ACML 2018 论文列表

Predicting medicines for patients with co-morbidity has long been recognized as a hard task due to complex dependencies between diseases and medicines. Efforts have been made recently to build high-order dependency between diseases and medicines by extracting knowledge from electronic health records (EHR). But current works failed to utilize additional knowledge and ignored the data skewness problem which lead to sub-optimal combination of medicines. In this paper, we formulate the medicines prediction task in multi-instance multi-label learning framework considering the multi-diagnoses as input instances and multi-medicines as output labels. We propose a knowledge-guided multi-instance multi-label networks called \mname where two types of additional knowledge are incorporated into a RNN encoder-decoder model. The utilization of structural knowledge like clinical ontology provides a way to learn better representation called tree embedding by utilizing the ancestors’ information. Contextual knowledge is a global summarization of input instances which is informative for personal prediction. Experiments are conducted on a real world clinical dataset which showed the necessity to combine both contextual and structural knowledge and the \mname performs better than baselines up to 4+% in terms of Jaccard similarity score.

Sparse representation has achieved great successes in various machine learning and image processing tasks. For image processing, typical patch-based sparse representation (PSR) models usually tend to generate undesirable visual artifacts, while group-based sparse representation (GSR) models produce over-smooth phenomena. In this paper, we propose a new sparse representation model, termed joint patch-group based sparse representation (JPG-SR). Compared with existing sparse representation models, the proposed JPG-SR provides a powerful mechanism to integrate the local sparsity and nonlocal self-similarity of images. We then apply the proposed JPG-SR model to a low-level vision problem, namely, image inpainting. To make the proposed scheme tractable and robust, an iterative algorithm based on the alternating direction method of multipliers (ADMM) framework is developed to solve the proposed JPG-SR model. Experimental results demonstrate that the proposed model is efficient and outperforms several state-of-the-art methods in both objective and perceptual quality.

Deep neural networks have a rich ability to learn complex representations and achieved remarkable results in various tasks. However, they are prone to overfitting due to the limited number of training samples; regularizing the learning process of neural networks is critical. In this paper, we propose a novel regularization method, which estimates parameters of a large convolutional neural network as implicit probabilistic distributions generated by a hypernetwork. Also, we can perform model averaging to improve the network performance. Experimental results demonstrate our regularization method outperformed the commonly-used maximum a posterior (MAP) estimation.

Gaussian Process Latent Variable Model (GPLVM) is a powerful nonlinear dimension reduction model and has been widely used in many machine learning scenarios. However, the original GPLVM and its variants do not explicitly model the correlations among the original features, leading to the underutilization of underlying information involved in the data. To compensate for this shortcoming, we propose a feature-correlation-aware GPLVM (fcaGPLVM) to simultaneously learn the latent variables and the feature correlations. The main contributions of this paper are 1) introducing a set of extra latent variables into the original GPLVM and proposing a feature-correlation-aware kernel function to explicitly model the feature-description information and infer the feature correlations; 2) defining a joint objective function and developing a stochastic optimization algorithm based on the stochastic variational inference (SVI) to learn all the latent variables. To the best of our knowledge, this is the first work that explicitly considers the feature correlations in the GPLVM and makes many existing GPLVMs become its special cases. Furthermore, it can be applied to both unsupervised and supervised learnings to improve the performance of dimension reduction. Experimental results show that in these two learning scenarios the proposed models outperform their state-of-the-art counterparts.

Principal component analysis (PCA) is widely used for dimension reduction and embedding of real data in social network analysis, information retrieval, and natural language processing, etc. In this work we propose a fast randomized PCA algorithm for processing large sparse data. The algorithm has similar accuracy to the basic randomized SVD (rPCA) algorithm (Halko et al., 2011), but is largely optimized for sparse data. It also has good flexibility to trade off runtime against accuracy for practical usage. Experiments on real data show that the proposed algorithm is up to 9.1X faster than the basic rPCA algorithm without accuracy loss, and is up to 20X faster than the \texttt{svds} in Matlab with little error. The algorithm computes the first 100 principal components of a large information retrieval data with 12,869,521 persons and 323,899 keywords in less than 400 seconds on a 24-core machine, while all conventional methods fail due to the out-of-memory issue.

