Input Space Research Articles

Work estimation of construction workers for productivity monitoring using kinematic data and deep learning

In this study, a novel method for work estimation is presented. The aim is to build an accurate and reliable work classification algorithm that can help monitor construction sites without unnecessarily constraining the workers or installing heavy sensing infrastructure. The method utilizes deep learning algorithms to classify multivariate time-series data collected from five inertial measurement units mounted on the worker. Three models are developed, differing in window sizes from 3 to 7 s. The best performing model achieves an accuracy of 90% and an F1 score of 0.876. The model is analyzed and pruned using expected gradients for feature selection. The process reduces the input space by 60%, equivalent to 3 sensors. This is an initial step towards a general model that can classify productivity measures for workers on construction sites, which will provide valuable input for monitoring construction site activities and future analyses such as forecasting productivity.

A data-driven intelligent decision support system that combines predictive and prescriptive analytics for the design of new textile fabrics

In this paper, we propose an Intelligent Decision Support System (IDSS) for the design of new textile fabrics. The IDSS uses predictive analytics to estimate fabric properties (e.g., elasticity) and composition values (% cotton) and then prescriptive techniques to optimize the fabric design inputs that feed the predictive models (e.g., types of yarns used). Using thousands of data records from a Portuguese textile company, we compared two distinct Machine Learning (ML) predictive approaches: Single-Target Regression (STR), via an Automated ML (AutoML) tool, and Multi-target Regression, via a deep learning Artificial Neural Network. For the prescriptive analytics, we compared two Evolutionary Multi-objective Optimization (EMO) methods (NSGA-II and R-NSGA-II) when optimizing 100 new fabrics, aiming to simultaneously minimize the physical property predictive error and the distance of the optimized values when compared with the learned input space. The two EMO methods were applied to design of 100 new fabrics. Overall, the STR approach provided the best results for both prediction tasks, with Normalized Mean Absolute Error values that range from 4% (weft elasticity) to 11% (pilling) in terms of the fabric properties and a textile composition classification accuracy of 87% when adopting a small tolerance of 0.01 for predicting the percentages of six types of fibers (e.g., cotton). As for the prescriptive results, they favored the R-NSGA-II EMO method, which tends to select Pareto curves that are associated with an average 11% predictive error and 16% distance.

Towards a new generation of artificial‐intelligence‐based infrared atmospheric sounding interferometer retrievals of surface temperature: Part II – Assessment

AbstractWhen using neural networks (NNs), the lack of input information characterizing the radiative transfer can result in regional biases, especially when retrieving surface properties. In the Part I companion article we explored localization techniques in an attempt to help the NN adjust its behaviour to local conditions. In this article we analyze results from an image‐processing approach, the novel localized convolutional NN (CNN) model for the retrieval of surface temperature (TS) over a fixed domain using infrared atmospheric sounding interferometer (IASI) observations. An in‐depth evaluation is performed. The localized‐CNN architecture is a promising artificial intelligence solution that provides retrievals similar to, if not better than, those of the European Organisation for the Exploitation of Meteorological Satellites' PWLR3 retrieval algorithm that also uses IASI observations, collocated with microwave data too. This shows the benefits of localizing the CNN retrieval. This image‐processing retrieval scheme allows interpolation of the TS below the clouds, and thus a preliminary analysis of the cloud impact on the TS is performed. The possibility to estimate retrieval uncertainties is investigated, and a practical solution, based on the binning of the input space, is proposed for CNN architectures. The best strategy for a global‐scale retrieval is yet to be found for such an image‐processing scheme, but potential solutions and their respective advantages and disadvantages are discussed.

