arXiv subjects

Yong Chen

Publications and source records attributed to Yong Chen.

At least 163 records · Page 9Linked to original sources

arXiv2024

From Basic to Extra Features: Hypergraph Transformer Pretrain-then-Finetuning for Balanced Clinical Predictions on EHR

Electronic Health Records (EHRs) contain rich patient information and are crucial for clinical research and practice. In recent years, deep learning models have been applied to EHRs, but they often rely on massive features, which may not be readily available for all patients. We propose HTP-Star, which leverages hypergraph structures with a pretrain-then-finetune framework for modeling EHR data, enabling seamless integration of additional features. Additionally, we design two techniques, namely (1) Smoothness-inducing Regularization and (2) Group-balanced Reweighting, to enhance the model's robustness during fine-tuning. Through experiments conducted on two real EHR datasets, we demonstrate that HTP-Star consistently outperforms various baselines while striking a balance between patients with basic and extra features.

cs.LG↗

arXiv2024

Research on the Tender Leaf Identification and Mechanically Perceptible Plucking Finger for High-quality Green Tea

BACKGROUND: Intelligent identification and precise plucking are the keys to intelligent tea harvesting robots, which are of increasing significance nowadays. Aiming at plucking tender leaves for high-quality green tea producing, in this paper, a tender leaf identification algorithm and a mechanically perceptible plucking finger have been proposed. RESULTS: Based on segmentation algorithm and color features, the tender leaf identification algorithm shows an average identification accuracy of over 92.8%. The mechanically perceptible plucking finger plucks tender leaves in a way that a human hand does so as to remain high quality of tea products. Though finite element analysis, we determine the ideal size of grippers and the location of strain gauge attachment on a gripper to enable the employment of feedback control of desired gripping force. Revealed from our experiments, the success rate of tender leaf plucking reaches 92.5%, demonstrating the effectiveness of our design. CONCLUSION: The results show that the tender leaf identification algorithm and the mechanically perceptible plucking finger are effective for tender leaves identification and plucking, providing a foundation for the development of an intelligent tender leaf plucking robot.

cs.RO↗

arXiv2024

Unleashing the Power of Multi-Task Learning: A Comprehensive Survey Spanning Traditional, Deep, and Pretrained Foundation Model Eras

MTL is a learning paradigm that effectively leverages both task-specific and shared information to address multiple related tasks simultaneously. In contrast to STL, MTL offers a suite of benefits that enhance both the training process and the inference efficiency. MTL's key advantages encompass streamlined model architecture, performance enhancement, and cross-domain generalizability. Over the past twenty years, MTL has become widely recognized as a flexible and effective approach in various fields, including CV, NLP, recommendation systems, disease prognosis and diagnosis, and robotics. This survey provides a comprehensive overview of the evolution of MTL, encompassing the technical aspects of cutting-edge methods from traditional approaches to deep learning and the latest trend of pretrained foundation models. Our survey methodically categorizes MTL techniques into five key areas: regularization, relationship learning, feature propagation, optimization, and pre-training. This categorization not only chronologically outlines the development of MTL but also dives into various specialized strategies within each category. Furthermore, the survey reveals how the MTL evolves from handling a fixed set of tasks to embracing a more flexible approach free from task or modality constraints. It explores the concepts of task-promptable and -agnostic training, along with the capacity for ZSL, which unleashes the untapped potential of this historically coveted learning paradigm. Overall, we hope this survey provides the research community with a comprehensive overview of the advancements in MTL from its inception in 1997 to the present in 2023. We address present challenges and look ahead to future possibilities, shedding light on the opportunities and potential avenues for MTL research in a broad manner. This project is publicly available at https://github.com/junfish/Awesome-Multitask-Learning.

cs.LG↗

arXiv2024

New Timing Results of MSPs from NICER Observations

Millisecond pulsars (MSPs) are known for their long-term stability. Using six years of observations from the Neutron Star Interior Composition Explorer (NICER), we have conducted an in-depth analysis of the X-ray timing results for six MSPs: PSRs B1937+21, B1821$-$24, J0437$-$4715, J0030+0451, J0218+4232, and J2124$-$3358. The timing stability parameter $σ_z$ has been calculated, revealing remarkable timing precision on the order of $10^{-14}$ for PSRs B1937+21 and J0437$-$4715, and $10^{-13}$ for PSRs B1821$-$24, J0218+4232, and J0030+0451 over a timescale of 1000 days. These findings underscore the feasibility of autonomous in-orbit timekeeping using X-ray observations of MSPs. In addition, the consistency of long-term spin-down noise in the X-ray and radio bands has been investigated by comparison with IPTA radio data.

