Distance In Feature Space Research Articles

Neuronal parts list and wiring diagram for a visual system

A catalogue of neuronal cell types has often been called a ‘parts list’ of the brain1, and regarded as a prerequisite for understanding brain function2,3. In the optic lobe of Drosophila, rules of connectivity between cell types have already proven to be essential for understanding fly vision4,5. Here we analyse the fly connectome to complete the list of cell types intrinsic to the optic lobe, as well as the rules governing their connectivity. Most new cell types contain 10 to 100 cells, and integrate information over medium distances in the visual field. Some existing type families (Tm, Li, and LPi)6–10 at least double in number of types. A new serpentine medulla (Sm) interneuron family contains more types than any other. Three families of cross-neuropil types are revealed. The consistency of types is demonstrated by analysing the distances in high-dimensional feature space, and is further validated by algorithms that select small subsets of discriminative features. We use connectivity to hypothesize about the functional roles of cell types in motion, object and colour vision. Connectivity with ‘boundary types’ that straddle the optic lobe and central brain is also quantified. We showcase the advantages of connectomic cell typing: complete and unbiased sampling, a rich array of features based on connectivity and reduction of the connectome to a substantially simpler wiring diagram of cell types, with immediate relevance for brain function and development.

Multi-scale locality preserving projection for partial multi-view incomplete multi-label learning

Amidst advancements in feature extraction techniques, research on multi-view multi-label classifications has attracted widespread interest in recent years. However, real-world scenarios often pose a challenge where the completeness of multiple views and labels cannot be ensured. At present, only a handful of techniques have attempted to address the complex issue of partial multi-view incomplete multi-label classification, and the majority of these approaches overlook the significance of manifold structures between instances. To tackle these challenges, we propose a novel partial multi-view incomplete multi-label learning model, termed MSLPP. Differing from existing studies, MSLPP emphasizes retaining the effective inherent structure of data during the feature extraction process, thereby facilitating a richer semantic information extraction. Specifically, MSLPP captures and integrates four types of information: the distance and similarity information in the original feature space, and the distance and similarity information in the extracted feature space. Further, by adopting the graph embedding technique, it simultaneously preserves the intrinsic structure with multi-scale information through a constraint term. Moreover, taking into account the negative impact of the missing views on the model and the possible impact of missing views on the data inherent structure, we further propose a shielding strategy for missing views, which not only eliminates the negative effects of missing views on the model but also more accurately captures the inherent data structure. The experimental results on five widely recognized datasets indicate that the model performs better than many excellent methods.

Analysis of machine learning prediction reliability based on sampling distance evaluation with feature decorrelation

Despite successful use in a wide variety of disciplines for data analysis and prediction, machine learning (ML) methods suffer from a lack of understanding of the reliability of predictions due to the lack of transparency and black-box nature of ML models. In materials science and other fields, typical ML model results include a significant number of low-quality predictions. This problem is known to be particularly acute for target systems which differ significantly from the data used for ML model training. However, to date, a general method for uncertainty quantification (UQ) of ML predictions has not been available. Focusing on the intuitive and computationally efficient similarity-based UQ, we show that a simple metric based on Euclidean feature space distance and sampling density together with the decorrelation of the features using Gram–Schmidt orthogonalization allows effective separation of the accurately predicted data points from data points with poor prediction accuracy. To demonstrate the generality of the method, we apply it to support vector regression models for various small data sets in materials science and other fields. We also show that this metric is a more effective UQ tool than the standard approach of using the average distance of k nearest neighbors (k = 1–10) in features space for similarity evaluation. Our method is computationally simple, can be used with any ML learning method and enables analysis of the sources of the ML prediction errors. Therefore, it is suitable for use as a standard technique for the estimation of ML prediction reliability for small data sets and as a tool for data set design.

LDMVFI: Video Frame Interpolation with Latent Diffusion Models

Existing works on video frame interpolation (VFI) mostly employ deep neural networks that are trained by minimizing the L1, L2, or deep feature space distance (e.g. VGG loss) between their outputs and ground-truth frames. However, recent works have shown that these metrics are poor indicators of perceptual VFI quality. Towards developing perceptually-oriented VFI methods, in this work we propose latent diffusion model-based VFI, LDMVFI. This approaches the VFI problem from a generative perspective by formulating it as a conditional generation problem. As the first effort to address VFI using latent diffusion models, we rigorously benchmark our method on common test sets used in the existing VFI literature. Our quantitative experiments and user study indicate that LDMVFI is able to interpolate video content with favorable perceptual quality compared to the state of the art, even in the high-resolution regime. Our code is available at https://github.com/danier97/LDMVFI.

