Dimensional Embedding Space Research Articles

Using the method of nonlinear recursive analysis for detecting DDOS anoma-lies in time series data

This research endeavors to address this gap by determining a qualitative char-acteristic for server network traffic and use it to construct the corresponding recur-rence plot (RP). The goal of this study is to develop and assess a novel technique based on nonlinear recursive analysis to detect Distributed Denial of Service (DDoS) anomalies in network traffic time series data. With the increasing frequency of DDoS attacks on modern digital infrastructures, there is a pressing need for more efficient and accurate detection methods. There has been some attempts to apply nonlinear analysis to network traffic [2-4], but those studies lack critical steps in determining parameters for embedding space dimension and delay time . More recent studies have explored machine learning and deep learning approaches [7], which offer improved accuracy but can be computationally intensive and require extensive training data. Despite the advance-ments, there remains a need for a method that is both accurate and efficient, espe-cially in real-time detection scenarios. The researchers employed nonlinear recursive analysis by estimating RQA pa-rameters and determining a qualitative characteristic of data points of DDoS attack contained in CIC-IDS2017 dataset. A technique for determining hidden information for this series and its use for constructing the corresponding recurrence diagram (RP) at the points of information retrieval are described. It is shown that the use of RP has significant drawbacks associated with the visualization of information on a computer monitor screen, so another way of research is proposed - the calculation of numerical indicators of RP The given calculated RP indicators made it possible to typify the received data and determine the type, which was named "DDOS-RP", which makes it possible to distinguish some types of DoS/DDoS type attacks. The study concludes by recom-mending further exploration of this method in diverse network environments and against more complex DDoS attack patterns.

Multifractal scaling analyses of the spatial diffusion pattern of COVID-19 pandemic in Chinese mainland

Revealing spatio-temporal evolution regularity in the spatial diffusion of epidemics is helpful for preventing and controlling the spread of epidemics. Based on the real-time COVID-19 datasets by prefecture-level cities, this paper is devoted to exploring the multifractal scaling in spatial diffusion pattern of COVID-19 pandemic and its evolution characteristics in Chinese mainland. The ArcGIS technology and box-counting method are employed to extract spatial data and the least square regression based on rescaling probability (μ-weight method) is used to calculate fractal parameters. The results show multifractal distribution of COVID-19 pandemic in China. The generalized correlation dimension spectrums are inverse S-shaped curves, but the fractal dimension values significantly exceed the Euclidean dimension of embedding space when moment order q«0. The local singularity spectrums are asymmetric unimodal curves, which slant to right. The fractal dimension growth curves are shown as quasi S-shaped curves. From these spectrums and growth curves, the main conclusions can be drawn as follows: First, self-similar patterns developed in the process of COVID-19 pandemic, which seems to be dominated by multifractal scaling law. Second, the spatial pattern of COVID-19 across China can be characterized by global clustering with local disordered diffusion. Third, the spatial diffusion process of COVID-19 in China experienced four stages, i.e., initial stage, the rapid diffusion stage, the hierarchical diffusion stage, and finally the contraction stage. This study suggests that multifractal theory can be utilized to characterize spatio-temporal diffusion of COVID-19 pandemic, and the case analyses may be instructive for further exploring natural laws of spatial diffusion.

Multi-Class Positive and Unlabeled Learning for High Dimensional Data Based on Outlier Detection in a Low Dimensional Embedding Space

Positive and unlabeled (PU) learning targets a binary classifier on labeled positive data and unlabeled data containing data samples of positive and unknown negative classes, whereas multi-class positive and unlabeled (MPU) learning aims to learn a multi-class classifier assuming labeled data from multiple positive classes. In this paper, we propose a two-step approach for MPU learning on high dimensional data. In the first step, negative samples are selected from unlabeled data using an ensemble of k-nearest neighbors-based outlier detection models in a low dimensional space which is embedded by a linear discriminant function. We present an approach for binary prediction which determines whether a data sample is a negative data sample. In the second step, the linear discriminant function is optimized on the labeled positive data and negative samples selected in the first step. It alternates between updating the parameters of the linear discriminant function and selecting reliable negative samples by detecting outliers in a low-dimensional space. Experimental results using high dimensional text data demonstrate the high performance of the proposed MPU learning method.