astro-ph.HE↗

arXiv2024

GraphAD: Interaction Scene Graph for End-to-end Autonomous Driving

Modeling complicated interactions among the ego-vehicle, road agents, and map elements has been a crucial part for safety-critical autonomous driving. Previous works on end-to-end autonomous driving rely on the attention mechanism for handling heterogeneous interactions, which fails to capture the geometric priors and is also computationally intensive. In this paper, we propose the Interaction Scene Graph (ISG) as a unified method to model the interactions among the ego-vehicle, road agents, and map elements. With the representation of the ISG, the driving agents aggregate essential information from the most influential elements, including the road agents with potential collisions and the map elements to follow. Since a mass of unnecessary interactions are omitted, the more efficient scene-graph-based framework is able to focus on indispensable connections and leads to better performance. We evaluate the proposed method for end-to-end autonomous driving on the nuScenes dataset. Compared with strong baselines, our method significantly outperforms in the full-stack driving tasks, including perception, prediction, and planning. Code will be released at https://github.com/zhangyp15/GraphAD.

cs.CV↗

arXiv2024

Leveraging Generative AI for Clinical Evidence Summarization Needs to Ensure Trustworthiness

Evidence-based medicine promises to improve the quality of healthcare by empowering medical decisions and practices with the best available evidence. The rapid growth of medical evidence, which can be obtained from various sources, poses a challenge in collecting, appraising, and synthesizing the evidential information. Recent advancements in generative AI, exemplified by large language models, hold promise in facilitating the arduous task. However, developing accountable, fair, and inclusive models remains a complicated undertaking. In this perspective, we discuss the trustworthiness of generative AI in the context of automated summarization of medical evidence.

cs.AI↗

arXiv2024

HAM-TTS: Hierarchical Acoustic Modeling for Token-Based Zero-Shot Text-to-Speech with Model and Data Scaling

Token-based text-to-speech (TTS) models have emerged as a promising avenue for generating natural and realistic speech, yet they grapple with low pronunciation accuracy, speaking style and timbre inconsistency, and a substantial need for diverse training data. In response, we introduce a novel hierarchical acoustic modeling approach complemented by a tailored data augmentation strategy and train it on the combination of real and synthetic data, scaling the data size up to 650k hours, leading to the zero-shot TTS model with 0.8B parameters. Specifically, our method incorporates a latent variable sequence containing supplementary acoustic information based on refined self-supervised learning (SSL) discrete units into the TTS model by a predictor. This significantly mitigates pronunciation errors and style mutations in synthesized speech. During training, we strategically replace and duplicate segments of the data to enhance timbre uniformity. Moreover, a pretrained few-shot voice conversion model is utilized to generate a plethora of voices with identical content yet varied timbres. This facilitates the explicit learning of utterance-level one-to-many mappings, enriching speech diversity and also ensuring consistency in timbre. Comparative experiments (Demo page: https://anonymous.4open.science/w/ham-tts/)demonstrate our model's superiority over VALL-E in pronunciation precision and maintaining speaking style, as well as timbre continuity.

cs.SD↗

arXiv2024

kNN-CTC: Enhancing ASR via Retrieval of CTC Pseudo Labels

The success of retrieval-augmented language models in various natural language processing (NLP) tasks has been constrained in automatic speech recognition (ASR) applications due to challenges in constructing fine-grained audio-text datastores. This paper presents kNN-CTC, a novel approach that overcomes these challenges by leveraging Connectionist Temporal Classification (CTC) pseudo labels to establish frame-level audio-text key-value pairs, circumventing the need for precise ground truth alignments. We further introduce a skip-blank strategy, which strategically ignores CTC blank frames, to reduce datastore size. kNN-CTC incorporates a k-nearest neighbors retrieval mechanism into pre-trained CTC ASR systems, achieving significant improvements in performance. By incorporating a k-nearest neighbors retrieval mechanism into pre-trained CTC ASR systems and leveraging a fine-grained, pruned datastore, kNN-CTC consistently achieves substantial improvements in performance under various experimental settings. Our code is available at https://github.com/NKU-HLT/KNN-CTC.