Unsupervised anomaly detection for manufacturing product images by significant feature space distance measurement

Using camera instruments to detect product anomalies is increasingly being applied in various manufacturing scenarios. Detection of anomaly product images and identification of anomaly areas can not only ensure product quality, but also help reveal potential problems in the manufacturing process. Unsupervised data environments and diverse engineering requirements are two challenges faced by modern manufacturing. To solve the problem of manufacturing product anomaly detection in the absence of anomaly data, and meet the requirements of model training cost, detection speed and accuracy, a fast deployable unsupervised image anomaly detection method is proposed without parameter training. The main idea is to design an efficient and training-free anomaly detection decoder. The encoder output is screened by non-parametric method, and the image anomaly score is calculated by the cumulative distance of the screened feature space. In addition, the proposed model is able to give the reference samples on which the prediction results are based to improve the interpretability of the predictions. Experiments are conducted with 15 real product image datasets, and the results show that the proposed method achieves performance improvement over other methods for both anomaly detection and segmentation.

P2S distance induced locally conjugated orthogonal subspace learning for feature extraction

When performing data classification tasks, it often occurs to them the curse of dimensionality problem. To address the issue, a manifold learning method termed locally conjugated orthogonal subspace (LCOS) is put forward for dimensionality reduction or feature extraction in this paper. Note that point to feature space (P2S) distance contributes to mining local geometry information, both a local margin characterizing data apartness and a locally conjugated orthogonal constraint beneficial to removing data redundancy are well studied from the P2S distance metric. They are all exploited to model the proposed LCOS. Then, a low dimensional subspace can be explored by maximizing the P2S distance induced local margin under the constraint. Compared with some other related dimensionality reduction methods, experimental results on benchmark face and object data sets validate the performance of the proposed method.

Calibrated simplex-mapping classification.

We propose a novel methodology for general multi-class classification in arbitrary feature spaces, which results in a potentially well-calibrated classifier. Calibrated classifiers are important in many applications because, in addition to the prediction of mere class labels, they also yield a confidence level for each of their predictions. In essence, the training of our classifier proceeds in two steps. In a first step, the training data is represented in a latent space whose geometry is induced by a regular (n - 1)-dimensional simplex, n being the number of classes. We design this representation in such a way that it well reflects the feature space distances of the datapoints to their own- and foreign-class neighbors. In a second step, the latent space representation of the training data is extended to the whole feature space by fitting a regression model to the transformed data. With this latent-space representation, our calibrated classifier is readily defined. We rigorously establish its core theoretical properties and benchmark its prediction and calibration properties by means of various synthetic and real-world data sets from different application domains.

SPECHT: Self-tuning Plausibility based object detection Enables quantification of Conflict in Heterogeneous multi-scale microscopy.

Identification of small objects in fluorescence microscopy is a non-trivial task burdened by parameter-sensitive algorithms, for which there is a clear need for an approach that adapts dynamically to changing imaging conditions. Here, we introduce an adaptive object detection method that, given a microscopy image and an image level label, uses kurtosis-based matching of the distribution of the image differential to express operator intent in terms of recall or precision. We show how a theoretical upper bound of the statistical distance in feature space enables application of belief theory to obtain statistical support for each detected object, capturing those aspects of the image that support the label, and to what extent. We validate our method on 2 datasets: distinguishing sub-diffraction limit caveolae and scaffold by stimulated emission depletion (STED) super-resolution microscopy; and detecting amyloid-β deposits in confocal microscopy retinal cross-sections of neuropathologically confirmed Alzheimer's disease donor tissue. Our results are consistent with biological ground truth and with previous subcellular object classification results, and add insight into more nuanced class transition dynamics. We illustrate the novel application of belief theory to object detection in heterogeneous microscopy datasets and the quantification of conflict of evidence in a joint belief function. By applying our method successfully to diffraction-limited confocal imaging of tissue sections and super-resolution microscopy of subcellular structures, we demonstrate multi-scale applicability.

