Pathological examination is regarded as the gold standard for cancer diagnosis in clinical medicine. Gigapixel histopathological images play an important role in developing automatic pathology diagnosis systems. Because of the expensive cost of pixel-level annotation, deep multiple instance learning (MIL) has become the main approach to classify histopatho-logical images with only slide-level annotation. However, three major challenges including lack of data efficiency because MIL approaches rely on task-agnostic feature extractor, overfitting challenges caused by high data imbalance between tumor and normal tissues, and the similarity between tumor and normal patches, are to be tackled. We proposed a three-stage deep MIL approach to address these challenges. The first stage generated pseudo instance-level labels by utilizing max-max ranking loss. The second stage refined the feature extractor into a task-specific feature extractor with pseudo labels. Focal loss was used to tackle the data imbalance challenge. Partial decoder component was adopted to better distinguish tumor tissues. We re-trained the MIL network with task-specific multi-level features in the third stage. Our model was able to predict both instance-level and bag-level labels in the inference stage. Experiment results showed our model outperformed baseline models in both accuracy and AUC metrics.

Sparse-matrix vector multiplication (SpMV) is a core routine in many applications. Its performance is limited by memory bandwidth, which is for matrix transport between processors and memory, and instruction latency in computations. Vectorized operations (SIMD) can dramatically improve the execution efficiency, but irregular matrices' sparsity pattern is not compatible with the style of SIMD execution. We present a new matrix format, Compressed Sparse Column Vector (CSCV), and a corresponding vectorized SpMV algorithm for matrices arising from integral equations. This SpMV algorithm can inherently suit wide SIMD instructions and reduce the memory bandwidth used. We implement this algorithm for Computed Tomography (CT) imaging reconstructions on both Intel and AMD x86 platforms and compare it with seven state-of-the-art SpMV implementations using different CT imaging matrices. Experimental results show that CSCV can achieve up to 96.9 GFLOP/s in single-precision tests, with speedup 3.70× to MKL and 3.48× to the second place implementation. Furthermore, the implementation of CSCV SpMV is performance portable, which excludes almost all SIMD assemble code and has promising performance with compiler-assisted vectorization. Code Availability: https://github.com/sysu-compsci/cscv

Synthetic digital mammography (SDM), a 2D image generated from digital breast tomosynthesis (DBT), is used as a potential substitute for full-field digital mammography (FFDM) in clinic to reduce the radiation dose for breast cancer screening. Previous studies exploited projection geometry and fused projection data and DBT volume, with different post-processing techniques applied on re-projection data which may generate different image appearance compared to FFDM. To alleviate this issue, one possible solution to generate an SDM image is using a learning-based method to model the transformation from the DBT volume to the FFDM image using current DBT/FFDM combo images. In this study, we proposed to use a deep convolutional neural network (DCNN) to learn the transformation to generate SDM using current DBT/FFDM combo images. Gradient guided conditional generative adversarial networks (GGGAN) objective function was designed to preserve subtle MCs and the perceptual loss was exploited to improve the performance of the proposed DCNN on perceptual quality. We used various image quality criteria for evaluation, including preserving masses and MCs which are important in mammogram. Experiment results demonstrated progressive performance improvement of network using different objective functions in terms of those image quality criteria. The methodology we exploited in the SDM generation task to analyze and progressively improve image quality by designing objective functions may be helpful to other image generation tasks.

The recent success of deep graph embedding innovates the graphical information characterization methodologies. However, in real-world applications, such a method still struggles with the challenges of heterogeneity, scalability, and multiplex. To address these challenges, in this study, we propose a novel solution, Genetic hEterogeneous gRaph eMbedding (GERM), which enables flexible and efficient task-driven vertex embedding in a complex heterogeneous graph. Unlike prior efforts for this track of studies, we employ a task-oriented genetic activation strategy to efficiently generate the “Edge Type Activated Vector” (ETAV) over the edge types in the graph. The generated ETAV can not only reduce the incompatible noise and navigate the heterogeneous graph random walk at the graph-schema level, but also activate an optimized subgraph for efficient representation learning. By revealing the correlation between the graph structure and task information, the model interpretability can be enhanced as well. Meanwhile, an activated heterogeneous skip-gram framework is proposed to encapsulate both topological and task-specific information of a given heterogeneous graph. Through extensive experiments on both scholarly and e-commerce datasets, we demonstrate the efficacy and scalability of the proposed methods via various search/recommendation tasks. GERM can significantly reduces the running time and remove expert-intervention without sacrificing the performance (or even modestly improve) by comparing with baselines.

Aspect category detection is an essential task for sentiment analysis and opinion mining. However, the cost of categorical data labeling, e.g., label the review aspect information for a large number of product domains, can be inevitable but unaffordable. In this study, we propose a novel problem, cross-domain aspect category transfer and detection, which faces three challenges: various feature spaces, different data distributions, and diverse output spaces. To address these problems, we propose an innovative solution, Traceable Heterogeneous Graph Representation Learning (THGRL). Unlike prior text-based aspect detection works, THGRL explores latent domain aspect category connections via massive user behavior information on a heterogeneous graph. Moreover, an innovative latent variable "Walker Tracer" is introduced to characterize the global semantic/aspect dependencies and capture the informative vertexes on the random walk paths. By using THGRL, we project different domains' feature spaces into a common one, while allowing data distributions and output spaces stay differently. Experiment results show that the proposed method outperforms a series of state-of-the-art baseline models.

This work proposes a dynamic-shape-prior guided snake model (DSP G-snake) that is designed to improve the overall stability of the point-based snake model. The dynamic shape prior is first proposed for snakes, that efficiently unifies different types of high-level priors into a new force term. To be specific, a global-topology regularity is first introduced that settles the inherent self-intersection problem with snakes. The problem that a snake's snaxels tend to unevenly distribute along the contour is also handled, leading to good parameterization. Unlike existing methods that employ learning templates or commonly enforce hard priors, the dynamic-template scheme strongly respects the deformation flexibility of the model, while retaining a decent global topology for the snake. It is verified by experiments that the proposed algorithm can effectively prevent snakes from self-crossing, or automatically untie an already selfintersected contour. In addition, the proposed model is combined with existing forces and applied to the very challenging task of tracking dense biological cell populations. The DSP G-snake model has enabled an improvement of up to 30% in tracking accuracy with respect to regular model-based approaches. Through experiments on real cellular datasets, with highly dense populations and relatively large displacements, it is confirmed that the proposed approach has enabled superior performance, in comparison to modern active-contour competitors as well as state-of-the-art cell tracking frameworks.

While the volume of scholarly publications has increased at a frenetic pace, accessing and consuming the useful candidate papers, in very large digital libraries, is becoming an essential and challenging task for scholars. Unfortunately, because of language barrier, some scientists (especially the junior ones or graduate students who do not master other languages) cannot efficiently locate the publications hosted in a foreign language repository. In this study, we propose a novel solution, cross-language citation recommendation via Hierarchical Representation Learning on Heterogeneous Graph (HRLHG), to address this new problem. HRLHG can learn a representation function by mapping the publications, from multilingual repositories, to a low-dimensional joint embedding space from various kinds of vertexes and relations on a heterogeneous graph. By leveraging both global (task specific) plus local (task independent) information as well as a novel supervised hierarchical random walk algorithm, the proposed method can optimize the publication representations by maximizing the likelihood of locating the important cross-language neighborhoods on the graph. Experiment results show that the proposed method can not only outperform state-of-the-art baseline models, but also improve the interpretability of the representation model for cross-language citation recommendation task.