cs.SD↗

arXiv2024

$PT$ Symmetric PINN for integrable nonlocal equations: Forward and inverse problems

Since the $PT$-symmetric nonlocal equations contain the physical information of the $PT$-symmetric, it is very appropriate to embed the physical information of the $PT$-symmetric into the loss function of PINN, named PTS-PINN. For general $PT$-symmetric nonlocal equations, especially those equations involving the derivation of nonlocal terms, due to the existence of nonlocal terms, directly using the original PINN method to solve such nonlocal equations will face certain challenges. This problem can be solved by the PTS-PINN method which can be illustrated in two aspects. First, we treat the nonlocal term of the equation as a new local component, so that the equation is coupled at this time. In this way, we successfully avoid differentiating nonlocal terms in neural networks. On the other hand, in order to improve the accuracy, we make a second improvement, which is to embed the physical information of the $PT$-symmetric into the loss function. Through a series of independent numerical experiments, we evaluate the efficacy of PTS-PINN in tackling the forward and inverse problems for the nonlocal nonlinear Schrödinger (NLS) equation, the nonlocal derivative NLS equation, the nonlocal (2+1)-dimensional NLS equation, and the nonlocal three wave interaction systems. The numerical experiments demonstrate that PTS-PINN has good performance. In particular, PTS-PINN has also demonstrated an extraordinary ability to learn large space-time scale rogue waves for nonlocal equations.

physics.comp-ph↗

arXiv2024

Large Model based Sequential Keyframe Extraction for Video Summarization

Keyframe extraction aims to sum up a video's semantics with the minimum number of its frames. This paper puts forward a Large Model based Sequential Keyframe Extraction for video summarization, dubbed LMSKE, which contains three stages as below. First, we use the large model "TransNetV21" to cut the video into consecutive shots, and employ the large model "CLIP2" to generate each frame's visual feature within each shot; Second, we develop an adaptive clustering algorithm to yield candidate keyframes for each shot, with each candidate keyframe locating nearest to a cluster center; Third, we further reduce the above candidate keyframes via redundancy elimination within each shot, and finally concatenate them in accordance with the sequence of shots as the final sequential keyframes. To evaluate LMSKE, we curate a benchmark dataset and conduct rich experiments, whose results exhibit that LMSKE performs much better than quite a few SOTA competitors with average F1 of 0.5311, average fidelity of 0.8141, and average compression ratio of 0.9922.

cs.CV↗

arXiv2024

Lax pairs informed neural networks solving integrable systems

Lax pairs are one of the most important features of integrable system. In this work, we propose the Lax pairs informed neural networks (LPNNs) tailored for the integrable systems with Lax pairs by designing novel network architectures and loss functions, comprising LPNN-v1 and LPNN-v2. The most noteworthy advantage of LPNN-v1 is that it can transform the solving of nonlinear integrable systems into the solving of a linear Lax pairs spectral problems, and it not only efficiently solves data-driven localized wave solutions, but also obtains spectral parameter and corresponding spectral function in Lax pairs spectral problems of the integrable systems. On the basis of LPNN-v1, we additionally incorporate the compatibility condition/zero curvature equation of Lax pairs in LPNN-v2, its major advantage is the ability to solve and explore high-accuracy data-driven localized wave solutions and associated spectral problems for integrable systems with Lax pairs. The numerical experiments focus on studying abundant localized wave solutions for very important and representative integrable systems with Lax pairs spectral problems, including the soliton solution of the Korteweg-de Vries (KdV) euqation and modified KdV equation, rogue wave solution of the nonlinear Schrödinger equation, kink solution of the sine-Gordon equation, non-smooth peakon solution of the Camassa-Holm equation and pulse solution of the short pulse equation, as well as the line-soliton solution of Kadomtsev-Petviashvili equation and lump solution of high-dimensional KdV equation. The innovation of this work lies in the pioneering integration of Lax pairs informed of integrable systems into deep neural networks, thereby presenting a fresh methodology and pathway for investigating data-driven localized wave solutions and Lax pairs spectral problems.

nlin.SI↗

arXiv2023

Fine-grained Disentangled Representation Learning for Multimodal Emotion Recognition