Deep Volumetric Descriptor Learning for Dense Correspondence of Cone-Beam Computed Tomography via Spectral Maps.

The deep neural network has achieved great success in 3D volumetric correspondence. These methods infer the dense displacement or velocity fields directly from the extracted volumetric features without addressing the intrinsic structure correspondence, being prone to shape and pose variations. On the other hand, the spectral maps address the intrinsic structure matching in the low dimensional embedding space, remain less involved in volumetric image correspondence. This paper presents an unsupervised deep volumetric descriptor learning neural network via the low dimensional spectral maps to address the dense volumetric correspondence. The neural network is optimized by a novel criterion on descriptor alignments in the spectral domain regarding the supervoxel graph. Aside from the deep convolved multi-scale features, we explicitly address the supervoxel-wise spatial and cross-channel dependencies to enrich deep descriptors. The dense volumetric correspondence is formulated as the low-dimensional spectral mapping. The proposed approach has been applied to both synthetic and clinically obtained cone-beam computed tomography images to establish dense supervoxel-wise and up-scaled voxel-wise correspondences. Extensive series of experimental results demonstrate the contribution of the proposed approach in volumetric descriptor extraction and consistent correspondence, facilitating attribute transfer for segmentation and landmark location. The proposed approach performs favorably against the state-of-the-art volumetric descriptors and the deep registration models, being resilient to pose or shape variations and independent of the prior transformations.

Non-linear Embedding Methods for Identifying Similar Brain Activity in 1 Million iEEG Records Captured From 256 RNS System Patients.

Finding electrophysiological features that are similar across patients with epilepsy may facilitate identifying treatment options for one patient that worked in patients with similar brain activity patterns. Three non-linear iEEG (intracranial electroencephalogram) embedding methods of finding similar cross-patient iEEG records in a large iEEG dataset were developed and compared. About 1 million iEEG records from 256 patients with drug-resistant focal onset seizures who were treated in prospective trials of the RNS System were used for analyses. Data from 200, 25, and 31 patients were randomly selected to be in the train, validation, and test datasets. In method 1, ResNet50 convolutional neural network (CNN) model pre-trained on the ImageNet dataset was used for extracting feature maps from spectrogram images (ImageNet-ResNet) of iEEG records. In method 2, ResNet50 custom trained on an iEEG classification task using ~138,000 manually labeled iEEG records was used as the feature extractor (ESC-ResNet). Feature maps were passed through dimensionality reduction and k nearest neighbors were found in the reduced feature space. In method 3, a 256 dimensional iEEG embedding space was learned via contrastive learning by training a ResNet50 model with triplet training sets generated using within-patient iEEG clustering (CL-ResNet). All three methods had comparable performance when identifying iEEG records from the search dataset similar to test iEEG records of baseline (non-seizure) and interictal spiking activity. Epileptic interictal spikes are represented by vertical (broadband) edges in spectrogram images, and hence even generic features extracted using models trained on everyday images appear to be sufficient to represent iEEG records with similar levels of interictal spiking activity in close proximity. In the case of electrographic seizures, however, the ESC-ResNet model, identified cross-patient iEEG records with electrographic seizure morphology features that were most similar to the test iEEG records. For nuanced electrographic seizure iEEG representation learning, domain specific model training with manually generated labels had the advantage. Finally, representative iEEG records were selected from every patient using an unsupervised clustering method which effectively reduced the number of iEEG records in the search dataset from ~750,000 to 2,148, thus substantially reducing the time required for finding similar cross-patient iEEG records.

