arXiv subjects

Yifan Sun

Publications and source records attributed to Yifan Sun.

At least 199 records · Page 11Linked to original sources

arXiv2022

Semisoft Task Clustering for Multi-Task Learning

Multi-task learning (MTL) aims to improve the performance of multiple related prediction tasks by leveraging useful information from them. Due to their flexibility and ability to reduce unknown coefficients substantially, the task-clustering-based MTL approaches have attracted considerable attention. Motivated by the idea of semisoft clustering of data, we propose a semisoft task clustering approach, which can simultaneously reveal the task cluster structure for both pure and mixed tasks as well as select the relevant features. The main assumption behind our approach is that each cluster has some pure tasks, and each mixed task can be represented by a linear combination of pure tasks in different clusters. To solve the resulting non-convex constrained optimization problem, we design an efficient three-step algorithm. The experimental results based on synthetic and real-world datasets validate the effectiveness and efficiency of the proposed approach. Finally, we extend the proposed approach to a robust task clustering problem.

cs.LG↗

arXiv2022

Regression-based heterogeneity analysis to identify overlapping subgroup structure in high-dimensional data

Heterogeneity is a hallmark of complex diseases. Regression-based heterogeneity analysis, which is directly concerned with outcome-feature relationships, has led to a deeper understanding of disease biology. Such an analysis identifies the underlying subgroup structure and estimates the subgroup-specific regression coefficients. However, most of the existing regression-based heterogeneity analyses can only address disjoint subgroups; that is, each sample is assigned to only one subgroup. In reality, some samples have multiple labels, for example, many genes have several biological functions, and some cells of pure cell types transition into other types over time, which suggest that their outcome-feature relationships (regression coefficients) can be a mixture of relationships in more than one subgroups, and as a result, the disjoint subgrouping results can be unsatisfactory. To this end, we develop a novel approach to regression-based heterogeneity analysis, which takes into account possible overlaps between subgroups and high data dimensions. A subgroup membership vector is introduced for each sample, which is combined with a loss function. Considering the lack of information arising from small sample sizes, an $l_2$ norm penalty is developed for each membership vector to encourage similarity in its elements. A sparse penalization is also applied for regularized estimation and feature selection. Extensive simulations demonstrate its superiority over direct competitors. The analysis of Cancer Cell Line Encyclopedia data and lung cancer data from The Cancer Genome Atlas shows that the proposed approach can identify an overlapping subgroup structure with favorable performance in prediction and stability.

stat.ME↗

arXiv2022

Identification of cancer omics commonality and difference via community fusion

The analysis of cancer omics data is a "classic" problem, however, still remains challenging. Advancing from early studies that are mostly focused on a single type of cancer, some recent studies have analyzed data on multiple "related" cancer types/subtypes, examined their commonality and difference, and led to insightful findings. In this article, we consider the analysis of multiple omics datasets, with each dataset on one type/subtype of "related" cancers. A Community Fusion (CoFu) approach is developed, which conducts marker selection and model building using a novel penalization technique, informatively accommodates the network community structure of omics measurements, and automatically identifies the commonality and difference of cancer omics markers. Simulation demonstrates its superiority over direct competitors. The analysis of TCGA lung cancer and melanoma data leads to interesting findings

stat.ME↗

arXiv2022

Robust nonparametric integrative analysis to decipher heterogeneity and commonality across subgroups using sparse boosting

In many biomedical problems, data are often heterogeneous, with samples spanning multiple patient subgroups, where different subgroups may have different disease subtypes, stages, or other medical contexts. These subgroups may be related, but they are also expected to have differences with respect to the underlying biology. The heterogeneous data presents a precious opportunity to explore the heterogeneities and commonalities between related subgroups. Unfortunately, effective statistical analysis methods are still lacking. Recently, several novel methods based on integrative analysis have been proposed to tackle this challenging problem. Despite promising results, the existing studies are still limited by ignoring data contamination and making strict assumptions of linear effects of covariates on response. As such, we develop a robust nonparametric integrative analysis approach to identify heterogeneity and commonality, as well as select important covariates and estimate covariate effects. Possible data contamination is accommodated by adopting the Cauchy loss function, and a nonparametric model is built to accommodate nonlinear effects. The proposed approach is based on a sparse boosting technique. The advantages of the proposed approach are demonstrated in extensive simulations. The analysis of The Cancer Genome Atlas data on glioblastoma multiforme and lung adenocarcinoma shows that the proposed approach makes biologically meaningful findings with satisfactory prediction.