Multimodal emotion recognition (MMER) is an active research field that aims to accurately recognize human emotions by fusing multiple perceptual modalities. However, inherent heterogeneity across modalities introduces distribution gaps and information redundancy, posing significant challenges for MMER. In this paper, we propose a novel fine-grained disentangled representation learning (FDRL) framework to address these challenges. Specifically, we design modality-shared and modality-private encoders to project each modality into modality-shared and modality-private subspaces, respectively. In the shared subspace, we introduce a fine-grained alignment component to learn modality-shared representations, thus capturing modal consistency. Subsequently, we tailor a fine-grained disparity component to constrain the private subspaces, thereby learning modality-private representations and enhancing their diversity. Lastly, we introduce a fine-grained predictor component to ensure that the labels of the output representations from the encoders remain unchanged. Experimental results on the IEMOCAP dataset show that FDRL outperforms the state-of-the-art methods, achieving 78.34% and 79.44% on WAR and UAR, respectively.

cs.SD↗

arXiv2023

The improved backward compatible physics-informed neural networks for reducing error accumulation and applications in data-driven higher-order rogue waves

Due to the dynamic characteristics of instantaneity and steepness, employing domain decomposition techniques for simulating rogue wave solutions is highly appropriate. Wherein, the backward compatible PINN (bc-PINN) is a temporally sequential scheme to solve PDEs over successive time segments while satisfying all previously obtained solutions. In this work, we propose improvements to the original bc-PINN algorithm in two aspects based on the characteristics of error propagation. One is to modify the loss term for ensuring backward compatibility by selecting the earliest learned solution for each sub-domain as pseudo reference solution. The other is to adopt the concatenation of solutions obtained from individual subnetworks as the final form of the predicted solution. The improved backward compatible PINN (Ibc-PINN) is applied to study data-driven higher-order rogue waves for the nonlinear Schrödinger (NLS) equation and the AB system to demonstrate the effectiveness and advantages. Transfer learning and initial condition guided learning (ICGL) techniques are also utilized to accelerate the training. Moreover, the error analysis is conducted on each sub-domain and it turns out that the slowdown of Ibc-PINN in error accumulation speed can yield greater advantages in accuracy. In short, numerical results fully indicate that Ibc-PINN significantly outperforms bc-PINN in terms of accuracy and stability without sacrificing efficiency.

physics.comp-ph↗

arXiv2023

Oceanic internal solitary wave interactions via the KP equation in a three-layer fluid with shear flow

The various patterns of internal solitary wave interactions are complex phenomena in the ocean, susceptible to the influence of shear flow and density distributions. Satellite imagery serves as an effective tool for investigating these interactions, but usually does not provide information on the structure of internal waves and their associated dynamics. Considering a three-layer configuration that approximates ocean stratification, we analytically investigate two-dimensional internal solitary waves (ISW) in a three-layer fluid with shear flow and continuous density distribution using the (2+1)-dimensional Kadomtsev-Petviashvili (KP) model. Firstly, the KP equation is derived from the basic governing equations which include mass and momentum conservations, along with free surface boundary conditions. The coefficients of the KP equation are determined by the vertical distribution of fluid density, shear flow, and layer depth. Secondly, it is found that the interactions of ISW can be carefully classified into five types: ordinary interactions including O-type, asymmetric interactions including P-type, TP-type and TO-type, and Miles resonance. The genuine existence of these interaction types is observed from satellite images in the Andaman Sea, the Malacca Strait, and the coast of Washington state. Finally, the ``bright" and ``dark" internal solitary interactions are discovered in the three-layer fluid, which together constitute the fluctuating forms of oceanic ISW. It is revealed that shear flow is the primary factor to determine whether these types of interactions are ``bright" or ``dark". Besides, a detailed analysis is conducted to show how the ratio of densities influences the properties of these interactions, such as amplitude, angle, and wave width.

physics.flu-dyn↗

arXiv2023

The Properties of Fractional Gaussian Process and Their Applications

The process $(G_t)_{t\in[0,T]}$ is referred to as a fractional Gaussian process if the first-order partial derivative of the difference between its covariance function and that of the fractional Brownian motion $(B^H_t)_{t\in[0,T ]}$ is a normalized bounded variation function. We quantify the relation between the associated reproducing kernel Hilbert space of $(G)$ and that of $(B^H)$. Seven types of Gaussian processes with non-stationary increments in the literature belong to it. In the context of applications, we demonstrate that the Gladyshev's theorem holds for this process, and we provide Berry-Esséen upper bounds associated with the statistical estimations of the ergodic fractional Ornstein-Uhlenbeck process driven by it. The second application partially builds upon the idea introduced in \cite{BBES 23}, where they assume that $(G)$ has stationary increments. Additionally, we briefly discuss a variant of this process where the covariance structure is not entirely linked to that of the fractional Brownian motion.

