Fisher Kernel Research Articles

Quantum Fisher kernel for mitigating the vanishing similarity issue

Quantum kernel (QK) methods exploit quantum computers to calculate QKs for the use of kernel-based learning models. Despite a potential quantum advantage of the method, the commonly used fidelity-based QK suffers from a detrimental issue, which we call the vanishing similarity issue; the exponential decay of the expectation value and the variance of the QK deteriorates implementation feasibility and trainability of the model with the increase of the number of qubits. This implies the need to design QKs alternative to the fidelity-based one. In this work, we propose a new class of QKs called the quantum Fisher kernels (QFKs) that take into account the geometric structure of the data source. We analytically and numerically demonstrate that the QFK can avoid the issue when shallow alternating layered ansatzes are used. In addition, the Fourier analysis numerically elucidates that the QFK can have the expressivity comparable to the fidelity-based QK. Moreover, we demonstrate synthetic classification tasks where QFK outperforms the fidelity-based QK in performance due to the absence of vanishing similarity. These results indicate that QFK paves the way for practical applications of quantum machine learning toward possible quantum advantages.

Large-scale Structures in COSMOS2020: Evolution of Star Formation Activity in Different Environments at 0.4 < z < 4

To study the role of environment in galaxy evolution, we reconstruct the underlying density field of galaxies based on COSMOS2020 (The Farmer catalog) and provide the density catalog for a magnitude-limited (K s < 24.5) sample of ∼210,000 galaxies at 0.4 < z < 5 within the COSMOS field. The environmental densities are calculated using a weighted kernel density estimation approach with the choice of a von Mises–Fisher kernel, an analog of the Gaussian kernel for periodic data. Additionally, we make corrections for the edge effect and masked regions in the field. We utilize physical properties extracted by LePhare to investigate the connection between star formation activity and the environmental density of galaxies in six mass-complete subsamples at different cosmic epochs within 0.4 < z < 4. Our findings confirm a strong anticorrelation between star formation rate (SFR)/specific SFR (sSFR) and environmental density out to z ∼ 1.1. At 1.1 < z < 2, there is no significant correlation between SFR/sSFR and density. At 2 < z < 4, we observe a reversal of the SFR/sSFR–density relation such that both SFR and sSFR increase by a factor of ∼10 with increasing density contrast, δ, from −0.4 to 5. This observed reversal at higher redshifts supports the scenario where an increased availability of gas supply, along with tidal interactions and a generally higher star formation efficiency in dense environments, could potentially enhance star formation activity in galaxies located in rich environments at z > 2.

Revisiting Probabilistic Latent Semantic Analysis: Extensions, Challenges and Insights

This manuscript provides a comprehensive exploration of Probabilistic latent semantic analysis (PLSA), highlighting its strengths, drawbacks, and challenges. The PLSA, originally a tool for information retrieval, provides a probabilistic sense for a table of co-occurrences as a mixture of multinomial distributions spanned over a latent class variable and adjusted with the expectation–maximization algorithm. The distributional assumptions and the iterative nature lead to a rigid model, dividing enthusiasts and detractors. Those drawbacks have led to several reformulations: the extension of the method to normal data distributions and a non-parametric formulation obtained with the help of Non-negative matrix factorization (NMF) techniques. Furthermore, the combination of theoretical studies and programming techniques alleviates the computational problem, thus making the potential of the method explicit: its relation with the Singular value decomposition (SVD), which means that PLSA can be used to satisfactorily support other techniques, such as the construction of Fisher kernels, the probabilistic interpretation of Principal component analysis (PCA), Transfer learning (TL), and the training of neural networks, among others. We also present open questions as a practical and theoretical research window.