A Robustness-Based Confidence Measure for Hybrid System Falsification

Verification of hybrid systems is very challenging, if not impossible, due to their continuous dynamics that leads to infinite state space. As a countermeasure, falsification is usually applied to show that a specification does not hold, by searching for a falsifying input as a counterexample that refutes the specification. A falsification algorithm exploits the quantitative robust semantics of temporal specifications, which provides a numerical robustness that tells how robustly a specification holds or not, and uses it as a guide to explore the input space towards the direction of robustness descent—once negative robustness is observed, it indicates that a falsifying input is found. However, if a falsification algorithm does not return any falsifying input, a user is not sure whether the specification does indeed hold, or there exist counterexamples that the algorithm did not manage to reach. In this case, a measurement on how likely there indeed exists no counterexample in the input space is necessary for better understanding the safety of the system and deciding whether more budget should be allocated for the falsification. To this end, we propose a confidence measure that assesses the likelihood that the system is not falsifiable, i.e., how confident a user should be that a specification holds, given the fact that an algorithm has sampled a set of inputs but did not find any falsifying one. The confidence measure is defined in terms of a coverage criterion of the input space that assesses to which extent the whole input space is explored and a local area is exploited where low robustness is observed. Experiments on commonly-used falsification benchmarks show that our proposed confidence measure is reasonable and can distinguish different specifications.

Design of Granular Model: A Method Driven by Hyper-Box Iteration Granulation.

Recently, granular models have been highlighted in system modeling and applied to many fields since their outcomes are information granules supporting human-centric comprehension and reasoning. In this study, a design method of granular model driven by hyper-box iteration granulation is proposed. The method is composed mainly of partition of input space, formation of input hyper-box information granules with confidence levels, and granulation of output data corresponding to input hyper-box information granules. Among them, the formation of input hyper-box information granules is realized through performing the hyper-box iteration granulation algorithm governed by information granularity on input space, and the granulation of out data corresponding to input hyper-box information granules is completed by the improved principle of justifiable granularity to produce triangular fuzzy information granules. Compared with the existing granular models, the resulting one can yield the more accurate numeric and preferable granular outcomes simultaneously. Experiments completed on the synthetic and publicly available datasets demonstrate the superiority of the granular model designed by the proposed method at granular and numeric levels. Also, the impact of parameters involved in the proposed design method on the performance of ensuing granular model is explored.

Robust Discriminant Subspace Clustering With Adaptive Local Structure Embedding.

Unsupervised dimension reduction and clustering are frequently used as two separate steps to conduct clustering tasks in subspace. However, the two-step clustering methods may not necessarily reflect the cluster structure in the subspace. In addition, the existing subspace clustering methods do not consider the relationship between the low-dimensional representation and local structure in the input space. To address the above issues, we propose a robust discriminant subspace (RDS) clustering model with adaptive local structure embedding. Specifically, unlike the existing methods which incorporate dimension reduction and clustering via regularizer, thereby introducing extra parameters, RDS first integrates them into a unified matrix factorization (MF) model through theoretical proof. Furthermore, a similarity graph is constructed to learn the local structure. A constraint is imposed on the graph to guarantee that it has the same connected components with low-dimensional representation. In this spirit, the similarity graph serves as a tradeoff that adaptively balances the learning process between the low-dimensional space and the original space. Finally, RDS adopts the l2,1 -norm to measure the residual error, which enhances the robustness to noise. Using the property of the l2,1 -norm, RDS can be optimized efficiently without introducing more penalty terms. Experimental results on real-world benchmark datasets show that RDS can provide more interpretable clustering results and also outperform other state-of-the-art alternatives.

Global-local least-squares support vector machine (GLocal-LS-SVM).

