Designing novel proteins that bind to small molecules is a long-standing challenge in computational biology, with applications in developing catalysts, biosensors, and more. Current computational methods rely on the assumption that the binding pose of the target molecule is known, which is not always feasible, as conformations of novel targets are often unknown and tend to change upon binding. In this work, we formulate proteins and molecules as unified biotokens, and present AtomFlow, a novel deep generative model under the flow-matching framework for the design of ligand-binding proteins from the 2D target molecular graph alone. Operating on representative atoms of biotokens, AtomFlow captures the flexibility of ligands and generates ligand conformations and protein backbone structures iteratively. We consider the multi-scale nature of biotokens and demonstrate that AtomFlow can be effectively trained on a subset of structures from the Protein Data Bank, by matching flow vector field using an SE(3) equivariant structure prediction network. Experimental results show that our method can generate high fidelity ligand-binding proteins and achieve performance comparable to the state-of-the-art model RFDiffusionAA, while not requiring bound ligand structures. As a general framework, AtomFlow holds the potential to be applied to various biomolecule generation tasks in the future.

q-bio.BM↗

arXiv2024

A Three-Coupled-Channel Analysis of $Z_c(3900)$ Involving $D\bar{D}^*$, $πJ/ψ$, and $ρη_c $

In this work, we conduct a three-coupled-channel analysis of the $Z_c(3900)$ structure, focusing on the $D\bar{D}^*$, $J/ψπ$, and $ρη_c$ channels, based on the one-boson exchange model. Drawing from previous study on the exotic state $T_{cc}$, we only utilize one more parameter to construct the interactions between the channels. Our model successfully reproduces the experimental line shapes of the invariant mass distribution at $\sqrt{s} = 4.23$ and $4.26$ GeV for the three channels. Additionally, the finite-volume energy levels in our model show agreement with current LQCD conclusion. Detailed analysis suggests that the $Z_c(3900)$ peaks in the $πJ/ψ$ and $ρη_c$ distributions primarily arise from the triangle loop involving the $D_1 \bar{D} D^*$ intermediate system. In the $D\bar{D}^*$ distribution, the threshold peak is generated by the cascade decay mechanism enhanced by a triangle diagram. Moreover, we find a virtual pole located far from the threshold, indicating that $Z_c(3900)$ peaks are not associated with physical pole. We conclude that the $Z_c(3900)$ peaks are predominantly caused by the threshold cusp.

hep-ph↗

arXiv2024

Investigation on the bottom analogs of the $T_{cc}^{+}$

We investigate the doubly bottom state $T^-_{bb}$ composed of two bottom mesons with $J^P=1^+$. The potentials are obtained using the one-boson exchange model. With the heavy quark flavor symmetry, all the parameters in the model are determined by fitting the experimental data of doubly charmed state $T_{cc}^{+}$ from our previous work. Our analysis indicates that the isospin symmetry is well-preserved. We identify a deeply bound $T^-_{bb}$ state with quantum numbers $I(J^{P})=0(1^+)$, in contrast to the loosely bound $T_{cc}^{+}$. Additionally, we discover a resonant $T^-_{bb}$ state with $I(J^{P})=1(1^+)$. Furthermore, our investigation into the $\bar{B}\bar{B}^*$-$\bar{B}^*\bar{B}^*$ coupled channel effect reveals its important impact. The binding energy of the bound $I(J^{P})=0(1^+)$ states becomes deeper, while the resonant $T^-_{bb}$ state with $I(J^{P})=1(1^+)$ remains nearly unchanged.

hep-ph↗

arXiv2024

New insight into the exotic states strongly coupled with the $D\bar{D}^*$ from the $T^+_{cc}$