Locating $$\gamma$$-ray sources on the celestial sphere via modal clustering

AbstractSky surveys represent the fundamental data basis for detecting and locating as yet undiscovered celestial objects. Since 2008, the Fermi LAT Collaboration has catalogued thousands of $$\gamma$$ γ -ray sources with the aim of extending our knowledge of the highly energetic physical mechanisms and processes that lie at the core of our Universe. In this article, we present a nonparametric clustering algorithm which identifies high-energy astronomical sources using the spatial information of the $$\gamma$$ γ -ray photons detected by the large area telescope onboard the Fermi spacecraft. In particular, the sources are identified using a von Mises–Fisher kernel estimate of the photon count density on the unit sphere via an adjustment of the mean-shift algorithm which accounts for the directional nature of the collected data and the need of local smoothing. This choice entails a number of desirable benefits. It allows us to bypass the difficulties inherent on the borders of any projection of the photon directions onto a 2-dimensional plane, while guaranteeing high flexibility. The smoothing parameter is chosen adaptively, by combining scientific input with optimal selection guidelines, as known from the literature. Using statistical tools from hypothesis testing and classification, we furthermore present an automatic way to skim off sound candidate sources from the $$\gamma$$ γ -ray emitting diffuse background and to quantify their significance. We calibrate and test our algorithm on simulated count maps provided by the Fermi LAT Collaboration.

Symbolic semantic design of industrial products based on Big data technology

Abstract Exploring the symbolic semantic design path of industrial products is to make industrial products more compatible with the diverse emotional needs of consumers. In this paper, starting from the sentiment analysis model, the PLSA-FSVM sentiment analysis method is constructed using a probabilistic latent potential semantic analysis method and support vector machine based on the Fisher kernel. The method’s validity is verified for comparative experiments and sentiment word frequency analysis evaluation. From the comparison experiments, the ten-fold cross-average precision and recall of PLSA-FSVM were 89.18% and 88.35%, respectively, 4.15% and 2.59% higher than PLSA-SVM. From the sentiment word frequency analysis, the percentages of sentiment words such as atmosphere, practical, and worthy are 23.08%, 22.59%, and 24.72%, respectively. This shows that the PLSA-FSVM sentiment analysis method can effectively realize the sentiment analysis of industrial product evaluation, promote the symbolic semantic design to be more in line with consumers’ emotional needs, and then realize the symbolic design of industrial products to reach the meaning with shape and enjoy with meaning.

Integration of Multivariate Beta-based Hidden Markov Models and Support Vector Machines with Medical Applications

In this paper, we propose a novel hybrid discriminative generative model by integrating a modified version of hidden Markov model (HMM), multivariate Beta-based HMM with support vector machine (SVM). We apply Fisher Kernel to define decision boundary and separate classes. In this model, we assume that HMM emission probabilities follow a Beta mixture model as generalizing the assumption of Gaussianity may not be practical in modeling real-world applications. HMM as a generative model needs less amount of data however, its accuracy is less than discriminative models such as SVM. Moreover, in some applications, data may have various feature-length. We tackle this problem with Fisher Kernel. We apply our proposed model to medical applications, lung cancer detection, colonoscopy image, and colon tissue analysis. The results indicate that our proposed model could be a promising alternative.

Hybrid Approach for Detection of Objects from Images Using Fisher Vector and PSO Based CNN

Owing to the near connection between object recognition and video processing and picture perception, a lot of research interest has been received in recent years. Standard methods of object detection are focused on manufactured technologies and slow-moving architectures. Fisher Vectors (FV) and Convolutional Neural Networks (CNN) are two picture arrangement pipelines with various qualities. While CNNs have indicated predominant exactness on various order assignments, FV classifiers are normally less exorbitant to prepare and assess. In this paper we propose a mechanism for detection of objects in image based on Fisher kernel and CNN with a PSO optimization technique. Here fisher kernel draws the global or statically features from the image object and CNN is used for local and more complex feature extraction from an image and here we use CNN with PSO to reduce the training complexity. Performance results shows that the proposed model is detect the object better than the existing models.

A Machine Learning Approach to the Classification of Acute Leukemias and Distinction From Nonneoplastic Cytopenias Using Flow Cytometry Data.

Flow cytometry (FC) is critical for the diagnosis and monitoring of hematologic malignancies. Machine learning (ML) methods rapidly classify multidimensional data and should dramatically improve the efficiency of FC data analysis. We aimed to build a model to classify acute leukemias, including acute promyelocytic leukemia (APL), and distinguish them from nonneoplastic cytopenias. We also sought to illustrate a method to identify key FC parameters that contribute to the model's performance. Using data from 531 patients who underwent evaluation for cytopenias and/or acute leukemia, we developed an ML model to rapidly distinguish among APL, acute myeloid leukemia/not APL, acute lymphoblastic leukemia, and nonneoplastic cytopenias. Unsupervised learning using gaussian mixture model and Fisher kernel methods were applied to FC listmode data, followed by supervised support vector machine classification. High accuracy (ACC, 94.2%; area under the curve [AUC], 99.5%) was achieved based on the 37-parameter FC panel. Using only 3 parameters, however, yielded similar performance (ACC, 91.7%; AUC, 98.3%) and highlighted the significant contribution of light scatter properties. Our findings underscore the potential for ML to automatically identify and prioritize FC specimens that have critical results, including APL and other acute leukemias.