This study introduces the global-local least-squares support vector machine (GLocal-LS-SVM), a novel machine learning algorithm that combines the strengths of localised and global learning. GLocal-LS-SVM addresses the challenges associated with decentralised data sources, large datasets, and input-space-related issues. The algorithm is a double-layer learning approach that employs multiple local LS-SVM models in the first layer and one global LS-SVM model in the second layer. The key idea behind GLocal-LS-SVM is to extract the most informative data points, known as support vectors, from each local region in the input space. Local LS-SVM models are developed for each region to identify the most contributing data points with the highest support values. The local support vectors are then merged at the final layer to form a reduced training set used to train the global model. We evaluated the performance of GLocal-LS-SVM using both synthetic and real-world datasets. Our results demonstrate that GLocal-LS-SVM achieves comparable or superior classification performance compared to standard LS-SVM and state-of-the-art models. In addition, our experiments show that GLocal-LS-SVM outperforms standard LS-SVM in terms of computational efficiency. For instance, on a training dataset of 9, 000 instances, the average training time for GLocal-LS-SVM was only 2% of the time required to train the LS-SVM model while maintaining classification performance. In summary, the GLocal-LS-SVM algorithm offers a promising solution to address the challenges associated with decentralised data sources and large datasets while maintaining high classification performance. Furthermore, its computational efficiency makes it a valuable tool for practical applications in various domains.

Input Distribution Coverage: Measuring Feature Interaction Adequacy in Neural Network Testing

Testing deep neural networks (DNNs) has garnered great interest in the recent years due to their use in many applications. Black-box test adequacy measures are useful for guiding the testing process in covering the input domain. However, the absence of input specifications makes it challenging to apply black-box test adequacy measures in DNN testing. The Input Distribution Coverage (IDC) framework addresses this challenge by using a variational autoencoder to learn a low dimensional latent representation of the input distribution, and then using that latent space as a coverage domain for testing. IDC applies combinatorial interaction testing on a partitioning of the latent space to measure test adequacy. Empirical evaluation demonstrates that IDC is cost-effective, capable of detecting feature diversity in test inputs, and more sensitive than prior work to test inputs generated using different DNN test generation methods. The findings demonstrate that IDC overcomes several limitations of white-box DNN coverage approaches by discounting coverage from unrealistic inputs and enabling the calculation of test adequacy metrics that capture the feature diversity present in the input space of DNNs.

An inverse mapping approach for process systems engineering using automatic differentiation and the implicit function theorem

AbstractThe objective in this work is to propose a novel approach for solving inverse problems from the output space to the input space using automatic differentiation coupled with the implicit function theorem and a path integration scheme. A common way of solving inverse problems in process systems engineering (PSE) and in science, technology, engineering and mathematics (STEM) in general is using nonlinear programming (NLP) tools, which may become computationally expensive when both the underlying process model complexity and dimensionality increase. The proposed approach takes advantage of recent advances in robust automatic differentiation packages to calculate the input space region by integration of governing differential equations of a given process. Such calculations are performed based on an initial starting point from the output space and are capable of maintaining accuracy and reducing computational time when compared to using NLP‐based approaches to obtain the inverse mapping. Two nonlinear case studies, namely a continuous stirred tank reactor (CSTR) and a membrane reactor for conversion of natural gas to value‐added chemicals are addressed using the proposed approach and compared against: (i) extensive (brute‐force) search for forward mapping and (ii) using NLP solvers for obtaining the inverse mapping. The obtained results show that the novel approach is in agreement with the typical approaches, while computational time and complexity are considerably reduced, indicating that a new direction for solving inverse problems is developed in this work.

Exploiting Active Subspaces for Geometric Optimization of Cavity-Stabilized Supersonic Flames

With the increased demand for hydrogen for carbon-neutral propulsion technologies, understanding the coupling effect of the fuel injection and cavity on flow characteristics is essential and an open issue for cavity-based hydrogen supersonic flame stabilization and optimizations. In this paper, the active subspace (AS) method is exploited to quantitatively reveal the intrinsic connections between the flow and flame stabilization with various fuel injection and cavity geometric parameters. A flame stabilization metric connecting the flame stabilization mode with nonreacting flows is proposed for the AS analysis and multiobjective optimizations. The results show that there exist one-dimensional active subspaces for key performance metrics, for example, the total pressure loss , residence time , and . The fuel injection pressure, cavity length-to-depth ratio, and cavity depth are three leading factors that govern the total pressure loss while the cavity residence time is dominated by the cavity depth. The proposed reasonably maps the flame stabilization mode with nonreacting flows. Using the response lines obtained by the AS, the tradeoff between and is quantitatively formulated with a clear visualization of candidate sets in the input space, demonstrating the advantages of the AS method for the combustor designs involving multiple objectives.

