Automatic feature selection in the markov random field model for information retrieval

ESVM: Evolutionary support vector machine for automatic feature selection and classification of microarray data

A common limitation of many information retrieval (IR) models is that relevance scores are solely based on exact (i.e., syntactic) matching of words in queries and documents under the simple Bag-of-Words (BoW) representation. This not only leads to the well-known vocabulary mismatch problem, but also does not allow semantically related words to contribute to the relevance score. Recent advances in word embedding have shown that semantic representations for words can be efficiently learned by distributional models. A natural generalization is then to represent both queries and documents as Bag-of-Word-Embeddings (BoWE), which provides a better foundation for semantic matching than BoW. Based on this representation, we introduce a novel retrieval model by viewing the matching between queries and documents as a non-linear word transportation (NWT) problem. With this formulation, we define the capacity and profit of a transportation model designed for the IR task. We show that this transportation problem can be efficiently solved via pruning and indexing strategies. Experimental results on several representative benchmark datasets show that our model can outperform many state-of-the-art retrieval models as well as recently introduced word embedding-based models. We also conducted extensive experiments to analyze the effect of different settings on our semantic matching model.

The exploding number of computational models produced by Systems Biologists over the last years is an invitation to structure and exploit this new wealth of information. Researchers would like to trace models relevant to specific scientific questions, to explore their biological content, to align and combine them, and to match them with experimental data. To automate these processes, it is essential to consider semantic annotations, which describe their biological meaning. As a prerequisite for a wide range of computational methods, we propose general and flexible similarity measures for Systems Biology models computed from semantic annotations. By using these measures and a large extensible ontology, we implement a platform that can retrieve, cluster, and align Systems Biology models and experimental data sets. At present, its major application is the search for relevant models in the BioModels Database, starting from initial models, data sets, or lists of biological concepts. Beyond similarity searches, the representation of models by semantic feature vectors may pave the way for visualisation, exploration, and statistical analysis of large collections of models and corresponding data.

Data-driven models were developed for monitoring the power consumption and wood quality during the lumber manufacturing process. The study proposes hybrid models using vibration signals combined with self-organizing maps (SOMs) for cutting power and waviness prediction in the circular sawing process of Douglas-fir wood under very high feed speed conditions. The acquired vibration signals were fed into SOMs with different topologies for data mapping and automatic sensory feature selection, which were fed into the adaptive neuro-fuzzy inference system (ANFIS) or multilayer perceptron (MLP) neural network (NN) for cutting power and waviness prediction. The monitoring performance of the hybrid SOM-MLP NN and SOM-ANFIS models was compared. The frequency response of vibration signals and its correlation with the cutting parameters as well as the cutting power and waviness were discussed. The study shows that the developed SOM models to select the optimal features from the vibration signals could accurately predict cutting power and waviness when combined with a machine learning model. The prediction performance is highly dependent on the optimal choice of SOM topology. Having a poor choice of SOM topology, fine-tuning the architecture of ANFIS and MLP NN is crucial and greatly impacts the monitoring performance. Based on the obtained results, SOM is recommended for the automatic feature in the machining or tool condition monitoring, where due to process high variability, manual feature selection is a challenging task. SOM combined with an ANN or ANFIS makes a powerful intelligent model for monitoring complex processes such as wood circular sawing.

A novel unsupervised corpus-based stemming technique using lexicon and corpus statistics

The increasing presence of hate speech on social media has drawn significant investment from governments, companies, and empirical research. Existing methods typically use a supervised text classification approach that depends on carefully engineered features. However, it is unclear if these features contribute equally to the performance of such methods. We conduct a feature selection analysis in such a task using Twitter as a case study, and show findings that challenge conventional perception of the importance of manual feature engineering: automatic feature selection can drastically reduce the carefully engineered features by over 90% and selects predominantly generic features often used by many other language related tasks; nevertheless, the resulting models perform better using automatically selected features than carefully crafted task-specific features.