Semantic Web Mining for Analyzing Retail Environment Using Word2Vec and CNN-FK

Digital Technology is becoming increasingly essential to organizations. Related knowledge is important for a company to allow optimal use of its IT services. The use of Big Data is relatively new to this field. Handling Big data is not, at this stage, a problem for large business organizations in particular; it has also become a challenge for small and medium-sized businesses. Although Semantic Web analysis is largely focused on fundamental advances that are expected to make the Semantic Web a reality, there has not been much work done to demonstrate the feasibility and effect of the Semantic Web on business issues. The infrastructure of electronic information executives and business types has provided various enhancements for companies, such as the automated process of buying and selling products. Nevertheless, undertakings are checked for the multifaceted nature of the extension required to deal with an ever-increasing number of electronic details and procedures. This paper suggests a model with a neural network design and a word representation system named Word2Vec for analyzing retail environment. Firstly, Word2vec manages the text data and shows it as a function diagram and a feature map is given to the Convolution Neural Network (CNN) that extracts the features and classifies them. The IMDB dataset, the Cornell dataset, the Amazon Products Dataset and the Twitter dataset were analyzed in the proposed model. The proposed Convolution Neural Network Fisher Kernel (CNN-FK) model is compared with the existing SVM model for analyzing retail environment in semantic web mining. The new approach has increased efficiency when compared to existing models.

Diversified Fisher kernel: encoding discrimination in Fisher features to compete deep neural models for visual classification task

Fisher kernels derived from stochastic probabilistic models such as restricted and deep Boltzmann machines have shown competitive visual classification results in comparison to widely popular deep discriminative models. This genre of Fisher kernels bridges the gap between shallow and deep learning paradigm by inducing the characteristics of deep architecture into Fisher kernel, further deployed for classification in discriminative classifiers. Despite their success, the memory and computational costs of Fisher vectors do not make them amenable for large-scale visual retrieval and classification tasks. This study introduces a novel feature selection technique inspired from the functional characteristics of neural architectures for learning discriminative feature representations to boost the performance of Fisher kernels against deep discriminative models. The proposed technique condenses the large dimensional Fisher features for kernel learning and shows improvement in its classification performance and storage cost on leading benchmark data sets. A comparison of the proposed method with other state-of-the-art feature selection techniques is made to demonstrate its performance supremacy as well as time complexity required to learn in reduced Fisher space.

Affective actions recognition in dyadic interactions based on generative and discriminative models

Purpose Dyadic interactions are significant for human life. Most body sensor networks-based research studies focus on daily actions, but few works have been done to recognize affective actions during interactions. The purpose of this paper is to analyze and recognize affective actions collected from dyadic interactions. Design/methodology/approach A framework that combines hidden Markov models (HMMs) and k-nearest neighbor (kNN) using Fisher kernel learning is presented in this paper. Furthermore, different features are considered according to the interaction situations (positive situation and negative situation). Findings Three experiments are conducted in this paper. Experimental results demonstrate that the proposed Fisher kernel learning-based framework outperforms methods using Fisher kernel-based approach, using only HMMs and kNN. Practical implications The research may help to facilitate nonverbal communication. Moreover, it is important to equip social robots and animated agents with affective communication abilities. Originality/value The presented framework may gain strengths from both generative and discriminative models. Further, different features are considered based on the interaction situations.