The power of typed affine decision structures: a case study

AbstractTADS are a novel, concise white-box representation of neural networks. In this paper, we apply TADS to the problem of neural network verification, using them to generate either proofs or concise error characterizations for desirable neural network properties. In a case study, we consider the robustness of neural networks to adversarial attacks, i.e., small changes to an input that drastically change a neural networks perception, and show that TADS can be used to provide precise diagnostics on how and where robustness errors a occur. We achieve these results by introducing Precondition Projection, a technique that yields a TADS describing network behavior precisely on a given subset of its input space, and combining it with PCA, a traditional, well-understood dimensionality reduction technique. We show that PCA is easily compatible with TADS. All analyses can be implemented in a straightforward fashion using the rich algebraic properties of TADS, demonstrating the utility of the TADS framework for neural network explainability and verification. While TADS do not yet scale as efficiently as state-of-the-art neural network verifiers, we show that, using PCA-based simplifications, they can still scale to medium-sized problems and yield concise explanations for potential errors that can be used for other purposes such as debugging a network or generating new training samples.

Few-shot short utterance speaker verification using meta-learning.

Short utterance speaker verification (SV) in the actual application is the task of accepting or rejecting the identity claim of a speaker based on a few enrollment utterances. Traditional methods have used deep neural networks to extract speaker representations for verification. Recently, several meta-learning approaches have learned a deep distance metric to distinguish speakers within meta-tasks. Among them, a prototypical network learns a metric space that may be used to compute the distance to the prototype center of speakers, in order to classify speaker identity. We use emphasized channel attention, propagation and aggregation in TDNN (ECAPA-TDNN) to implement the necessary function for the prototypical network, which is a nonlinear mapping from the input space to the metric space for either few-shot SV task. In addition, optimizing only for speakers in given meta-tasks cannot be sufficient to learn distinctive speaker features. Thus, we used an episodic training strategy, in which the classes of the support and query sets correspond to the classes of the entire training set, further improving the model performance. The proposed model outperforms comparison models on the VoxCeleb1 dataset and has a wide range of practical applications.

Machines Perceive Emotions: Identifying Affective States from Human Gait Using On-Body Smart Devices

Emotions are a crucial part of our daily lives, and they are defined as an organism’s complex reaction to significant objects or events, which include subjective and physiological components. Human emotion recognition has a variety of commercial applications, including intelligent automobile systems, affect-sensitive systems for customer service and contact centres, and the entertainment sector. In this work, we present a novel deep neural network of the Convolutional Neural Network - Bidirectional Gated Recurrent Unit (CNN-RNN) that can classify six basic emotions with an accuracy of above 95%. The deep model was trained on human gait data captured with body-mounted inertial sensors. We also proposed a reduction in the input space by utilizing 1D magnitudes of 3D accelerations and 3D angular velocities (maga^, magω^), which not only minimizes the computational complexity but also yields better classification accuracies. We compared the performance of the proposed model with existing methodologies and observed that the model outperforms the state-of-the-art.

Comparison of hierarchical clustering and neural network clustering: an analysis on precision dominance

