Wenting Tu

A subject transfer framework for EEG classification

This paper proposes a subject transfer framework for EEG classification. It aims to improve the classification performance when the training set of the target subject (namely user) is small owing to the need to reduce the calibration session. Our framework pursues improvement not only at the feature extraction stage, but also at the classification stage. At the feature extraction stage, we first obtain a candidate filter set for each subject through a previously proposed feature extraction method. Then, we design different criterions to learn two sparse subsets of the candidate filter set, which are called the robust filter bank and adaptive filter bank, respectively. Given robust and adaptive filter banks, at the classification step, we learn classifiers corresponding to these filter banks and employ a two-level ensemble strategy to dynamically and locally combine their outcomes to reach a single decision output. The proposed framework, as validated by experimental results, can achieve positive knowledge transfer for improving the performance of EEG classification.

View construction for multi-view semi-supervised learning

Recent developments on semi-supervised learning have witnessed the effectiveness of using multiple views, namely integrating multiple feature sets to design semi-supervised learning methods. However, the so-called multiview semi-supervised learning methods require the availability of multiple views. For many problems, there are no ready multiple views, and although the random split of the original feature sets can generate multiple views, it is definitely not the most effective approach for view construction. In this paper, we propose a feature selection approach to construct multiple views by means of genetic algorithms. Genetic algorithms are used to find promising feature subsets, two of which having maximum classification agreements are then retained as the best views constructed from the original feature set. Besides conducting experiments with single-task support vector machine (SVM) classifiers, we also apply multitask SVM classifiers to the multi-view semi-supervised learning problem. The experiments validate the effectiveness of the proposed view construction method.

Semi-supervised feature extraction for EEG classification

Two semi-supervised feature extraction methods are proposed for electroencephalogram (EEG) classification. They aim to alleviate two important limitations in brain–computer interfaces (BCIs). One is on the requirement of small training sets owing to the need of short calibration sessions. The second is the time-varying property of signals, e.g., EEG signals recorded in the training and test sessions often exhibit different discriminant features. These limitations are common in current practical applications of BCI systems and often degrade the performance of traditional feature extraction algorithms. In this paper, we propose two strategies to obtain semi-supervised feature extractors by improving a previous feature extraction method extreme energy ratio (EER). The two methods are termed semi-supervised temporally smooth EER and semi-supervised importance weighted EER, respectively. The former constructs a regularization term on the preservation of the temporal manifold of test samples and adds this as a constraint to the learning of spatial filters. The latter defines two kinds of weights by exploiting the distribution information of test samples and assigns the weights to training data points and trials to improve the estimation of covariance matrices. Both of these two methods regularize the spatial filters to make them more robust and adaptive to the test sessions. Experimental results on data sets from nine subjects with comparisons to the previous EER demonstrate their better capability for classification.

Transferable Discriminative Dimensionality Reduction

In transfer learning scenarios, previous discriminative dimensionality reduction methods tend to perform poorly owing to the difference between source and target distributions. In such cases, it is unsuitable to only consider discrimination in the low-dimensional source latent space since this would generalize badly to target domains. In this paper, we propose a new dimensionality reduction method for transfer learning scenarios, which is called transferable discriminative dimensionality reduction (TDDR). By resolving an objective function that encourages the separation of the domain-merged data and penalizes the distance between source and target distributions, we can find a low-dimensional latent space which guarantees not only the discrimination of projected samples, but also the transferability to enable later classification or regression models constructed in the source domain to generalize well to the target domain. In the experiments, we firstly analyze the perspective of transfer learning in brain-computer interface (BCI) research and then test TDDR on two real datasets from BCI applications. The experimental results show that the TDDR method can learn a low-dimensional latent feature space where the source models can perform well in the target domain.

Cross-domain representation-learning framework with combination of class-separate and domain-merge objectives

Recently, cross-domain learning has become one of the most important research directions in data mining and machine learning. In multi-domain learning, one problem is that the classification patterns and data distributions are different among domains, which leads to that the knowledge (e.g. classification hyperplane) can not be directly transferred from one domain to another. This paper proposes a framework to combine class-separate objectives (maximize separability among classes) and domain-merge objectives (minimize separability among domains) to achieve cross-domain representation learning. Three special methods called DMCS_CSF, DMCS_FDA and DMCS_PCDML upon this framework are given and the experimental results valid their effectiveness.