Mapping the presence and distribution of tree species in Canada's forested ecosystems

Knowledge of tree species is required to inform management, planning, and monitoring of forests as well as to characterize habitat and ecosystem function. Remotely sensed data and spatial modeling enable mapping of tree species presence and distribution. Following an assessment of tree species identified in the sample-based National Forest Inventory (NFI), we mapped 37 tree species over the 650-Mha, forest-dominated ecosystems of Canada representing 2019 conditions. Landsat imagery and related spectral indices, geographic and climate data, elevation derivatives, and remote sensing-derived phenology are used as predictor variables trained with calibration samples from Canada's NFI using the Random Forests machine learning algorithm. Based upon prior knowledge of tree species distributions, classification models were implemented on a regional basis, meaning only the tree species that are expected in a given mapping region were modeled using local calibration samples. Modeling resulted in class membership probabilities values for each regionally eligible tree species for all treed pixels as well as an indicator of attribution confidence derived from the distance in feature space between the two leading classes. Accuracy assessment was conducted using independent validation data also drawn from the NFI following the same selection rules and indicated an overall accuracy of 93.1% ± 0.1% (95%-confidence interval). Predictor variables informing on geographic, climatic and topographic conditions had the largest importance on the classification models. Nationally, the most common leading tree species were black spruce (Picea mariana; 203 Mha or 57.3% of the treed area), trembling aspen (Populus tremuloides; 34.7 Mha, 9.8%), and lodgepole pine (Pinus contorta; 21.1 Mha, 5.9%). Regionally, there was ecozone-level dominance of other tree species, including subalpine fir (Abies lasiocarpa; Montane Cordillera), western hemlock (Tsuga heterophylla; Pacific Maritime), and balsam fir (Abies balsamea; Atlantic Maritime). Based upon the per-pixel class membership probabilities, species assemblages akin to those in forest inventories can also be produced. Further, given the calibrated reflectance of Landsat imagery, the methods presented herein can be implemented over a time series of images. The approach uses open data as predictor variables, making the method portable to other areas given availability of tree species training data.

Tsbagging: A Novel Cross-Project Software Defect Prediction Algorithm Based on Semisupervised Clustering

Software defect prediction (SDP) is an important technology which is widely applied to improve software quality and reduce development costs. It is difficult to train the SDP model when software to be test only has limited historical data. Cross-project defect prediction (CPDP) has been proposed to solve this problem by using source project data to train the defect prediction model. Most of CPDP methods build defect prediction models based on the similarity of feature space or data distance between different projects. However, when the target project has a small amount of label data, these methods usually do not consider this part of data information. Therefore, when the distribution between source project and target project is quite different, these methods are difficult to achieve good prediction performance. To solve this problem, this paper proposes a CPDP method based on a semisupervised clustering (namely, Tsbagging). Tsbagging has two stages; in the first stage, we cluster to the source project data based on the limited labeled data in the target project and assign different weights to these source project data according to the clustering results. In the second stage, we use bagging method to train the prediction model based on the weight assigned in the first stage. The experimental results show that the performance achieved by Tsbagging is better than other existing SDP methods.

Optimal design of a cathode flow field with a new arrangement of baffle plates for a high clean power generation of a polymer electrolyte membrane fuel cell

Polymer electrolyte membrane fuel cell is a clean and promising energy device in the near future; the present work proposes newly designed baffle plates that feature unequal height and space distance arrangements in the cathode flow field of fuel cells. Using a three dimensions multiphase fuel cell model, the effects of various newly designed baffle plates with different numbers, unequal height and space distance arrangements on the oxygen diffusion, water transport behavior and performance of a fuel cell are comprehensively evaluated. The results show that the baffle plates significantly promote oxygen diffusion and water elimination in fuel cells compared to the regular straight flow field, especially in the regions around the baffle plates due to the convection effect. A gradually increasing height and a decreasing space distance between the baffle plates can further relieve water flooding and oxygen starvation in the porous electrodes downstream of the channel under high current densities; hence, they significantly improve the maximum power density by 9.67% compared to the regular flow field without baffle plates. In addition, the optimal newly designed baffle plate flow field pattern exhibits uniform physical quantities with a relatively low-pressure loss along the flow channel, which reduces the pumping power to supply reactants and favors the energy efficiency of fuel cells.

High-precision inversion of dynamic radiography using hydrodynamic features.

While radiography is routinely used to probe complex, evolving density fields in research areas ranging from materials science to shock physics to inertial confinement fusion and other national security applications, complications resulting from noise, scatter, complex beam dynamics, etc. prevent current methods of reconstructing density from being accurate enough to identify the underlying physics with sufficient confidence. In this work, we show that using only features that are robustly identifiable in radiographs and combining them with the underlying hydrodynamic equations of motion using a machine learning approach of a conditional generative adversarial network (cGAN) provides a new and effective approach to determine density fields from a dynamic sequence of radiographs. In particular, we demonstrate the ability of this method to outperform a traditional, direct radiograph to density reconstruction in the presence of scatter, even when relatively small amounts of scatter are present. Our experiments on synthetic data show that the approach can produce high quality, robust reconstructions. We also show that the distance (in feature space) between a testing radiograph and the training set can serve as a diagnostic of the accuracy of the reconstruction.