Feature selection is the process of finding the set of inputs to a machine learning algorithm that will yield the best performance. Developing a way to solve this problem automatically would make current machine learning methods much more useful. Previous efforts to automate feature selection rely on expensive meta-learning or are applicable only when labeled training data is available. This paper presents a novel method called FS-NEAT which extends the NEAT neuroevolution method to automatically determine an appropriate set of inputs for the networks it evolves. By learning the network's inputs, topology, and weights simultaneously, FS-NEAT addresses the feature selection problem without relying on meta-learning or labeled data. Initial experiments in an autonomous car racing simulation demonstrate that FS-NEAT can learn better and faster than regular NEAT. In addition, the networks it evolves are smaller and require fewer inputs. Furthermore, FS-NEAT's performance remains robust even as the feature selection task it faces is made increasingly difficult.

This thesis is about the generation of playlists for music. The distances between songs are calculated based on spectral shape features. Minimal user feedback (approval or rejection of proposed songs) is used besides song distances to generate playlists. Automatic feature selection is used to better fit the user’s needs. There are 4 different contributions to the current state of art: • The comparison of different distance measures and statistical models in the context of playlist generation. • A new way of using spectral shape features, extracting them in multiple dimensions and modelling them independently from each other instead of putting all spectral features together and just estimating one probability density function. • Introduction of additional song selection strategies for playlist generation. • Adaptive automatic feature selection in the context of playlist generation. The results of research work indicate that using a likelihood ratio hypothesis test distance using single Gaussians returns results comparable to the results of the more complicated distances operating on Gaussian mixture models. Furthermore, results show that more-sophisticated selection strategies are capable of handling feature spaces including less important features. The quality of the playlists can be further improved by means of user-adaptive automatic feature selection.

Regression is a fundamental problem in statistical data analysis, which aims at estimating the conditional mean of output given input. However, regression is not informative enough if the conditional probability density is multi-modal, asymmetric, and heteroscedastic. To overcome this limitation, various estimators of conditional densities themselves have been developed, and a kernel-based approach called least-squares conditional density estimation (LS-CDE) was demonstrated to be promising. However, LS-CDE still suffers from large estimation error if input contains many irrelevant features. In this paper, we therefore propose an extension of LS-CDE called sparse additive CDE (SA-CDE), which allows automatic feature selection in CDE. SA-CDE applies kernel LS-CDE to each input feature in an additive manner and penalizes the whole solution by a group-sparse regularizer. We also give a subgradient-based optimization method for SA-CDE training that scales well to high-dimensional large data sets. Through experiments with benchmark and humanoid robot transition datasets, we demonstrate the usefulness of SA-CDE in noisy CDE problems.

BackgroundUsing DNA microarrays, we have developed two novel models for tumor classification and target gene prediction. First, gene expression profiles are summarized by optimally selected Self-Organizing Maps (SOMs), followed by tumor sample classification by Fuzzy C-means clustering. Then, the prediction of marker genes is accomplished by either manual feature selection (visualizing the weighted/mean SOM component plane) or automatic feature selection (by pair-wise Fisher's linear discriminant).ResultsThe proposed models were tested on four published datasets: (1) Leukemia (2) Colon cancer (3) Brain tumors and (4) NCI cancer cell lines. The models gave class prediction with markedly reduced error rates compared to other class prediction approaches, and the importance of feature selection on microarray data analysis was also emphasized.ConclusionsOur models identify marker genes with predictive potential, often better than other available methods in the literature. The models are potentially useful for medical diagnostics and may reveal some insights into cancer classification. Additionally, we illustrated two limitations in tumor classification from microarray data related to the biology underlying the data, in terms of (1) the class size of data, and (2) the internal structure of classes. These limitations are not specific for the classification models used.