Deep Gaussian processes for music mood estimation and retrieval with locally aggregated acoustic Fisher vector

Due to the subjective nature of music mood, it is challenging to computationally model the affective content of the music. In this work, we propose novel features known as locally aggregated acoustic Fisher vectors based on the Fisher kernel paradigm. To preserve the temporal context, onset-detected variable-length segments of the audio songs are obtained, for which a variational Bayesian approach is used to learn the universal background Gaussian mixture model (GMM) representation of the standard acoustic features. The local Fisher vectors obtained with the soft assignment of GMM are aggregated to obtain a better performance relative to the global Fisher vector. A deep Gaussian process (DGP) regression model inspired by the deep learning architectures is proposed to learn the mapping between the proposed Fisher vector features and the mood dimensions of valence and arousal. Since the exact inference on DGP is intractable, the pseudo-data approximation is used to reduce the training complexity and the Monte Carlo sampling technique is used to solve the intractability problem during training. A detailed derivation of a 3-layer DGP is presented that can be easily generalized to an L-layer DGP. The proposed work is evaluated on the PMEmo dataset containing valence and arousal annotations of Western popular music and achieves an improvement in $$R^2$$ of $$25\%$$ for arousal and $$52\%$$ for valence for music mood estimation and an improvement in the Gamma statistic of $$68\%$$ for music mood retrieval relative to the baseline single-layer Gaussian process.

Application of Support Vector Machine (SVM) in the Sentiment Analysis of Twitter DataSet

At present, in the mainstream sentiment analysis methods represented by the Support Vector Machine, the vocabulary and the latent semantic information involved in the text are not well considered, and sentiment analysis of text is dependent overly on the statistics of sentiment words. Thus, a Fisher kernel function based on Probabilistic Latent Semantic Analysis is proposed in this paper for sentiment analysis by Support Vector Machine. The Fisher kernel function based on the model is derived from the Probabilistic Latent Semantic Analysis model. By means of this method, latent semantic information involving the probability characteristics can be used as the classification characteristics, along with the improvement of the effect of classification for support vector machine, and the problem of ignoring the latent semantic characteristics in text sentiment analysis can be addressed. The results show that the effect of the method proposed in this paper, compared with the comparison method, is obviously improved.

Spatio-temporal modeling of global ozone data using convolution

Large spatial data are becoming more and more popular in environmental science and other related fields. Observations are often made over a substantial fraction of the surface of the Earth over a long period of time. It is necessary to model spatio-temporal random processes on the sphere which is challenging both conceptually and computationally. Convolution modeling method can be utilized to generate a random field with valid covariance structure on spheres. A latent dynamic process is defined on a grid covering the globe. The data vector is first projected onto the low-dimensional space spanned by those grids at each available time point. The resulting time series are fitted with seasonal ARIMA models. Forecasting is made by convolving the latent dynamic processes at all grid points using von Mises–Fisher kernel function. The procedure is illustrated by the total ozone data collected by Total Ozone Mapping Spectrometer during a 12-year period of time.

Bayesian network Fisher kernel for categorical feature spaces

We address the problem of defining similarity between vectors of possibly dependent categorical variables by deriving formulas for the Fisher kernel for Bayesian networks. While both Bayesian networks and Fisher kernels are established techniques, this result does not seem to appear in the literature. Such a kernel naturally opens up the possibility to conduct kernel-based analyses in completely categorical feature spaces with dependent features. We show experimentally how this kernel can be used to find subsets of observations that we see as representative for the underlying Bayesian network model.

Discriminative Viewer Identification using Generative Models of Eye Gaze

Abstract We study the problem of identifying viewers of arbitrary images based on their eye gaze. Psychological research has derived generative stochastic models of eye movements. In order to exploit this background knowledge within a discriminatively trained classification model, we derive Fisher kernels from different generative models of eye gaze. Experimentally, we find that the performance of the classifier strongly depends on the underlying generative model. Using an SVM with Fisher kernel improves the classification performance over the underlying generative model.

Optimal Reduction of Dirac Mixture Densities on the 2-Sphere

This paper is concerned with optimal approximation of a given Dirac mixture density on the S2 manifold, i.e., a set of weighted samples located on the unit sphere, by an equally weighted Dirac mixture with a reduced number of components. The sample locations of the approximating density are calculated by minimizing a smooth global distance measure, a generalization of the well-known Cramér-von Mises Distance. First, the Localized Cumulative Distribution (LCD) together with the von Mises–Fisher kernel provides a continuous characterization of Dirac mixtures on the S2 manifold. Second, the L2 norm of the difference of two LCDs is a unique and symmetric distance between the corresponding Dirac mixtures. Thereby we integrate over all possible kernel sizes instead of choosing one specific kernel size. The resulting approximation method facilitates various efficient nonlinear sample-based state estimation methods.