Memory Linked Anomaly Metric Learning of Thermography Rail Defects Detection System

With the development of railway, high speed and intensive rolling will significantly increase the probability of fatigue damage and they lay a hidden danger for major safety accidents. Therefore, it is vital to detect defects of the rail. The commonly used algorithms mostly trained with the defect samples whereas this data is rare and cannot cover all types of defects. On the contrary, the background samples are relatively easy to obtain. In this work, a new algorithm is proposed which models the background of rail samples with memory items in order to exploit the full use of the background information. The samples that do not fit with the models are considered as defect candidates. Our model is only trained with the normal samples and a small amount of defect samples. When new patterns occur, they will be considered as defect candidates and the model is not limited to the known defect patterns. Additionally, the proposed metric learning modules enable the model to learn representative memory items, and enlarges the feature space distance between defect and normal samples as well as reducing the influence of disturbance. In order to show the efficacy of the proposed algorithm, it has been validated with various types of rail defects, ranging from the visually apparent defects to visually weak defects. In the experiments, the thermography inspection system scans at a constant speed while inductively heats the rail tracks continuously to record the infrared thermal image sequences from the rail surface. Recent anomaly detection algorithms are used as platform for objective performance evaluation. The algorithms are evaluated quantitatively based on Area under ROC curve (AUC) at both image level and pixel level along with the visual comparative analysis.

Comparison of Novelty Detection Methods for Detection of Various Rotary Machinery Faults.

Condition monitoring is an indispensable element related to the operation of rotating machinery. In this article, the monitoring system for the parallel gearbox was proposed. The novelty detection approach is used to develop the condition assessment support system, which requires data collection for a healthy structure. The measured signals were processed to extract quantitative indicators sensitive to the type of damage occurring in this type of structure. The indicator’s values were used for the development of four different novelty detection algorithms. Presented novelty detection models operate on three principles: feature space distance, probability distribution, and input reconstruction. One of the distance-based models is adaptive, adjusting to new data flowing in the form of a stream. The authors test the developed algorithms on experimental and simulation data with a similar distribution, using the training set consisting mainly of samples generated by the simulator. Presented in the article results demonstrate the effectiveness of the trained models on both data sets.

Learning Domain‐Invariant and Discriminative Features for Homogeneous Unsupervised Domain Adaptation

A classifier trained on the label-rich source dataset tends to perform poorly on the unlabeled target dataset because of the distribution discrepancy across different datasets. Unsupervised domain adaptation aims to transfer knowledge from the labeled source dataset to the unlabeled target dataset to solve this problem. Most of the existing unsupervised domain adaptation methods only concentrate on learning domain-invariant features across different domains, but they neglect the discriminability of the learned features to satisfy the cluster assumption. In this paper, we propose Semantic pairwise centroid alignment (SPCA), which is a point-wise method to learn both domain-invariant and discriminative features for homogeneous unsupervised domain adaptation. SPCA utilizes a novel semantic centroid loss to reduce the intraclass distance in feature space by using source data and target High-confidence centroid points (HCCPs). Then a classifier trained on source features is expected to generalize well on target features. Extensive experiments on visual recognition tasks verify the effectiveness of the proposed SPCA and also demonstrate that both domaininvariant and discriminative features learned by SPCA can significantly boost the performance of homogeneous unsupervised domain adaptation.

Density-based outlier scoring on Kepler data

ABSTRACT In the present era of large-scale surveys, big data present new challenges to the discovery process for anomalous data. Such data can be indicative of systematic errors, extreme (or rare) forms of known phenomena, or most interestingly, truly novel phenomena that exhibit as-of-yet unobserved behaviours. In this work, we present an outlier scoring methodology to identify and characterize the most promising unusual sources to facilitate discoveries of such anomalous data. We have developed a data mining method based on k-nearest neighbour distance in feature space to efficiently identify the most anomalous light curves. We test variations of this method including using principal components of the feature space, removing select features, the effect of the choice of k, and scoring to subset samples. We evaluate the performance of our scoring on known object classes and find that our scoring consistently scores rare (<1000) object classes higher than common classes. We have applied scoring to all long cadence light curves of Quarters 1–17 of Kepler’s prime mission and present outlier scores for all 2.8 million light curves for the roughly 200k objects.