Data anomaly detection with automatic feature selection and deep learning

Self-adaptive fusion of local-temporal features for tool condition monitoring: A human experience free model

The application of machine learning in health care often necessitates the use of hierarchical codes such as the International Classification of Diseases (ICD) and Anatomical Therapeutic Chemical (ATC) systems. These codes classify diseases and medications, respectively, thereby forming extensive data dimensions. Unsupervised feature selection tackles the "curse of dimensionality" and helps to improve the accuracy and performance of supervised learning models by reducing the number of irrelevant or redundant features and avoiding overfitting. Techniques for unsupervised feature selection, such as filter, wrapper, and embedded methods, are implemented to select the most important features with the most intrinsic information. However, they face challenges due to the sheer volume of ICD and ATC codes and the hierarchical structures of these systems. The objective of this study was to compare several unsupervised feature selection methods for ICD and ATC code databases of patients with coronary artery disease in different aspects of performance and complexity and select the best set of features representing these patients. We compared several unsupervised feature selection methods for 2 ICD and 1 ATC code databases of 51,506 patients with coronary artery disease in Alberta, Canada. Specifically, we used the Laplacian score, unsupervised feature selection for multicluster data, autoencoder-inspired unsupervised feature selection, principal feature analysis, and concrete autoencoders with and without ICD or ATC tree weight adjustment to select the 100 best features from over 9000 ICD and 2000 ATC codes. We assessed the selected features based on their ability to reconstruct the initial feature space and predict 90-day mortality following discharge. We also compared the complexity of the selected features by mean code level in the ICD or ATC tree and the interpretability of the features in the mortality prediction task using Shapley analysis. In feature space reconstruction and mortality prediction, the concrete autoencoder-based methods outperformed other techniques. Particularly, a weight-adjusted concrete autoencoder variant demonstrated improved reconstruction accuracy and significant predictive performance enhancement, confirmed by DeLong and McNemar tests (P<.05). Concrete autoencoders preferred more general codes, and they consistently reconstructed all features accurately. Additionally, features selected by weight-adjusted concrete autoencoders yielded higher Shapley values in mortality prediction than most alternatives. This study scrutinized 5 feature selection methods in ICD and ATC code data sets in an unsupervised context. Our findings underscore the superiority of the concrete autoencoder method in selecting salient features that represent the entire data set, offering a potential asset for subsequent machine learning research. We also present a novel weight adjustment approach for the concrete autoencoders specifically tailored for ICD and ATC code data sets to enhance the generalizability and interpretability of the selected features.

The features used may have an important effect on the performance of credit scoring models. The process of choosing the best set of features for credit scoring models is usually unsystematic and dominated by somewhat arbitrary trial. This paper presents an empirical study of four machine learning feature selection methods. These methods provide an automatic data mining technique for reducing the feature space. The study illustrates how four feature selection methods—‘ReliefF’, ‘Correlation-based’, ‘Consistency-based’ and ‘Wrapper’ algorithms help to improve three aspects of the performance of scoring models: model simplicity, model speed and model accuracy. The experiments are conducted on real data sets using four classification algorithms—‘model tree (M5)’, ‘neural network (multi-layer perceptron with back-propagation)’, ‘logistic regression’, and ‘k-nearest-neighbours’.

A variety of flexible models have been proposed to detect objects in challenging real world scenes. Motivated by some of the most successful techniques, we propose a hierarchical multi-feature representation and automatically learn flexible hierarchical object models for a wide variety of object classes. To that end we not only rely on automatic selection of relevant individual features, but go beyond previous work by automatically selecting and modeling complex, long-range feature couplings within this model. To achieve this generality and flexibility our work combines structure learning in conditional random fields and discriminative parameter learning of classifiers using hierarchical features. We adopt an efficient gradient based heuristic for model selection and carry it forward to discriminative, multidimensional selection of features and their couplings for improved detection performance. Experimentally we consistently outperform the currently leading method on all 20 classes of the PASCAL VOC 2007 challenge and achieve the best published results on 16 of 20 classes.