Neural decoding, which aims to predict external visual stimuli information from evoked brain activities, plays an important role in understanding human system. Many existing methods are based on linear models, and most of them only focus either the activity pattern classification or identification. Accurate reconstruction perceived images measured activities still remains challenging. In this paper, we propose a novel deep generative multiview model for accurate image by functional magnetic...

Spatial item recommendation has become an important means to help people discover interesting locations, especially when pay a visit unfamiliar regions. Some current researches are focusing on modelling individual and collective geographical preferences for spatial based users' check-in records, but they fail explore the phenomenon of user interest drift across regions, i.e., users would show different interests travel Besides, ignore influence public comments subsequent behaviors....

There are threefold challenges in emotion recognition. First, it is difficult to recognize human's emotional states only considering a single modality. Second, expensive manually annotate the data. Third, data often suffers from missing modalities due unforeseeable sensor malfunction or configuration issues. In this paper, we address all these problems under novel multi-view deep generative framework. Specifically, propose model statistical relationships of multi-modality using multiple...

Transfer learning focuses on the scenarios when test data from target domains and training source are drawn similar but different distributions with respect to raw features. Along this line, some recent studies revealed that high-level concepts, such as word clusters, could help model differences of distributions, thus more appropriate for classification. In other words, these methods assume all have same set shared which used bridge knowledge transfer. However, in addition each domain may...

Multi-task multi-view learning deals with the scenarios where multiple tasks are associated each other through shared feature views. All previous works for this problem assume that use same set of class labels. However, in real world there exist quite a few applications several views correspond to different This new scenario is called Multi-view Learning Heterogeneous Tasks study. Then, we propose Multi-tAsk MUlti-view Discriminant Analysis (MAMUDA) method solve problem. Specifically,...

Decoding human brain activities via functional magnetic resonance imaging (fMRI) has gained increasing attention in recent years. While encouraging results have been reported states classification tasks, reconstructing the details of visual experience still remains difficult. Two main challenges that hinder development effective models are perplexing fMRI measurement noise and high dimensionality limited data instances. Existing methods generally suffer from one or both these issues yield...

Transfer learning focuses on the scenarios when test data from target domains and training source are drawn similar but different distribution with respect to raw features. Some recent studies argued that high-level concepts (e.g. word clusters) can help model difference, thus more appropriate for classification. Specifically, these methods assume all have same set of shared concepts, which used as bridge knowledge transfer. However, besides each domain may its own distinct concepts. To...

The reconstruction of visual information from human brain activity is a very important research topic in decoding. Existing methods ignore the structural underlying activities and features, which severely limits their performance interpretability. Here, we propose hierarchically structured neural decoding framework by using multitask transfer learning deep network (DNN) representations matrix-variate Gaussian prior. Our consists two stages, Voxel2Unit Unit2Pixel. In Voxel2Unit, decode...

Many important data mining problems can be modeled as learning a (bidirectional) multidimensional mapping between two domains. Based on the generative adversarial networks (GANs), particularly conditional ones, cross-domain joint distribution matching is an increasingly popular kind of methods addressing such problems. Though significant advances have been achieved, there are still main disadvantages existing models, i.e., requirement large amount paired training samples and notorious...

Stock comments from analysts contain important consulting information for investors to foresee stock volatility and market trends. Existing studies on usually focused capturing coarse-grained opinion polarities or understanding fundamentals. However, are often overwhelmed confused by massive with huge noises ambiguous opinions. Therefore, it is an emerging need have a fine-grained comment analysis tool identify more reliable comments. To this end, paper provides solution called...

Multi-task Learning (MTL) aims to learn multiple related tasks simultaneously instead of separately improve generalization performance each task. Most existing MTL methods assumed that the be learned have same feature representation. However, this assumption may not hold for many real-world applications. In paper, we study problem with heterogeneous features To address problem, first construct an integrated graph a set bipartite graphs build connection among different tasks. We then propose...

Decoding human brain activities via functional magnetic resonance imaging (fMRI) has gained increasing attention in recent years. While encouraging results have been reported states classification tasks, reconstructing the details of visual experience still remains difficult. Two main challenges that hinder development effective models are perplexing fMRI measurement noise and high dimensionality limited data instances. Existing methods generally suffer from one or both these issues yield...

Supervised dimensionality reduction has shown great advantages in finding predictive subspaces. Previous methods rarely consider the popular maximum margin principle and are prone to overfitting usually small training data, especially for those under likelihood framework. In this paper, we present a posterior-regularized Bayesian approach combine Principal Component Analysis (PCA) with max-margin learning. Based on data augmentation idea learning probabilistic interpretation of PCA, our...

Decoding visual contents from human brain activity is a challenging task with great scientific value. Two main facts that hinder existing methods producing satisfactory results are 1) typically small paired training data; 2) under-exploitation of the structural information underlying data. In this paper, we present novel conditional deep generative neural decoding approach structured intermediate feature prediction. Specifically, our first decodes to multilayer features pretrained...

Multi-task learning has proven to be useful boost the of multiple related but different tasks. Meanwhile, latent semantic models such as LSA and LDA are popular effective methods extract discriminative features high dimensional dyadic data. In this paper, we present a method combine these two techniques together by introducing new matrix tri-factorization based formulation for semi-supervised learning, which can incorporate labeled information into traditional unsupervised semantics. Our...

Spatial item recommendation has become an important means to help people discover interesting locations, especially when pay a visit unfamiliar regions. Some current researches are focusing on modelling individual and collective geographical preferences for spatial based users' check-in records, but they fail explore the phenomenon of user interest drift across regions, i.e., users would show different interests travel Besides, ignore influence public comments subsequent behaviors....

Predicting tags for a given item and leveraging to assist recommendation are two popular research topics in the field of recommender system. Previous studies mostly focus only one them make contributions. However, we believe that these tasks inherently correlated with each other: can provide additional information profile items more accurate recommendation; user behaviors help infer relationships benefit tagging process. In order take advantages such mutually influential signals, propose...

Learning over incomplete multi-modality data is a challenging problem with strong practical applications. Most existing multi-modal imputation approaches have two limitations: (1) they are unable to accurately control the semantics of imputed modalities; and (2) without shared low-dimensional latent space, do not scale well multiple modalities. To overcome limitations, we propose novel doubly semi-supervised learning framework (DSML) modality-shared space modality-specific generators,...

The enlarging volumes of data resources produced in real world makes classification very large scale a challenging task. Therefore, parallel process high dimensional is important. Hyper-Surface Classification (HSC) approved to be an effective and efficient algorithm handle two three data. Though HSC can extended deal with dimension reduction or ensemble techniques, it not trivial tackle directly. Inspired by the decision tree idea, improvement proposed directly this work. Furthermore, we...