We have investigated the internal structure of the open- and hidden-charmed ($DD^*$/$\bar DD^*$) molecules in the unified framework. We first fit the experimental lineshape of the $T^+_{cc}$ state and extract the $DD^*$ interaction, from which the $T^+_{cc}$ is assumed to arise solely. Then we obtain the $D\bar{D}^*$ interaction by charge conjugation. Our results show that the $D\bar{D}^*$ interaction is attractive but insufficient to form $X(3872)$ as a bound state. Instead, its formation requires the crucial involvement of the coupled channel effect between the $D\bar D^*$ and $c\bar c$ components, although the $c\bar c$ accounts for approximately $1\%$ only. Besides $X(3872)$, we have obtained a higher-energy state around $3957.9$ MeV with a width of $16.7$ MeV, which may be a potential candidate for the $X(3940)$. In $J^{PC}=1^{+-}$ sector, we have found two states related to the iso-scalar $\tilde X(3872)$ and $h_c(2P)$, respectively. Our combined study provides valuable insights into the nature of these $DD^*$/$D\bar D^*$ exotic states.

hep-ph↗

arXiv2024

Graph Neural Networks Automated Design and Deployment on Device-Edge Co-Inference Systems

The key to device-edge co-inference paradigm is to partition models into computation-friendly and computation-intensive parts across the device and the edge, respectively. However, for Graph Neural Networks (GNNs), we find that simply partitioning without altering their structures can hardly achieve the full potential of the co-inference paradigm due to various computational-communication overheads of GNN operations over heterogeneous devices. We present GCoDE, the first automatic framework for GNN that innovatively Co-designs the architecture search and the mapping of each operation on Device-Edge hierarchies. GCoDE abstracts the device communication process into an explicit operation and fuses the search of architecture and the operations mapping in a unified space for joint-optimization. Also, the performance-awareness approach, utilized in the constraint-based search process of GCoDE, enables effective evaluation of architecture efficiency in diverse heterogeneous systems. We implement the co-inference engine and runtime dispatcher in GCoDE to enhance the deployment efficiency. Experimental results show that GCoDE can achieve up to $44.9\times$ speedup and $98.2\%$ energy reduction compared to existing approaches across various applications and system configurations.

cs.LG↗

arXiv2024

WAL-Net: Weakly supervised auxiliary task learning network for carotid plaques classification

The classification of carotid artery ultrasound images is a crucial means for diagnosing carotid plaques, holding significant clinical relevance for predicting the risk of stroke. Recent research suggests that utilizing plaque segmentation as an auxiliary task for classification can enhance performance by leveraging the correlation between segmentation and classification tasks. However, this approach relies on obtaining a substantial amount of challenging-to-acquire segmentation annotations. This paper proposes a novel weakly supervised auxiliary task learning network model (WAL-Net) to explore the interdependence between carotid plaque classification and segmentation tasks. The plaque classification task is primary task, while the plaque segmentation task serves as an auxiliary task, providing valuable information to enhance the performance of the primary task. Weakly supervised learning is adopted in the auxiliary task to completely break away from the dependence on segmentation annotations. Experiments and evaluations are conducted on a dataset comprising 1270 carotid plaque ultrasound images from Wuhan University Zhongnan Hospital. Results indicate that the proposed method achieved an approximately 1.3% improvement in carotid plaque classification accuracy compared to the baseline network. Specifically, the accuracy of mixed-echoic plaques classification increased by approximately 3.3%, demonstrating the effectiveness of our approach.