Activity-Partner Recommendation

In many activities, such as watching movies or having dinner, people prefer to find partners before participation. Therefore, when recommending activity items (e.g., movie tickets) to users, it makes sense to also recommend suitable activity partners. This way, (i) the users save time for finding activity partners, (ii) the effectiveness of the item recommendation is increased (users may prefer activity items more if they can find suitable activity partners), (iii) recommender systems become more interesting and enkindle users’ social enthusiasm. In this paper, we identify the usefulness of suggesting activity partners together with items in recommender systems. In addition, we propose and compare several methods for activity-partner recommendation. Our study includes experiments that test the practical value of activity-partner recommendation and evaluate the effectiveness of all suggested methods as well as some alternative strategies.

LCCT: A Semi-supervised Model for Sentiment Classification

Analyzing public opinions towards products, services and social events is an important but challenging task. An accurate sentiment analyzer should take both lexicon-level information and corpus-level information into account. It also needs to exploit the domainspecific knowledge and utilize the common knowledge shared across domains. In addition, we want the algorithm being able to deal with missing labels and learning from incomplete sentiment lexicons. This paper presents a LCCT (Lexicon-based and Corpus-based, Co-Training) model for semi-supervised sentiment classification. The proposed method combines the idea of lexicon-based learning and corpus-based learning in a unified cotraining framework. It is capable of incorporating both domain-specific and domainindependent knowledge. Extensive experiments show that it achieves very competitive classification accuracy, even with a small portion of labeled data. Comparing to state-ofthe-art sentiment classification methods, the LCCT approach exhibits significantly better performances on a variety of datasets in both English and Chinese.

Semi-supervised feature extraction with local temporal regularization for EEG classification

Extreme energy ratio (EER) is a recently proposed feature extractor to learn spatial filters for electroencephalogram (EEG) signal classification. It is theoretically equivalent and computationally superior to the common spatial patterns (CSP) method which is a widely used technique in brain-computer interfaces (BCIs). However, EER may seriously overfit on small training sets due to the presence of large noise. Moreover, it is a totally supervised method that cannot take advantage of unlabeled data. To overcome these limitations, we propose a regularization constraint utilizing local temporal information of unlabeled trails. It can encourage the temporal smoothness of source signals discovered, and thus alleviate their tendency to overfit. By combining this regularization trick with the EER method, we present a semi-supervised feature extractor termed semi-supervised extreme energy ratio (SEER). After solving two eigenvalue decomposition problems, SEER recovers latent source signals that not only have discriminative energy features but also preserve the local temporal structure of test trails. Compared to the features found by EER, the energy features of these source signals have a stronger generalization ability, as shown by the experimental results. As a nonlinear extension of SEER, we further present the kernel SEER and provide the derivation of its solutions.

Investment Recommendation using Investor Opinions in Social Media

Investor social media, such as StockTwist, are gaining increasing popularity. These sites allow users to post their investing opinions and suggestions in the form of microblogs. Given the growth of the posted data, a significant and challenging research problem is how to utilize the personal wisdom and different viewpoints in these opinions to help investment. Previous work aggregates sentiments related to stocks and generates buy or hold recommendations for stocks obtaining favorable votes while suggesting sell or short actions for stocks with negative votes. However, considering the fact that there always exist unreasonable or misleading posts, sentiment aggregation should be improved to be robust to noise. In this paper, we improve investment recommendation by modeling and using the quality of each investment opinion. To model the quality of an opinion, we use multiple categories of features generated from the author information, opinion content and the characteristics of stocks to which the opinion refers. Then, we discuss how to perform investment recommendation (including opinion recommendation and portfolio recommendation) with predicted qualities of investor opinions. Experimental results on real datasets demonstrate effectiveness of our work in recommending high-quality opinions and generating profitable investment decisions.

Personalized location recommendation by aggregating multiple recommenders in diversity

Location recommendation is an important feature of social network applications and location-based services. Most existing studies focus on developing one single method or model for all users. By analyzing data from two real location-based social networks (Foursquare and Gowalla), in this paper we reveal that the decisions of users on place visits depend on multiple factors, and different users may be affected differently by these factors. We design a location recommendation framework that combines results from various recommenders that consider different factors. Our framework estimates, for each individual user, the underlying influence of each factor to her. Based on the estimation, we aggregate suggestions from different recommenders to derive personalized recommendations. Experiments on Foursquare and Gowalla show that our proposed method outperforms the state-of-the-art methods on location recommendation.