Inductive Latent Space Sparse and Low-rank Subspace Clustering Algorithm

Sparse subspace clustering (SSC) and low-rank representation (LRR) are the most popular algorithms for subspace clustering. However, SSC and LRR are transductive methods and cannot deal with the new data not involved in the training data. When a new data comes, SSC and LRR need to calculate over all the data again, which is a time-consuming thing. On the other hand, for high-dimensional data, dimensionality reduction is firstly performed before running SSC and LRR algorithms which isolate the dimensionality reduction and the following subspace clustering. To overcome these shortcomings, in this paper, two sparse and low-rank subspace clustering algorithms based on simultaneously dimensionality reduction and subspace clustering which can deal with out-of-sample data were proposed. The proposed algorithms divide the whole data set into in-sample data and out-of-sample data. The in-sample data are used to learn the projection matrix and the sparse or low-rank representation matrix in the low-dimensional space. The membership of in-sample data is obtained by spectral clustering. In the low dimensional embedding space, the membership of out of sample data is obtained by collaborative representation classification (CRC). Experimental results on a variety of data sets verify that our proposed algorithms can handle new data in an efficient way.

Multi‐label based view learning for vehicle re‐identification

Due to the high similarity of different vehicles with similar appearances and the great diversity of camera viewpoints, vehicle re-identification (ReID) is still a challenging task. It commonly maps query set image into a high dimensional embedding space and then retrieve gallery vehicle images according to the distance. In this letter, a multi-label based view learning (MLVL) model to enhance the distinguishability of intra-class and inter-class through multiple labels learning, and decrease the distance of intra-class features caused by multi-view appearances of vehicles. Specifically, the model includes two main parts. First, the vehicle orientation estimation module is responsible for detecting key-points of the vehicle and predicting its orientation, which provides the orientation label for the multi-label based learning network to enhance view-invariant representation. Second, a novel dual-triplet and quadruplet loss function is designed to optimise intra-class and inter-class distance with the help of multi-label information, i.e. vehicle colour, type and orientation. In particular, a multi-label based sampler is proposed to generate training mini-batches instead of random sampler. Extensive experiments results show that the proposed MLVL model achieves 5.6 % $\%$ mAP improvement on VeRi-776 and 1.9 % $\%$ on VERI-Wild datasets compared with baseline model.

A digital cardiac disease biomarker from a generative progressive cardiac cine-MRI representation.

Cardiac cine-MRI is one of the most important diagnostic tools used to assess the morphology and physiology of the heart during the cardiac cycle. Nonetheless, the analysis on cardiac cine-MRI is poorly exploited and remains highly dependent on the observer's expertise. This work introduces an imaging cardiac disease representation, coded as an embedding vector, that fully exploits hidden mapping between the latent space and a generated cine-MRI data distribution. The resultant representation is progressively learned and conditioned by a set of cardiac conditions. A generative cardiac descriptor is achieved from a progressive generative-adversarial network trained to produce MRI synthetic images, conditioned to several heart conditions. The generator model is then used to recover a digital biomarker, coded as an embedding vector, following a backpropagation scheme. Then, an UMAP strategy is applied to build a topological low dimensional embedding space that discriminates among cardiac pathologies. Evaluation of the approach is carried out by using an embedded representation as a potential disease descriptor in 2296 pathological cine-MRI slices. The proposed strategy yields an average accuracy of 0.8 to discriminate among heart conditions. Furthermore, the low dimensional space shows a remarkable grouping of cardiac classes that may suggest its potential use as a tool to support diagnosis. The learned progressive and generative representation, from cine-MRI slices, allows retrieves and coded complex descriptors that results useful to discriminate among heart conditions. The cardiac disease representation expressed as a hidden embedding vector could potentially be used to support cardiac analysis on cine-MRI sequences.