eess.IV↗

arXiv2023

Detecting bulk carbon ferromagnetism in graphene multi-edge structure

The emergence of bulk carbon ferromagnetism is long-expected over years. At nanoscale, carbon ferromagnetism was detected by analyzing the magnetic edge states via scanning tunneling microscopy(STM), and its origin can be explained by local redistribution of electron wave function. In larger scale, carbon ferromagnetism can be created by deliberately producing defects in graphite, and detected by macroscopic technical magnetization. Meanwhile, it becomes crucial to determine that the detected magnetization is originated from carbon rather than from magnetic impurities. One solution is X-ray magnetic circular dichroism (XMCD). Nonetheless, a reproducible, full section of XMCD spectrum across C-1s absorption energy has not appeared yet, which should be decisive for assuring the indisputable existence of bulk carbon ferromagnetism. Besides, the lack of direct observation on the atomic structure of the ferromagnetic carbon leaves the structural origin of its ferromagnetism still in mist. In this work, for detecting bulk carbon ferromagnetism, we managed to grow all-carbon film consisting of vertically aligned graphene multi-edge (VGME), which wove into a three-dimensional hyperfine-porous network. Magnetization (M-H) curves and XMCD spectra co-confirmed bulk carbon ferromagnetism of VGME at room temperature, with the average unit magnetic momentum of ~0.0006 miuB/atom. The influence of magnetic impurities on magnetization was excluded by both absorption spectra and inductively coupled plasma mass spectrometry measurements. The spin transfer behavior also verified the long-range and robust feature of the bulk carbon ferromagnetism. Our work provides direct evidence of elementary resolved bulk carbon ferromagnetism at room temperature and clarifies its origin from pi-electrons at graphene edges.

cond-mat.mtrl-sci↗

arXiv2023

Self-training with dual uncertainty for semi-supervised medical image segmentation

In the field of semi-supervised medical image segmentation, the shortage of labeled data is the fundamental problem. How to effectively learn image features from unlabeled images to improve segmentation accuracy is the main research direction in this field. Traditional self-training methods can partially solve the problem of insufficient labeled data by generating pseudo labels for iterative training. However, noise generated due to the model's uncertainty during training directly affects the segmentation results. Therefore, we added sample-level and pixel-level uncertainty to stabilize the training process based on the self-training framework. Specifically, we saved several moments of the model during pre-training, and used the difference between their predictions on unlabeled samples as the sample-level uncertainty estimate for that sample. Then, we gradually add unlabeled samples from easy to hard during training. At the same time, we added a decoder with different upsampling methods to the segmentation network and used the difference between the outputs of the two decoders as pixel-level uncertainty. In short, we selectively retrained unlabeled samples and assigned pixel-level uncertainty to pseudo labels to optimize the self-training process. We compared the segmentation results of our model with five semi-supervised approaches on the public 2017 ACDC dataset and 2018 Prostate dataset. Our proposed method achieves better segmentation performance on both datasets under the same settings, demonstrating its effectiveness, robustness, and potential transferability to other medical image segmentation tasks. Keywords: Medical image segmentation, semi-supervised learning, self-training, uncertainty estimation

cs.CV↗

arXiv2023

Fusing Structural and Functional Connectivities using Disentangled VAE for Detecting MCI

Brain network analysis is a useful approach to studying human brain disorders because it can distinguish patients from healthy people by detecting abnormal connections. Due to the complementary information from multiple modal neuroimages, multimodal fusion technology has a lot of potential for improving prediction performance. However, effective fusion of multimodal medical images to achieve complementarity is still a challenging problem. In this paper, a novel hierarchical structural-functional connectivity fusing (HSCF) model is proposed to construct brain structural-functional connectivity matrices and predict abnormal brain connections based on functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI). Specifically, the prior knowledge is incorporated into the separators for disentangling each modality of information by the graph convolutional networks (GCN). And a disentangled cosine distance loss is devised to ensure the disentanglement's effectiveness. Moreover, the hierarchical representation fusion module is designed to effectively maximize the combination of relevant and effective features between modalities, which makes the generated structural-functional connectivity more robust and discriminative in the cognitive disease analysis. Results from a wide range of tests performed on the public Alzheimer's Disease Neuroimaging Initiative (ADNI) database show that the proposed model performs better than competing approaches in terms of classification evaluation. In general, the proposed HSCF model is a promising model for generating brain structural-functional connectivities and identifying abnormal brain connections as cognitive disease progresses.