Intelligent Fault Diagnosis by Fusing Domain Adversarial Training and Maximum Mean Discrepancy via Ensemble Learning

Nowadays, the industrial Internet of Things (IIoT) has been successfully utilized in smart manufacturing. The massive amount of data in IIoT promote the development of deep learning-based health monitoring for industrial equipment. Since monitoring data for mechanical fault diagnosis collected on different working conditions or equipment have domain mismatch, models trained with training data may not work in practical applications. Therefore, it is essential to study fault diagnosis methods with domain adaptation ability. In this article, we propose an intelligent fault diagnosis method based on an improved domain adaptation method. Specifically, two feature extractors concerning feature space distance and domain mismatch are trained using maximum mean discrepancy and domain adversarial training respectively to enhance feature representation. Since separate classifiers are trained for feature extractors, ensemble learning is further utilized to obtain final results. Experimental results indicate that the proposed method is effective and applicable in diagnosing faults with domain mismatch.

An Identity-Preserved Model for Face Sketch-Photo Synthesis

Face sketch-photo synthesis can be regarded as an image-to-image translation problem. Although many generative models achieve good translations from sketches to photos, they still have limitations in preserving face identity due to the huge modality gap of the two domains. To this end, we propose an identity-preserved adversarial model (IPAM), which includes an extended U-Net to increase the weight of the original sketch in translation, two discriminators focusing on the real or fake image concatenation of two domains to learn more styles of the target domain, and an identity constraint to request the fakes and the real targets to have zero cosine distance in feature space. We evaluate our method on two face sketch databases with face recognition. The results demonstrate our translation method is superior to the existing methods in maintaining face identity information.

Joint Gaze Correction and Face Beautification for Conference Video Using Dual Sparsity Prior

Gaze mismatch is a common problem in video conferencing, where the viewpoint captured by a camera (usually located above or below a display monitor) is not aligned with the gaze direction of the human subject, who typically looks at his counterpart in the center of the screen. This means that the two parties cannot converse eye-to-eye, hampering the visual communication experience. A recent popular approach to the gaze mismatch problem is to synthesize a gaze-corrected face image as viewed from the screen center via depth-image-based rendering (DIBR), assuming texture and depth maps are both available at the camera-captured viewpoint. Due to self-occlusion, however, there will be missing pixels in the DIBR-synthesized view image that require satisfactory filling. In this paper, we propose to jointly solve the hole-filling problem and the face beautification problem (subtle modifications of facial components and contour to enhance attractiveness of the rendered face) using dual sparsity prior. Specifically, prior to the start of a video conference session, we first train two dictionaries separately offline using two large data sets: one with general face images, and the other with “beautiful” human faces, which means faces with high beauty scores. During the actual conference session, we solve the hole-filling and facial components beautification problems simultaneously by seeking two code vectors—one is sparse in the first dictionary and explains the available DIBR-synthesized pixels, the other is sparse in the second dictionary and matches well with the first vector in terms of feature space distance. This ensures an acceptable level of recognizability of the conference subject, while increasing proximity to “beautiful” facial features to improve attractiveness. Experimental results show naturally rendered human faces with noticeably improved attractiveness.

Online Fast Adaptive Low-Rank Similarity Learning for Cross-Modal Retrieval

The semantic similarity among cross-modal data objects, e.g., similarities between images and texts, are recognized as the bottleneck of cross-modal retrieval. However, existing batch-style correlation learning methods suffer from prohibitive time complexity and extra memory consumption in handling large-scale high dimensional cross-modal data. In this paper, we propose a Cross-Modal Online Low-Rank Similarity function learning (CMOLRS) method, which learns a low-rank bilinear similarity measurement for cross-modal retrieval. We model the cross-modal relations by relative similarities on the training data triplets and formulate the relative relations as convex hinge loss. By adapting the margin in hinge loss with pair-wise distances in feature space and label space, CMOLRS effectively captures the multi-level semantic correlation and adapts to the content divergence among cross-modal data. Imposed with a low-rank constraint, the similarity function is trained by online learning in the manifold of low-rank matrices. The low-rank constraint not only endows the model learning process with faster speed and better scalability, but also improves the model generality. We further propose fast-CMOLRS combining multiple triplets for each query instead of standard process using single triplet at each model update step, which further reduces the times of gradient updates and retractions. Extensive experiments are conducted on four public datasets, and comparisons with state-of-the-art methods show the effectiveness and efficiency of our approach.