A comparison of neural network clustering (NNC) and hierarchical clustering (HC) is conducted to assess computing dominance of two machine learning (ML) methods for classifying a populous data of large number of variables into clusters. An accurate clustering disposition is imperative to investigate assembly-influence of predictors on a system over a course of time. Moreover, categorically designated representation of variables can assist in scaling down a wide data without loss of essential system knowledge. For NNC, a self-organizing map (SOM)-training was used on a local aqua system to learn distribution and topology of variables in an input space. Ternary features of SOM; sample hits, neighbouring weight distances and weight planes were investigated to institute an optical inference of system’s structural attributes. For HC, constitutional partitioning of the data was executed through a coupled dissimilarity-linkage matrix operation. The validation of this approach was established through a higher value of cophenetic coefficient. Additionally, an HC-feature of stem-division was used to determine cluster boundaries. SOM visuals reported two locations’ samples for remarkable concentration analogy and presence of 4 extremely out of range concentration parameter from among 16 samples. NNC analysis also demonstrated that singular conduct of 18 independent components over a period of time can be comparably inquired through aggregate influence of 6 clusters containing these components. However, a precise number of 7 clusters was retrieved through HC analysis for segmentation of the system. Composing elements of each cluster were also distinctly provided. It is concluded that simultaneous categorization of system’s predictors (water components) and inputs (locations) through NNC and HC is valid to the precision probability of 0.8, as compared to data segmentation conducted with either of the methods exclusively. It is also established that cluster genesis through combined HC’s linkage and dissimilarity algorithms and NNC is more reliable than individual optical assessment of NNC, where varying a map size in SOM will alter the association of inputs’ weights to neurons, providing a new consolidation of clusters.

New perspective on reliability updating with equality information under line sampling

The observations provided by monitoring and sensor can be used to reevaluate the important structure performance for deepening the understanding of structural model and the reliability. Thus it is necessary to perform the reliability updating and estimate the posterior failure probability based on the collected observations. The posterior failure probability can be viewed as a corrected failure probability based on the onsite observation information and the prior failure probability, and it is more practical to reflect the current structure safety level than the prior failure probability. Current methods for estimating the posterior failure probability with equality observation information are to transform it into an inequality one by introducing an auxiliary variable, which enables it to be estimated by the structural reliability analysis methods. However, this is a challenging task because directly estimating the rare posterior failure probability often requires large number of samples. A new line sampling-based perspective on reliability updating is proposed in this paper, on which the posterior failure probability can be expressed by a more concise mathematical form and completed in the original random input space without introducing the auxiliary variable. Benefit from the high sampling efficiency of line sampling, the number of required samples is greatly reduced compared with Monte Carlo simulation. Additionally, the adaptive Kriging model is employed to identify the failure samples in the proposed reliability updating model, which can further reduce the computational cost. The efficiency and accuracy of the proposed method are demonstrated on the numerical and engineering examples.

Interpretable machine learning models for the estimation of seismic drifts in CLT buildings

An accurate estimation of drift demands is crucial for designing and assessing structures under seismic loads. Given the novelty of massive timber buildings, predictive models for the estimation of drifts in mid-to high-rise CLT structures are lacking, particularly in the form of simple models suitable for preliminary design evaluations or regional seismic assessments. In this paper, we present and compare several Machine Learning (ML) models for the estimation of peak inter-storey and roof drifts in multi-storey Cross-Laminated Timber (CLT) walled structures. The ML techniques used include: Multiple Linear Regression, Regression Trees, Random Forest, K-nearest Neighbor, and Support Vector Regression. To this end, 69 structures spanning mid-rise to tall timber buildings are subjected to a large collection of acceleration records and used to create the training and testing datasets. Different structural configurations and behaviour factors, related to the assumed energy dissipation capacity of the buildings, are considered. A diversity of feature selection techniques informs our choice of parameters to the reduced input space leading to a set of six most efficient features: the spectral acceleration at the building's fundamental period (Sa(T1)), the Peak Ground Velocity (PGV), tuning ratio (T1/Tm), behaviour factor (q), wall height (Hw), and the wall subdivision ratio (Wr). After verifying the high accuracy of our model predictions, the SHapley Additive exPlanation method (SHAP) is used to gain insight into the influence of key input features on the ML model outputs. Finally, our ML drift estimations are compared against previous proposals and design code assumptions, and the potential causes of disagreement are discussed.