Quantile–Quantile Embedding for distribution transformation and manifold embedding with ability to choose the embedding distribution

We propose a new embedding method, named Quantile–Quantile Embedding (QQE), for distribution transformation and manifold embedding with the ability to choose the embedding distribution. QQE, which uses the concept of quantile–quantile plot from visual statistical tests, can transform the distribution of data to any theoretical desired distribution or empirical reference sample. Moreover, QQE gives the user a choice of embedding distribution in embedding the manifold of data into the low dimensional embedding space. It can also be used for modifying the embedding distribution of other dimensionality reduction methods, such as PCA, t-SNE, and deep metric learning, for better representation or visualization of data. We propose QQE in both unsupervised and supervised forms. QQE can also transform a distribution to either an exact reference distribution or its shape. We show that QQE allows for better discrimination of classes in some cases. Our experiments on different synthetic and image datasets show the effectiveness of the proposed embedding method.

Echo-SyncNet: Self-Supervised Cardiac View Synchronization in Echocardiography

In echocardiography (echo), an electrocardiogram (ECG) is conventionally used to temporally align different cardiac views for assessing critical measurements. However, in emergencies or point-of-care situations, acquiring an ECG is often not an option, hence motivating the need for alternative temporal synchronization methods. Here, we propose Echo-SyncNet, a self-supervised learning framework to synchronize various cross-sectional 2D echo series without any human supervision or external inputs. The proposed framework takes advantage of two types of supervisory signals derived from the input data: spatiotemporal patterns found between the frames of a single cine (intra-view self-supervision) and interdependencies between multiple cines (inter-view self-supervision). The combined supervisory signals are used to learn a feature-rich and low dimensional embedding space where multiple echo cines can be temporally synchronized. Two intra-view self-supervisions are used, the first is based on the information encoded by the temporal ordering of a cine (temporal intra-view) and the second on the spatial similarities between nearby frames (spatial intra-view). The inter-view self-supervision is used to promote the learning of similar embeddings for frames captured from the same cardiac phase in different echo views. We evaluate the framework with multiple experiments: 1) Using data from 998 patients, Echo-SyncNet shows promising results for synchronizing Apical 2 chamber and Apical 4 chamber cardiac views, which are acquired spatially perpendicular to each other; 2) Using data from 3070 patients, our experiments reveal that the learned representations of Echo-SyncNet outperform a supervised deep learning method that is optimized for automatic detection of fine-grained cardiac cycle phase; 3) We go one step further and show the usefulness of the learned representations in a one-shot learning scenario of cardiac key-frame detection. Without any fine-tuning, key frames in 1188 validation patient studies are identified by synchronizing them with only one labeled reference cine. We do not make any prior assumption about what specific cardiac views are used for training, and hence we show that Echo-SyncNet can accurately generalize to views not present in its training set. Project repository: github.com/fatemehtd/Echo-SyncNet>.

Transductive Mutual Information Encoder Network for Few Shot Learning

Deep learning has made breakthroughs in recent decades and has been widely used in many domains. However, most of those methods heavily rely on large labeled datasets, which results in poor performance when provided with limited labeled data. Few-shot learning (FSL), which aims at learning a novel task with limited samples, has attracted a lot of research recently. The previous metric-based methods ignore the internal bias between the training and testing datasets since the categories of the testing dataset are completely different from the training set. Transfer learning methods also suffer from few labeled data and tends to be overfitting in this situation. This paper proposes Transductive Mutual Information Encoder Network (TMIN) for few-shot learning problems. TMIN typically trains a convolutional neural network with a mutual information maximization module in an unsupervised manner. The trained network maps images to a high dimensional embedding space. Then the embeddings are exploited to measure the similarity between samples by a distance metric. Experiments indicate that the proposed model achieves competitive performance compared with the counterparts.