stat.ME↗

arXiv2022

Robust structured heterogeneity analysis approach for high-dimensional data

Revealing relationships between genes and disease phenotypes is a critical problem in biomedical studies. This problem has been challenged by the heterogeneity of diseases. Patients of a perceived same disease may form multiple subgroups, and different subgroups have distinct sets of important genes. It is hence imperative to discover the latent subgroups and reveal the subgroup-specific important genes. Some heterogeneity analysis methods have been proposed in recent literature. Despite considerable successes, most of the existing studies are still limited as they cannot accommodate data contamination and ignore the interconnections among genes. Aiming at these shortages, we develop a robust structured heterogeneity analysis approach to identify subgroups, select important genes as well as estimate their effects on the phenotype of interest. Possible data contamination is accommodated by employing the Huber loss function. A sparse overlapping group lasso penalty is imposed to conduct regularization estimation and gene identification, while taking into account the possibly overlapping cluster structure of genes. This approach takes an iterative strategy in the similar spirit of K-means clustering. Simulations demonstrate that the proposed approach outperforms alternatives in revealing the heterogeneity and selecting important genes for each subgroup. The analysis of Cancer Cell Line Encyclopedia data leads to biologically meaningful findings with improved prediction and grouping stability.

stat.ME↗

arXiv2022

Target observation of complex networks

How to observe the state of a network from a limited number of measurements has become an important issue in complex networks, engineering, communication, epidemiology, etc. Under some scenarios, it is neither unfeasible nor unnecessary to observe the entire network. Therefore, we investigate the target observation of a network in this paper. We propose a target minimal dominating set problem corresponding to target observation, which is a natural generalization of classical minimal dominating set problem. Three algorithms are proposed to approximate the minimum set of occupied nodes sufficient for target observation. Extensive numerical results on computer-generated random networks and real-world networks demonstrate that the proposed algorithms offer superior performance in identification of a target minimal dominating set.

physics.soc-ph↗

arXiv2022

An adaptive shortest-solution guided decimation approach to sparse high-dimensional linear regression

High-dimensional linear regression model is the most popular statistical model for high-dimensional data, but it is quite a challenging task to achieve a sparse set of regression coefficients. In this paper, we propose a simple heuristic algorithm to construct sparse high-dimensional linear regression models, which is adapted from the shortest solution-guided decimation algorithm and is referred to as ASSD. This algorithm constructs the support of regression coefficients under the guidance of the least-squares solution of the recursively decimated linear equations, and it applies an early-stopping criterion and a second-stage thresholding procedure to refine this support. Our extensive numerical results demonstrate that ASSD outperforms LASSO, vector approximate message passing, and two other representative greedy algorithms in solution accuracy and robustness. ASSD is especially suitable for linear regression problems with highly correlated measurement matrices encountered in real-world applications.

cs.LG↗

arXiv2022

An integrative sparse boosting analysis of cancer genomic commonality and difference

In cancer research, high-throughput profiling has been extensively conducted. In recent studies, the integrative analysis of data on multiple cancer patient groups/subgroups has been conducted. Such analysis has the potential to reveal the genomic commonality as well as difference across groups/subgroups. However, in the existing literature, methods with a special attention to the genomic commonality and difference are very limited. In this study, a novel estimation and marker selection method based on the sparse boosting technique is developed to address the commonality/difference problem. In terms of technical innovation, a new penalty and computation of increments are introduced. The proposed method can also effectively accommodate the grouping structure of covariates. Simulation shows that it can outperform direct competitors under a wide spectrum of settings. The analysis of two TCGA (The Cancer Genome Atlas) datasets is conducted, showing that the proposed analysis can identify markers with important biological implications and have satisfactory prediction and stability.