math.PR↗

arXiv2023

Scalable high-dimensional Bayesian varying coefficient models with unknown within-subject covariance

Nonparametric varying coefficient (NVC) models are useful for modeling time-varying effects on responses that are measured repeatedly for the same subjects. When the number of covariates is moderate or large, it is desirable to perform variable selection from the varying coefficient functions. However, existing methods for variable selection in NVC models either fail to account for within-subject correlations or require the practitioner to specify a parametric form for the correlation structure. In this paper, we introduce the nonparametric varying coefficient spike-and-slab lasso (NVC-SSL) for Bayesian high-dimensional NVC models. Through the introduction of functional random effects, our method allows for flexible modeling of within-subject correlations without needing to specify a parametric covariance function. We further propose several scalable optimization and Markov chain Monte Carlo (MCMC) algorithms. For variable selection, we propose an Expectation Conditional Maximization (ECM) algorithm to rapidly obtain maximum a posteriori (MAP) estimates. Our ECM algorithm scales linearly in the total number of observations $N$ and the number of covariates $p$. For uncertainty quantification, we introduce an approximate MCMC algorithm that also scales linearly in both $N$ and $p$. We demonstrate the scalability, variable selection performance, and inferential capabilities of our method through simulations and a real data application. These algorithms are implemented in the publicly available R package NVCSSL on the Comprehensive R Archive Network.

stat.ME↗

arXiv2023

Long time and Painlevé-type asymptotics for the defocusing Hirota equation with finite density initial data

In this work, we consider the Cauchy problem for the defocusing Hirota equation with a nonzero background \begin{align} \begin{cases} iq_{t}+α\left[q_{xx}-2\left(\left\vert q\right\vert^{2}-1\right)q\right]+iβ\left(q_{xxx}-6\left\vert q\right\vert^{2}q_{x}\right)=0,\quad (x,t)\in \mathbb{R}\times(0,+\infty),\\ q(x,0)=q_{0}(x),\qquad \underset{x\rightarrow\pm\infty 1}{\lim} q_{0}(x)=\pm 1, \qquad q_{0}\mp 1\in H^{4,4}(\mathbb{R}). \end{cases} \nonumber \end{align} According to the Riemann-Hilbert problem representation of the Cauchy problem and the $\bar{\partial}$ generalization of the nonlinear steepest descent method, we find different long time asymptotics types for the defocusing Hirota equation in oscillating region and transition region, respectively. For the oscillating region $ξ<-8$, four phase points appear on the jump contour $\mathbb{R}$, which arrives at an asymptotic expansion,given by \begin{align} q(x,t)=-1+t^{-1/2}h+O(t^{-3/4}).\nonumber \end{align} It consists of three terms. The first term $-1$ is leading term representing a nonzero background, the second term $t^{-1/2}h$ originates from the continuous spectrum and the third term $O(t^{-3/4})$ is the error term due to pure $\bar{\partial}$-RH problem. For the transition region $\vertξ+8\vert t^{2/3}<C$, three phase points raise on the jump contour $\mathbb{R}$. Painlevé asymptotics expansion is obtained \begin{align} q(x,t)=-1-(\frac{15}{4}t)^{-1/3}\varrho+O(t^{-1/2}),\nonumber \end{align} in which the leading term is a solution to the Painlevé II equation, the last term is a residual error being from pure $\bar{\partial}$-RH problem and parabolic cylinder model.

nlin.SI↗

arXiv2023

Long-time asymptotics for the Elastic Beam equation in the solitonless region via $\bar{\partial}$ methods

In this work, we study the Cauchy problem of the Elastic Beam equation with initial value in weighted Sobolev space $H^{1,1}(\mathbb{R})$ via the $\bar{\partial}$-steepset descent method. Begin with the Lax pair of the Elastic Beam equation, we successfully derive the basic Riemann-Hilbert problem, which can be used to represent the solutions of the Elastic Beam equation. Then, considering the solitonless region and using the $\bar{\partial}$-steepset descent method, we analyse the long-time asymptotic behaviors of the solutions for the Elastic Beam equation.

math.AP↗