Multifractal scaling analyses of urban street network structure: The cases of twelve megacities in China

Traffic networks have been proved to be fractal systems. However, previous studies mainly focused on monofractal networks, while complex systems are of multifractal structure. This paper is devoted to exploring the general regularities of multifractal scaling processes in the street network of 12 Chinese cities. The city clustering algorithm is employed to identify urban boundaries for defining comparable study areas; box-counting method and the direct determination method are utilized to extract spatial data; the least squares calculation is employed to estimate the global and local multifractal parameters. The results showed multifractal structure of urban street networks. The global multifractal dimension spectrums are inverse S-shaped curves, while the local singularity spectrums are asymmetric unimodal curves. If the moment order q approaches negative infinity, the generalized correlation dimension will seriously exceed the embedding space dimension 2, and the local fractal dimension curve displays an abnormal decrease for most cities. The scaling relation of local fractal dimension gradually breaks if the q value is too high, but the different levels of the network always keep the scaling reflecting singularity exponent. The main conclusions are as follows. First, urban street networks follow multifractal scaling law, and scaling precedes local fractal structure. Second, the patterns of traffic networks take on characteristics of spatial concentration, but they also show the implied trend of spatial deconcentration. Third, the development space of central area and network intensive areas is limited, while the fringe zone and network sparse areas show the phenomenon of disordered evolution. This work may be revealing for understanding and further research on complex spatial networks by using multifractal theory.

Fast hard negative mining for deep metric learning

Deep metric learning methods aim to measure similarity of data points (e.g. images) by calculating their distance in a high dimensional embedding space. These methods are usually trained by optimizing a ranking loss function, which is designed to bring together samples from the same class while separating them from samples from all other classes. The most challenging part of these methods is the selection of samples that contribute to effective network training. In this paper we present Bag of Negatives (BoN), a fast hard negative mining method, that provides a set, triplet or pair of potentially relevant training samples. BoN is an efficient method that selects a bag of hard negatives based on a novel online hashing strategy. We show the superiority of BoN against state-of-the-art hard negative mining methods in terms of accuracy and training time over three large datasets.

Scaling behavior of crystalline membranes: An ε-expansion approach

We study the scaling behavior of two-dimensional (2D) crystalline membranes in the flat phase by a renormalization group (RG) method and an ε-expansion. Generalization of the problem to non-integer dimensions, necessary to control the ε-expansion, is achieved by dimensional continuation of a well-known effective theory describing out-of-plane fluctuations coupled to phonon-mediated interactions via a scalar composite field, equivalent for small deformations to the local Gaussian curvature. The effective theory, which will be referred to as Gaussian curvature interaction (GCI) model, is equivalent to theories of elastic D-dimensional manifolds fluctuating in a (D+dc)-dimensional embedding space in the physical case D=2 for arbitrary dc. For D≠2, instead, the GCI model is not equivalent to a direct dimensional continuation of elastic membrane theory and it defines an alternative generalization to generic internal dimension D. After decoupling interactions through a Hubbard-Stratonovich transformation, we study the GCI model by perturbative field-theoretic RG within the framework of an expansion in ε=(4−D). We calculate explicitly RG functions at two-loop order and determine the exponent η characterizing the long-wavelength scaling of correlation functions to order ε2. The value of η at this order is shown to be insensitive to Feynman diagrams involving vertex corrections. As a consequence, the self-consistent screening approximation for the GCI model is shown to be exact to O(ε2). In the physical case of a single out-of-plane displacement field, dc=1, the O(ε2) correction is suppressed by a small numerical prefactor. As a result, despite the large value of ε=2, extrapolation of the first and second order results to D=2 leads to very close numbers, η=0.8 and η≃0.795. The calculated exponent values are close to earlier reference results obtained by non-perturbative renormalization group, the self-consistent screening approximation and numerical simulations. These indications suggest that a perturbative analysis of the GCI model could provide an useful framework for accurate quantitative predictions of the scaling exponent even at D=2.