Most of the existing tracking methods based on CNN(convolutional neural networks) are too slow for real-time application despite the excellent tracking precision compared with the traditional ones. In this paper, a fast dynamic visual tracking algorithm combining CNN based MDNet(Multi-Domain Network) and RoIAlign was developed. The major problem of MDNet also lies in the time efficiency. Considering the computational complexity of MDNet is mainly caused by the large amount of convolution operations and fine-tuning of the network during tracking, a RoIPool layer which could conduct the convolution over the whole image instead of each RoI is added to accelerate the convolution and a new strategy of fine-tuning the fully-connected layers is used to accelerate the update. With RoIPool employed, the computation speed has been increased but the tracking precision has dropped simultaneously. RoIPool could lose some positioning precision because it can not handle locations represented by floating numbers. So RoIAlign, instead of RoIPool, which can process floating numbers of locations by bilinear interpolation has been added to the network. The results show the target localization precision has been improved and it hardly increases the computational cost. These strategies can accelerate the processing and make it 7 $\times$ faster than MDNet with very low impact on precision and it can run at around 7 fps. The proposed algorithm has been evaluated on two benchmarks: OTB100 and VOT2016, on which high precision and speed have been obtained. The influence of the network structure and training data are also discussed with experiments. The source code is publicly available on github: https://github.com/ZhiyanCui/fdn-release

Convolutional Neural Networks achieve state of the art results in many image recognition tasks. While their structure makes predictions invariant to small translations, some recognition tasks require invariance to other transformations, like rotation and reflection. We apply group convolutions to build an Equivariant Autoencoder with embeddings that change predictably under the specified set of transformations. We then introduce two approaches to extracting invariant features from these embeddings—Gram Pooling and Equivariant Attention. These two methods separate transformation-relevant information from all other image features. We use obtained embeddings in classification and clustering tasks and show an improvement of the classification quality on the learned embeddings compared to pure autoencoder and average pooling method. A visualization of the learned manifold shows that objects of the same class tend to cluster together, which was not observed for the pure autoencoder embeddings.

Time series imputation (replacing missing data) plays an important role in time series analysis due to missing values in real world data. How to recover missing values and model the underlying dynamic dependencies from incomplete time series remains a challenge. A recent work has found that residual networks help build very deep networks by leveraging short paths due to skip connections (Veit et al., 2016). Inspired by this, we observe that these short paths can model underlying correlations between missing items and their previous non-missing observations in a graph-like way. Hence, we propose an end-to-end imputation network with residual short paths, called Residual IMPutation LSTM (RIMP-LSTM), a flexible combination of residual short paths with graph-based temporal dependencies. We construct a residual sum unit (RSU), which enables RIMP-LSTM to make full use of previous revealed information to model incomplete time series and reduce the negative impact of missing values. Moreover, a switch unit is designed to detect the missing values and a new loss function is then developed to train our model with time series in the presence of missing values in an end-to-end way, which also allows simultaneous imputation and prediction. Extensive empirical comparisons with other competitive imputation approaches over several synthetic and real world time series with various rates of missing data verify the superiority of our model.