stat.ME↗

arXiv2022

Semiparametric integrative interaction analysis for non-small-cell lung cancer

In the genomic analysis, it is significant while challenging to identify markers associated with cancer outcomes or phenotypes. Based on the biological mechanisms of cancers and the characteristics of datasets as well, this paper proposes a novel integrative interaction approach under the semiparametric model, in which the genetic factors and environmental factors are included as the parametric and nonparametric components, respectively. The goal of this approach is to identify the genetic factors and gene-gene interactions associated with cancer outcomes, and meanwhile, estimate the nonlinear effects of environmental factors. The proposed approach is based on the threshold gradient directed regularization (TGDR) technique. Simulation studies indicate that the proposed approach outperforms in the identification of main effects and interactions, and has favorable estimation and prediction accuracy compared with the alternative methods. The analysis of non-small-cell lung carcinomas (NSCLC) datasets from The Cancer Genome Atlas (TCGA) are conducted, showing that the proposed approach can identify markers with important implications and have favorable performance in prediction accuracy, identification stability, and computation cost.

stat.ME↗

arXiv2022

Generalizable Re-Identification from Videos with Cycle Association

In this paper, we are interested in learning a generalizable person re-identification (re-ID) representation from unlabeled videos. Compared with 1) the popular unsupervised re-ID setting where the training and test sets are typically under the same domain, and 2) the popular domain generalization (DG) re-ID setting where the training samples are labeled, our novel scenario combines their key challenges: the training samples are unlabeled, and collected form various domains which do no align with the test domain. In other words, we aim to learn a representation in an unsupervised manner and directly use the learned representation for re-ID in novel domains. To fulfill this goal, we make two main contributions: First, we propose Cycle Association (CycAs), a scalable self-supervised learning method for re-ID with low training complexity; and second, we construct a large-scale unlabeled re-ID dataset named LMP-video, tailored for the proposed method. Specifically, CycAs learns re-ID features by enforcing cycle consistency of instance association between temporally successive video frame pairs, and the training cost is merely linear to the data size, making large-scale training possible. On the other hand, the LMP-video dataset is extremely large, containing 50 million unlabeled person images cropped from over 10K Youtube videos, therefore is sufficient to serve as fertile soil for self-supervised learning. Trained on LMP-video, we show that CycAs learns good generalization towards novel domains. The achieved results sometimes even outperform supervised domain generalizable models. Remarkably, CycAs achieves 82.2% Rank-1 on Market-1501 and 49.0% Rank-1 on MSMT17 with zero human annotation, surpassing state-of-the-art supervised DG re-ID methods. Moreover, we also demonstrate the superiority of CycAs under the canonical unsupervised re-ID and the pretrain-and-finetune scenarios.

cs.CV↗

arXiv2022

Far Ultra-Violet Insights Into NGC 1399's Globular Cluster Population

We investigate archival Hubble Space Telescope ACS/SBC F140LP observations of NGC~1399 to search for evidence of multiple stellar populations in extragalactic globular clusters. Enhanced FUV populations are thought to be indicators of He-enhanced second generation populations in globular clusters, specifically extreme/blue horizontal branch stars. Out of 149 globular clusters in the field of view, 58 have far ultraviolet (FUV) counterparts with magnitudes brighter than 28.5. Six of these FUV-deteced globular clusters are also detected in X-rays, including one ultraluminous X-ray source ($L_X > 10^{39}$ erg/s). While optically bright clusters corresponded to brighter FUV counterparts, we observe FUV emission from both metal-rich and metal-poor clusters, which implies that the FUV excess is not dependent on optical colour. We also find no evidence that the cluster size influences the FUV emission. The clusters with X-ray emission are not unusually FUV bright, which suggests that even the ultraluminous X-ray source does not provide significant FUV contributions. NGC 1399 is only the fourth galaxy to have its globular cluster system probed for evidence of FUV-enhanced populations, and we compare these clusters to previous studies of the Milky Way, M31, M87, and the brightest cluster in M81. These sources indicate that many globular clusters likely host extreme HB stars and/or second generation stars, and highlight the need for more complete FUV observations of extragalactic globular cluster systems.