Kernel compositional embedding and its application in linguistic structured data classification

In many applications such as natural language processing, speech recognition, and computer vision, there are inputs with hierarchical compositional structure and long relation among their subcomponents. Introducing this information in the definition of a model can improve its performance in dealing with this type of data. On the other side, the high generalization power of kernel methods is proven in traditional machine learning problems. If we can employ some idea of these methods on handling structured objects, we can benefit from improving the performance and the generalization capability of compositional models. Accordingly, a new approach is introduced in this paper to realize the idea of simultaneously leveraging advantages of both kernel methods and compositional embedding to provide powerful representations and classifiers for structured data. Based on this approach, which is named Kernel Compositional Embedding (KCE), we propose two methods: Direct KCE (DKCE), and Indirect KCE (IKCE). In the DKCE method, we directly deal with a potentially infinite dimensional embedding space, i.e., the embeddings are used implicitly in the classification task. In IKCE method, instead of implicitly performing all operations inside embedding space, only some elements of one or more reproducing kernel Hilbert spaces are used to embed structured objects into low-dimensional Euclidean spaces. To evaluate the performance of the proposed methods, we apply them on two common computational linguistic tasks, i.e, sentiment analysis and natural language inference. The experimental results illustrate that the classification performance of the proposed methods is higher than or competitive to some well-known competing methods.

Machine Learning of Time Series Using Time-Delay Embedding and Precision Annealing.

Tasking machine learning to predict segments of a time series requires estimating the parameters of a ML model with input/output pairs from the time series. We borrow two techniques used in statistical data assimilation in order to accomplish this task: time-delay embedding to prepare our input data and precision annealing as a training method. The precision annealing approach identifies the global minimum of the action (). In this way, we are able to identify the number of training pairs required to produce good generalizations (predictions) for the time series. We proceed from a scalar time series and, using methods of nonlinear time series analysis, show how to produce a -dimensional time-delay embedding space in which the time series has no false neighbors as does the observed time series. In that -dimensional space, we explore the use of feedforward multilayer perceptrons as network models operating on -dimensional input and producing -dimensional outputs.

Fault Diagnosis of Bearing by Utilizing LWT‐SPSR‐SVD‐Based RVM with Binary Gravitational Search Algorithm

The fault diagnosis method of bearing based on lifting wavelet transform (LWT)‐self‐adaptive phase space reconstruction (SPSR)‐singular value decomposition (SVD)‐based relevance vector machine (RVM) with binary gravitational search algorithm (BGSA) is presented in this study, among which LWT‐SPSR‐SVD (LSS) is presented for feature extraction of the bearing vibration signal, the dynamic characteristics of lifting wavelet coefficients′ (LWCs′) reconstructed signals of the bearing vibration signal can be reflected by SPSR for LWCs′ reconstructed signals of the bearing vibration signal, and BGSA is used to select the embedding space dimension and time delay of phase space reconstruction (PSR) and kernel parameter of RVM. In order to show the superiority of LWT‐SPSR‐SVD‐based RVM with BGSA (LSS‐BGSA‐RVM), the traditional RVM trained by the training samples with the features based on LWT‐SVD (LS‐RVM) is used to compare with the proposed LSS‐BGSA‐RVM method. The experimental result demonstrates that compared with LS‐RVM, LSS‐BGSA‐RVM can achieve the higher diagnosis accuracy for bearing.

An integrated segmentation and shape-based classification scheme for distinguishing adenocarcinomas from granulomas on lung CT.