Attention boosted bilinear pooling for remote sensing image retrieval

ABSTRACTRemote sensing image retrieval is to find the most identical or similar images to a query image in the vast archive of remote sensing images. A key process is to extract the most distinctive features. In this study, we introduce a second-order pooling named compact bilinear pooling (CBP) into convolutional neural networks (CNNs) for remote sensing image retrieval. The retrieval algorithm has three stages, pretraining, fine-tuning and retrieval. In the pretraining stage, two classic CNN structures, VGG16 and ResNet34, are pretrained respectively with the ImageNet consisting of close-range images. A CBP layer is introduced before the fully connected layers in the two networks. To extract globally consistent representations, a channel and spatial integrated attention mechanism is proposed to refine features from the last convolution layer and the features are used as the input of the CBP. In the fine-tuning stage, the new network is fine-tuned on a remote sensing dataset to train discriminable features. In the retrieval stage, the network, with fully connected layers being replaced by a PCA (principal component analysis) module, is applied to new remote sensing datasets. Our retrieval algorithm with the combination of CBP and PCA obtained the best performance and outperformed several mainstream pooling or encoding methods such as full-connected layer, IFK (Improved Fisher Kernel), BoW (Bag-of-Words) and maxpooling, etc. The channel and spatial attention mechanism contributes to the CBP based retrieval method and obtained the best performance on all the datasets, as well as outperformed several recent attention methods. Source code is available at http://study.rsgis/whu.edu.cn/pages/download.

Tangent Fisher Vector on Matrix Manifolds for Action Recognition.

In this paper, we address the problem of representing and recognizing human actions from videos on matrix manifolds. For this purpose, we propose a new vector representation method, named tangent Fisher vector, to describe video sequences in the Fisher kernel framework. We first extract dense curved spatio-temporal cuboids from each video sequence. Compared with the traditional 'straight cuboids', the dense curved spatio-temporal cuboids contain much more local motion information. Each cuboid is then described using a linear dynamical system (LDS) to simultaneously capture the local appearance and dynamics. Furthermore, a simple yet efficient algorithm is proposed to learn the LDS parameters and approximate the observability matrix at the same time. Each video sequence is thus represented by a set of LDSs. Considering that each LDS can be viewed as a point in a Grassmann manifold, we propose to learn an intrinsic GMM on the manifold to cluster the LDS points. Finally a tangent Fisher vector is computed by first accumulating all the tangent vectors in each Gaussian component, and then concatenating the normalized results across all the Gaussian components. A kernel is defined to measure the similarity between tangent Fisher vectors for classification and recognition of a video sequence. This approach is evaluated on the state-of-the-art human action benchmark datasets. The recognition performance is competitive when compared with current state-of-the-art results.

Feature and Model Level Fusion of Pretrained CNN for Remote Sensing Scene Classification

Convolutional neural networks (CNN) have attracted tremendous attention in the remote sensing community due to its excellent performance in different domains. Especially for remote sensing scene classification, the CNN-based methods have brought a great breakthrough. However, it is not feasible to fully design and train a new CNN model for remote sensing scene classification, as this usually requires a large number of training samples and high computational costs. To alleviate these limitations of fully training a new model, some work attempts to use the pretrained CNN models as feature extractors to build feature representation of scene images for classification and has achieved impressive results. In this scheme, how to construct feature representation of scene image via the pretrained CNN model becomes the key process. Existing studies paid a little attention to build more discriminative feature representation by exploring the potential benefits of multilayer features from a single CNN model and different feature representations from multiple CNN models. To this end, this paper presents a fusion strategy to build the feature representation of the scene images by integrating multilayer features of a single pretrained CNN model, and extends it to a framework of multiple CNN models. For these purposes, a multiscale improved Fisher kernel coding method is used to build feature representation of the scene images on convolutional layers, and a feature fusion approach based on two feature subspace learning methods [principal component analysis (PCA)/spectral regression kernel discriminant analysis and PCA/spectral regression kernel locality preserving projection] is proposed to construct final fused features for scene classification. For validation and comparison purposes, the proposed approaches are evaluated with two challenging high-resolution remote sensing datasets and shows the competitive performance compared with existing state-of-the-art baselines such as fully trained CNN models, fine tuning CNN models, and other related works.