Deep neural networks have shown excellent performance in stereo matching task. Recently CNN-based methods have shown that stereo matching can be formulated as a supervised learning task. However, less attention is paid on the fusion of contextual semantic information and details. To tackle this problem, we propose a network for disparity estimation based on abundant contextual details and semantic information, called Multi-scale Features Network (MSFNet). First, we design a new structure to encode rich semantic information and fine-grained details by fusing multi-scale features. And we combine the advantages of element-wise addition and concatenation, which is conducive to merge semantic information with details. Second, a guidance mechanism is introduced to guide the network to automatically focus more on the unreliable regions. Third, we formulate the consistency check as an error map, obtained by the low stage features with fine-grained details. Finally, we adopt the consistency checking between the left feature and the synthetic left feature to refine the initial disparity. Experiments on Scene Flow and KITTI 2015 benchmark demonstrated that the proposed method can achieve the state-of-the-art performance.

Many scenarios in our daily life require us to infer some ranking over items or people based on limited information. In this paper, we consider two such scenarios, one for ranking student papers in massive online open courses and one for identifying the best player (or team) in sports tournaments. For the peer grading problem, we design a mechanism with a new way of matching graders to papers. This allows us to aggregate partial rankings from graders into a global one, with an accuracy rate matching the best in previous works, but with a much simpler analysis. For the winner selection problem in sports tournaments, we cast it as the well-known dueling bandit problem and identify a new measure to minimize: the number of parallel rounds, as one normally would not like a large tournament to last too long. We provide mechanisms which can determine the optimal or an almost optimal player in a small number of parallel rounds and at the same time using a small number of competitions.

The questions in a crowdsourcing task typically exhibit varying degrees of

In the past few years, person re-identification (reID) has developed rapidly due to the success of deep convolutional neural networks. The softmax loss function is an important component for learning discriminative features. However, the classifier trained by the softmax loss is difficult to distinguish the hard samples. In this work, we introduce a new auxiliary loss function, called batch-contrastive loss, for person reID to further separate the features of different identities and pulls the features of same identity closer. Furthermore, the proposed loss function does not rely on the pairwise or triplet sampling which is commonly used in the Siamese model. We test our loss function on two large-scale person reID benchmarks, Market-1501 and DukeMTMC datasets. Under the combination of the batch-contrastive loss and the softmax loss, even only employing the generic L2-distance metric, we can achieve competitive results among the state-of-the-arts.

End-to-end optimization of multi-instance learning (MIL) using neural networks is an important problem with many applications, in which a core issue is how to design a permutation-invariant pooling function without losing much instance-level information. Inspired by the dynamic routing in recent capsule networks, we propose a novel dynamic pooling function for MIL. It is an adaptive scheme for both key instance selection and modeling the contextual information among instances in a bag. The dynamic pooling iteratively updates the instance contribution to its bag. It is permutation-invariant and can interpret instance-to-bag relationship. The proposed dynamic pooling based multi-instance neural network has been validated on many MIL tasks and outperforms other MIL methods.

We propose a 2D multi-level appearance representation of the human body in RGB images, spatially modelled using a fully-connected graphical model. The appearance model is based on a CNN body part detector, which uses shared features in a cascade architecture to simultaneously detect body parts with different levels of granularity. We use a fully-connected Conditional Random Field (CRF) as our spatial model, over which approximate inference is efficiently performed using the Mean-Field algorithm, implemented as a Recurrent Neural Network (RNN). The stronger visual support from body parts with different levels of granularity, along with the fully-connected pairwise spatial relations, which have their weights learnt by the model, improve the performance of the bottom-up part detector. We adopt an end-to-end training strategy to leverage the potential of both our appearance and spatial models, and achieve competitive results on the MPII and LSP datasets.

Deep Embedded Clustering (DEC) surpasses traditional clustering algorithms by jointly performing feature learning and cluster assignment. Although a lot of variants have emerged, they all ignore a crucial ingredient, \emph{data augmentation}, which has been widely employed in supervised deep learning models to improve the generalization. To fill this gap, in this paper, we propose the framework of Deep Embedded Clustering with Data Augmentation (DEC-DA). Specifically, we first train an autoencoder with the augmented data to construct the initial feature space. Then we constrain the embedded features with a clustering loss to further learn clustering-oriented features. The clustering loss is composed of the target (pseudo label) and the actual output of the feature learning model, where the target is computed by using clean (non-augmented) data, and the output by augmented data. This is analogous to supervised training with data augmentation and expected to facilitate unsupervised clustering too. Finally, we instantiate five DEC-DA based algorithms. Extensive experiments validate that incorporating data augmentation can improve the clustering performance by a large margin. Our DEC-DA algorithms become the new state of the art on various datasets.