astro-ph.GA↗

arXiv2022

A Multistep Frank-Wolfe Method

The Frank-Wolfe algorithm has regained much interest in its use in structurally constrained machine learning applications. However, one major limitation of the Frank-Wolfe algorithm is the slow local convergence property due to the zig-zagging behavior. We observe the zig-zagging phenomenon in the Frank-Wolfe method as an artifact of discretization, and propose multistep Frank-Wolfe variants where the truncation errors decay as $O(Δ^p)$, where $p$ is the method's order. This strategy "stabilizes" the method, and allows tools like line search and momentum to have more benefits. However, our results suggest that the worst case convergence rate of Runge-Kutta-type discretization schemes cannot improve upon that of the vanilla Frank-Wolfe method for a rate depending on $k$. Still, we believe that this analysis adds to the growing knowledge of flow analysis for optimization methods, and is a cautionary tale on the ultimate usefulness of multistep methods.

math.OC↗

arXiv2022

Feature-Proxy Transformer for Few-Shot Segmentation

Few-shot segmentation (FSS) aims at performing semantic segmentation on novel classes given a few annotated support samples. With a rethink of recent advances, we find that the current FSS framework has deviated far from the supervised segmentation framework: Given the deep features, FSS methods typically use an intricate decoder to perform sophisticated pixel-wise matching, while the supervised segmentation methods use a simple linear classification head. Due to the intricacy of the decoder and its matching pipeline, it is not easy to follow such an FSS framework. This paper revives the straightforward framework of "feature extractor $+$ linear classification head" and proposes a novel Feature-Proxy Transformer (FPTrans) method, in which the "proxy" is the vector representing a semantic class in the linear classification head. FPTrans has two keypoints for learning discriminative features and representative proxies: 1) To better utilize the limited support samples, the feature extractor makes the query interact with the support features from the bottom to top layers using a novel prompting strategy. 2) FPTrans uses multiple local background proxies (instead of a single one) because the background is not homogeneous and may contain some novel foreground regions. These two keypoints are easily integrated into the vision transformer backbone with the prompting mechanism in the transformer. Given the learned features and proxies, FPTrans directly compares their cosine similarity for segmentation. Although the framework is straightforward, we show that FPTrans achieves competitive FSS accuracy on par with state-of-the-art decoder-based methods.

cs.CV↗

arXiv2022

Visualization Design Practices in a Crisis: Behind the Scenes with COVID-19 Dashboard Creators

During the COVID-19 pandemic, a number of data visualizations were created to inform the public about the rapidly evolving crisis. Data dashboards, a form of information dissemination used during the pandemic, have facilitated this process by visualizing statistics regarding the number of COVID-19 cases over time. In this research, we conducted a qualitative interview study among dashboard creators from federal agencies, state health departments, mainstream news media outlets, and other organizations that created (often widely-used) COVID-19 dashboards to answer the following questions: how did visualization creators engage in COVID-19 dashboard design, and what tensions, conflicts, and challenges arose during this process? Our findings detail the trajectory of design practices -- from creation to expansion, maintenance, and termination -- that are shaped by the complex interplay between design goals, tools and technologies, labor, emerging crisis contexts, and public engagement. We particularly examined the tensions between designers and the general public involved in these processes. These conflicts, which often materialized due to a divergence between public demands and standing policies, centered around the type and amount of information to be visualized, how public perceptions shape and are shaped by visualization design, and the strategies utilized to deal with (potential) misinterpretations and misuse of visualizations. Our findings and lessons learned shed light on new ways of thinking in visualization design, focusing on the bundled activities that are invariably involved in human and nonhuman participation throughout the entire trajectory of design practice.

cs.HC↗

arXiv2022

V$^2$L: Leveraging Vision and Vision-language Models into Large-scale Product Retrieval