eess.IV↗

arXiv2023

HD-GCN:A Hybrid Diffusion Graph Convolutional Network

The information diffusion performance of GCN and its variant models is limited by the adjacency matrix, which can lower their performance. Therefore, we introduce a new framework for graph convolutional networks called Hybrid Diffusion-based Graph Convolutional Network (HD-GCN) to address the limitations of information diffusion caused by the adjacency matrix. In the HD-GCN framework, we initially utilize diffusion maps to facilitate the diffusion of information among nodes that are adjacent to each other in the feature space. This allows for the diffusion of information between similar points that may not have an adjacent relationship. Next, we utilize graph convolution to further propagate information among adjacent nodes after the diffusion maps, thereby enabling the spread of information among similar nodes that are adjacent in the graph. Finally, we employ the diffusion distances obtained through the use of diffusion maps to regularize and constrain the predicted labels of training nodes. This regularization method is then applied to the HD-GCN training, resulting in a smoother classification surface. The model proposed in this paper effectively overcomes the limitations of information diffusion imposed only by the adjacency matrix. HD-GCN utilizes hybrid diffusion by combining information diffusion between neighborhood nodes in the feature space and adjacent nodes in the adjacency matrix. This method allows for more comprehensive information propagation among nodes, resulting in improved model performance. We evaluated the performance of DM-GCN on three well-known citation network datasets and the results showed that the proposed framework is more effective than several graph-based semi-supervised learning methods.

cs.LG↗

arXiv2022

The investigations of the $P$-wave $B_s$ states combining quark model and lattice QCD in the coupled channel framework

Combining the quark model, the quark-pair-creation mechanism and $B^{(*)}\bar K$ interaction, we have investigated the near-threshold $P$-wave $B_s$ states in the framework of the Hamiltonian effective field theory. With the heavy quark flavor symmetry, all the parameters are determined in the $D_s$ sector by fitting the lattice data. The masses of the bottom-strange partners of the $D^{*}_{s0}(2317)$ and $D^{*}_{s1}(2460)$ are predicted to be $M_{B^{*}_{s0}}=5730.2_{-1.5}^{+2.4}$ MeV and $M_{B^{*}_{s1}}= 5769.6_{-1.6}^{+2.4}$ MeV, respectively, which are well consistent with the lattice QCD simulation. The two $P$-wave $B_s$ states are the mixtures of the bare $\bar b s$ core and $B^{(*)}\bar K$ component. Moreover, we find a crossing point between the energy levels with and without the interaction Hamiltonian in the finite volume spectrum in the $0^+$ case, which corresponds to a CDD (Castillejo-Dalitz-Dyson) zero in the $T$-matrix of the $B\bar K$ scattering. This CDD zero will help deepen the insights of the near-threshold states and can be examined by future lattice calculation.

hep-lat↗

arXiv2022

Semi-inclusive electroproduction of hidden-charm and double-charm hadronic molecules

The semi-inclusive electroproduction of exotic hadrons, including the $T_{cc}$, $P_{cs}$, and hidden-charm baryon-antibaryon states, is explored under the assumption that they are $S$-wave hadronic molecules of a pair of charmed hadrons. We employ the Monte Carlo event generator Pythia to produce the hadron pairs and then bind them together to form hadronic molecules. With the use of such a production mechanism, the semi-inclusive electroproduction rates are estimated at the order-of-magnitude level. Our results indicate that a larger number of $P_{cs}$ states and $Λ_c\barΛ_c$ molecules can be produced at the proposed electron-ion colliders in China (EicC) and in the US (EIC). The results also suggest that the $T_{cc}$ states and other hidden-charm baryon-antibaryon states can be searched for at EIC. Besides, the potential 24-GeV upgrade of the Continuous Beam Accelerator Facility at the Thomas Jefferson National Accelerator Facility can play an important role in the search for the hidden-charm tetraquark and pentaquark states due to its high luminosity.

hep-ph↗

arXiv2022

EvoMoE: An Evolutional Mixture-of-Experts Training Framework via Dense-To-Sparse Gate