Label embedding is an effective and efficient method which can jointly extract the information of all labels for better performance of multi-label classification. However, most existing embedding methods ignore information of feature space or intrinsic structure of previous label space, such that their learned latent space will not have strong predictability and discriminant ability. We propose a novel deep neural network (DNN) based model, namely Deep Correlation Structure Preserved Label Space Embedding (DCSPE). Specifically, DCSPE derives a deep latent space by performing feature-aware label space embedding with deep canonical correlation analysis (DCCA) and preserving the intrinsic structure of the previous label space with proposed deep multidimensional scaling (DMDS). Our DCSPE is achieved by integrating the DNN architectures of the two DNN based models and can learn a feature-aware structure preserved deep latent space. Furthermore, extensive experimental results on datasets with many labels demonstrate that our proposed approach is significantly better than the existing label embedding algorithms.

Highly incoherent dictionaries have broad applications in machine learning. Minimizing the mutual coherence is a common intuition to construct incoherent dictionaries in the previous methods. However, as pointed out by Tropp(2004), mutual coherence does not offer a very subtle description and Babel function, as a generalization of mutual coherence, is a more attractive alternative. However, it is much more challenging to optimize. In this work, we minimize the Babel function directly to construct incoherent dictionaries. As far as we know, this is the first work to optimize the Babel function. We propose an augmented Lagrange multiplier based algorithm to solve this nonconvex and nonsmooth problem with the convergence guarantee that every accumulation point is a KKT point. We define a new norm $\|\X\|_{\infty,max_p}$ and propose an efficient method to compute its proximal operation with $O(n^2\mbox{log}n)$ complexity, which dominates the running time of our algorithm, where $max_p$ means the sum of the largest $p$ elements and $n$ is the number of the atoms. Numerical experiments testify to the advantage of our method.

Neural conversational models are widely used in applications such as personal assistants and chat bots. These models seem to give better performance when operating on the word level. However, for fusional languages such as French, Russian, or Polish, the vocabulary size can become infeasible since most of the words have multiple of word forms. To reduce vocabulary size, we propose a new pipeline for building conversational models: first generate words in a standard (lemmatized) form and then transform them into a grammatically correct sentence. In this work, we focus on the \emph{morphological agreement} part of the pipeline, i.e., reconstructing proper word forms from lemmatized sentences. For this task, we propose a neural network architecture that outperforms character-level models while being twice faster in training and 20% faster in inference. The proposed pipeline yields better performance than character-level conversational models according to human assessor testing.

Deep convolutional neural networks provide a powerful feature learning capability for image classification. The deep image features can be utilized to deal with many image understanding tasks like image classification and object recognition. However, the robustness obtained in one dataset can be hardly reproduced in the other domain, which leads to inefficient models far from state-of-the-art. We propose a deep collaborative weight-based classification (DeepCWC) method to resolve this problem, by providing a novel option to fully take advantage of deep features in classic machine learning. It firstly performs the $l_2$-norm based collaborative representation on the original images, as well as the deep features extracted by deep CNN models. Then, two distance vectors, obtained based on the pair of linear representations, are fused together via a novel collaborative weight. This collaborative weight enables deep and classic representations to weigh each other. We observed the complementarity between two representations in a series of experiments on 10 facial and object datasets. The proposed DeepCWC produces very promising classification results, and outperforms many other benchmark methods, especially the ones claimed for Fashion-MNIST. The code is going to be published in our public repository\footnote{https://github.com/zengsn/research}.