Product retrieval is of great importance in the ecommerce domain. This paper introduces our 1st-place solution in eBay eProduct Visual Search Challenge (FGVC9), which is featured for an ensemble of about 20 models from vision models and vision-language models. While model ensemble is common, we show that combining the vision models and vision-language models brings particular benefits from their complementarity and is a key factor to our superiority. Specifically, for the vision models, we use a two-stage training pipeline which first learns from the coarse labels provided in the training set and then conducts fine-grained self-supervised training, yielding a coarse-to-fine metric learning manner. For the vision-language models, we use the textual description of the training image as the supervision signals for fine-tuning the image-encoder (feature extractor). With these designs, our solution achieves 0.7623 MAR@10, ranking the first place among all the competitors. The code is available at: \href{https://github.com/WangWenhao0716/V2L}{V$^2$L}.

cs.CV↗

arXiv2022

UFO: Unified Feature Optimization

This paper proposes a novel Unified Feature Optimization (UFO) paradigm for training and deploying deep models under real-world and large-scale scenarios, which requires a collection of multiple AI functions. UFO aims to benefit each single task with a large-scale pretraining on all tasks. Compared with the well known foundation model, UFO has two different points of emphasis, i.e., relatively smaller model size and NO adaptation cost: 1) UFO squeezes a wide range of tasks into a moderate-sized unified model in a multi-task learning manner and further trims the model size when transferred to down-stream tasks. 2) UFO does not emphasize transfer to novel tasks. Instead, it aims to make the trimmed model dedicated for one or more already-seen task. With these two characteristics, UFO provides great convenience for flexible deployment, while maintaining the benefits of large-scale pretraining. A key merit of UFO is that the trimming process not only reduces the model size and inference consumption, but also even improves the accuracy on certain tasks. Specifically, UFO considers the multi-task training and brings two-fold impact on the unified model: some closely related tasks have mutual benefits, while some tasks have conflicts against each other. UFO manages to reduce the conflicts and to preserve the mutual benefits through a novel Network Architecture Search (NAS) method. Experiments on a wide range of deep representation learning tasks (i.e., face recognition, person re-identification, vehicle re-identification and product retrieval) show that the model trimmed from UFO achieves higher accuracy than its single-task-trained counterpart and yet has smaller model size, validating the concept of UFO. Besides, UFO also supported the release of 17 billion parameters computer vision (CV) foundation model which is the largest CV model in the industry.

cs.CV↗

arXiv2022

Approximate Frank-Wolfe Algorithms over Graph-structured Support Sets

In this paper, we consider approximate Frank-Wolfe (FW) algorithms to solve convex optimization problems over graph-structured support sets where the linear minimization oracle (LMO) cannot be efficiently obtained in general. We first demonstrate that two popular approximation assumptions (additive and multiplicative gap errors) are not applicable in that no cheap gap-approximate LMO oracle exists. Thus, approximate dual maximization oracles (DMO) are proposed, which approximate the inner product rather than the gap. We prove that the standard FW method using a $δ$-approximate DMO converges as $\mathcal{O}((1-δ) \sqrt{s}/δ)$ in the worst case, and as $\mathcal{O}(L/(δ^2 t))$ over a $δ$-relaxation of the constraint set. Furthermore, when the solution is on the boundary, a variant of FW converges as $\mathcal{O}(1/t^2)$ under the quadratic growth assumption. Our empirical results suggest that even these improved bounds are pessimistic, showing fast convergence in recovering real-world images with graph-structured sparsity.

math.OC↗

arXiv2022

Multimode optical fiber beam-by-beam cleanup

We introduce and experimentally demonstrate the concept of all-optical beam switching in graded-index multimode optical fibers. Nonlinear coupling between orthogonally polarized seed and signal beams permits to control the spatial beam quality at the fiber output. Remarkably, we show that even a weak few-mode control beam may substantially enhance the quality of an intense, highly multimode signal beam. We propose a simple geometrical representation of the beam switching operation, whose validity is quantitatively confirmed by the experimental mode decomposition of the output beam. All-optical switching of multimode beams may find important applications in high-power beam delivery and fiber lasers.

physics.optics↗