Distinguishing between benign granulmoas and adenocarcinomas is confounded by their similar visual appearance on routine CT scans. Unfortunately, owing to the inability to discriminate these lesions radigraphically, many patients with benign granulomas are subjected to unnecessary surgical wedge resections and biopsies for pathologic confirmation of cancer presence or absence. This suggests the need for improved computerized characterization of these nodules in order to distinguish between these two classes of lesions on CT scans. While there has been substantial interest in the use of textural analysis for radiomic characterization of lung nodules, relatively less work has been done in shape based characterization of lung nodules, particularly with respect to granulmoas and adenocarcinomas. The primary goal of this study is to evaluate the role of 3D shape features for discrimination of benign granulomas from malignant adenocarcinomas on lung CT images. Towards this end we present an integrated framework for segmentation, feature characterization and classification of these nodules on CT. The nodule segmentation method starts with separation of lung regions from the surrounding lung anatomy. Next, the lung CT scans are projected into and represented in a three dimensional spectral embedding (SE) space, allowing for better determination of the boundaries of the nodule. This then enables the application of a gradient vector flow active contour (SEGvAC) model for nodule boundary extraction. A set of 24 shape features from both 2D slices and 3D surface of the segmented nodules are extracted, including features pertaining to the angularity, spiculation, elongation and nodule compactness. A feature selection scheme, PCA-VIP, is employed to identify the most discriminating set of features to distinguish granulmoas from adenocarcinomas within a learning set of 82 patients. The features thus identified were then combined with a support vector machine classifier and independently validated on a distinct test set comprising 67 patients. The performance of the classifier for both of the training and validation cohorts was evaluated by the area under receiver characteristic curve (ROC). We used 82 and 67 studies from two different institutions respectively for training and independent validation of the model and the shape features. The Dice coefficient between automatically segmented nodules by SEGvAC and the manual delineations by expert radiologists (readers) was 0.84± 0.04 whereas inter-reader segmentation agreement was 0.79± 0.12. We also identified a set of consistent features (Roughness, Convexity and Spherecity) that were found to be strongly correlated across both manual and automated nodule segmentations (R > 0.80, p < 0.0001) and capture the marginal smoothness and 3D compactness of the nodules. On the independent validation set of 67 studies our classifier yielded a ROC AUC of 0.72 and 0.64 for manually- and automatically segmented nodules respectively. On a subset of 20 studies, the AUCs for the two expert radiologists and 1 pulmonologist were found to be 0.82, 0.68 and 0.58 respectively. The major finding of this study was that certain shape features appear to differentially express between granulomas and adenocarcinomas and thus computer extracted shape cues could be used to distinguish these radiographically similar pathologies.

DEEP ATTRACTOR NETWORK FOR SINGLE-MICROPHONE SPEAKER SEPARATION.

Despite the overwhelming success of deep learning in various speech processing tasks, the problem of separating simultaneous speakers in a mixture remains challenging. Two major difficulties in such systems are the arbitrary source permutation and unknown number of sources in the mixture. We propose a novel deep learning framework for single channel speech separation by creating attractor points in high dimensional embedding space of the acoustic signals which pull together the time-frequency bins corresponding to each source. Attractor points in this study are created by finding the centroids of the sources in the embedding space, which are subsequently used to determine the similarity of each bin in the mixture to each source. The network is then trained to minimize the reconstruction error of each source by optimizing the embeddings. The proposed model is different from prior works in that it implements an end-to-end training, and it does not depend on the number of sources in the mixture. Two strategies are explored in the test time, K-means and fixed attractor points, where the latter requires no post-processing and can be implemented in real-time. We evaluated our system on Wall Street Journal dataset and show 5.49% improvement over the previous state-of-the-art methods.

Spike pattern recognition by supervised classification in low dimensional embedding space.

Epileptiform discharges in interictal electroencephalography (EEG) form the mainstay of epilepsy diagnosis and localization of seizure onset. Visual analysis is rater-dependent and time consuming, especially for long-term recordings, while computerized methods can provide efficiency in reviewing long EEG recordings. This paper presents a machine learning approach for automated detection of epileptiform discharges (spikes). The proposed method first detects spike patterns by calculating similarity to a coarse shape model of a spike waveform and then refines the results by identifying subtle differences between actual spikes and false detections. Pattern classification is performed using support vector machines in a low dimensional space on which the original waveforms are embedded by locality preserving projections. The automatic detection results are compared to experts’ manual annotations (101 spikes) on a whole-night sleep EEG recording. The high sensitivity (97 %) and the low false positive rate (0.1 min−1), calculated by intra-patient cross-validation, highlight the potential of the method for automated interictal EEG assessment.