Mixture-of-experts (MoE) is becoming popular due to its success in improving the model quality, especially in Transformers. By routing tokens with a sparse gate to a few experts (i.e., a small pieces of the full model), MoE can easily increase the model parameters to a very large scale while keeping the computation cost in a constant level. Most existing works just initialize some random experts, set a fixed gating strategy (e.g., Top-k), and train the model from scratch in an ad-hoc way. We identify that these MoE models are suffering from the immature experts and unstable sparse gate, which are harmful to the convergence performance. In this paper, we propose an efficient end-to-end MoE training framework called EvoMoE. EvoMoE starts from training one single expert and gradually evolves into a large and sparse MoE structure. EvoMoE mainly contains two phases: the expert-diversify phase to train the base expert for a while and spawn multiple diverse experts from it, and the gate-sparsify phase to learn an adaptive sparse gate and activate a dynamic number of experts. EvoMoE naturally decouples the joint learning of both the experts and the sparse gate and focuses on learning the basic knowledge with a single expert at the early training stage. Then it diversifies the experts and continues to train the MoE with a novel Dense-to-Sparse gate (DTS-Gate). Specifically, instead of using a permanent sparse gate, DTS-Gate begins as a dense gate that routes tokens to all experts, then gradually and adaptively becomes sparser while routes to fewer experts. Evaluations are conducted on three popular models and tasks, including RoBERTa for masked language modeling task, GPT for language modeling task and Transformer for machine translation task. The results show that EvoMoE outperforms existing baselines, including Switch, BASE Layer, Hash Layer and StableMoE.

cs.LG↗

arXiv2022

Large Deviation Principle for Backward Stochastic Differential Equations with a stochastic Lipschitz condition on $z$

In this paper, a probabilistic interpretation for the viscosity solution of a parabolic partial differential equation is obtained by virtue of the solution of a class of quadratic backward stochastic differential equations (BSDEs, for short). Furthermore, we prove the convergence and the large deviation principle for the solution of this class of quadratic BSDEs, which is associated with a family of Markov processes with the diffusion coefficients that tend to be zero.

math.PR↗

arXiv2022

On the uniqueness result for the BSDE with continuous coefficient

In this paper, we study one-dimensional backward stochastic differential equation (BSDE, for short), whose coefficient $f$ is Lipschitz in $y$ but only continuous in $z$. In addition, if the terminal condition $ξ$ has bounded Malliavin derivative, we prove some uniqueness results for the BSDE with quadratic and linear growth in $z$, respectively.

math.PR↗

arXiv2022

Unified Normalization for Accelerating and Stabilizing Transformers

Solid results from Transformers have made them prevailing architectures in various natural language and vision tasks. As a default component in Transformers, Layer Normalization (LN) normalizes activations within each token to boost the robustness. However, LN requires on-the-fly statistics calculation in inference as well as division and square root operations, leading to inefficiency on hardware. What is more, replacing LN with other hardware-efficient normalization schemes (e.g., Batch Normalization) results in inferior performance, even collapse in training. We find that this dilemma is caused by abnormal behaviors of activation statistics, including large fluctuations over iterations and extreme outliers across layers. To tackle these issues, we propose Unified Normalization (UN), which can speed up the inference by being fused with other linear operations and achieve comparable performance on par with LN. UN strives to boost performance by calibrating the activation and gradient statistics with a tailored fluctuation smoothing strategy. Meanwhile, an adaptive outlier filtration strategy is applied to avoid collapse in training whose effectiveness is theoretically proved and experimentally verified in this paper. We demonstrate that UN can be an efficient drop-in alternative to LN by conducting extensive experiments on language and vision tasks. Besides, we evaluate the efficiency of our method on GPU. Transformers equipped with UN enjoy about 31% inference speedup and nearly 18% memory reduction. Code will be released at https://github.com/hikvision-research/Unified-Normalization.

cs.CV↗