Batch size: go big or go home? Counterintuitive improvement in medical autoencoders with smaller batch size.

Batch size is a key hyperparameter in training deep learning models. Conventional wisdom suggests larger batches produce improved model performance. Here we present evidence to the contrary, particularly when using autoencoders to derive meaningful latent spaces from data with spatially global similarities and local differences, such as electronic health records (EHR) and medical imaging. We investigate batch size effects in both EHR data from the Baltimore Longitudinal Study of Aging and medical imaging data from the multimodal brain tumor segmentation (BraTS) challenge. We train fully connected and convolutional autoencoders to compress the EHR and imaging input spaces, respectively, into 32-dimensional latent spaces via reconstruction losses for various batch sizes between 1 and 100. Under the same hyperparameter configurations, smaller batches improve loss performance for both datasets. Additionally, latent spaces derived by autoencoders with smaller batches capture more biologically meaningful information. Qualitatively, we visualize 2-dimensional projections of the latent spaces and find that with smaller batches the EHR network better separates the sex of the individuals, and the imaging network better captures the right-left laterality of tumors. Quantitatively, the analogous sex classification and laterality regressions using the latent spaces demonstrate statistically significant improvements in performance at smaller batch sizes. Finally, we find improved individual variation locally in visualizations of representative data reconstructions at lower batch sizes. Taken together, these results suggest that smaller batch sizes should be considered when designing autoencoders to extract meaningful latent spaces among EHR and medical imaging data driven by global similarities and local variation.

Equivariant Observers for Second-Order Systems on Matrix Lie Groups

This paper develops an equivariant symmetry for second order kinematic systems on matrix Lie groups and uses this symmetry for observer design. The state of a second order kinematic system on a matrix Lie group is naturally posed on the tangent bundle of the group with the inputs lying in the tangent of the tangent bundle known as the double tangent bundle. We provide a simple parameterization of both the tangent bundle state space and the input space (the fiber space of the double tangent bundle) and then introduce a semi-direct product group and group actions onto both the state and input spaces. We show that with the proposed group actions the second order kinematics are equivariant. An equivariant lift of the kinematics onto the symmetry group is derived and used to design nonlinear observers on the lifted state space using nonlinear constructive design techniques. The observer design is specialised to kinematics on groups that themselves admit a semi-direct product structure and include applications in rigid-body motion amongst others. A simulation based on an ideal hovercraft model verifies the performance of the proposed observer architecture.

Exploring the Relationship Between Center and Neighborhoods: Central Vector Oriented Self-Similarity Network for Hyperspectral Image Classification

To mine the spectral-spatial information of target pixel in hyperspectral image classification (HSIC), convolutional neural network (CNN)-based models widely adopt patch-based input pattern, where a patch represents its central pixel and the neighbor pixels play auxiliary roles in the classification process. However, compared to the central pixel, its neighbor pixels often have different contributions for classification. Although many existing patch-based CNNs could adaptively emphasize the spatial neighbor information, most of them ignore the latent relationship between the center pixel and its neighbor pixels. Moreover, efficient spectral-spatial feature extraction has been a difficult yet vital topic for HSIC. To address the mentioned problems, a central vector oriented self-similarity network (CVSSN) is proposed for HSIC. Specifically, based on two similarity measures, we firstly design an adaptive weight addition based spectral vector self-similarity module (AWA-SVSS) in input space and a Euclidean distance based feature vector self-similarity module (ED-FVSS) in feature space to fully mine the central vector oriented spatial relationships. Besides, a spectral-spatial information fusion module (SSIF) is formulated as a new pattern to fuse the central 1D spectral vector and the corresponding 3D patch for efficient spectral-spatial feature learning of the subsequent modules. Moreover, we implement a channel spatial separation convolution module (CSS-Conv) and a scale information complementary convolution module (SIC-Conv) for efficient spectral-spatial feature learning. Extensive experimental results on four popular HSI data sets demonstrate the effectiveness and efficiency of the proposed method compared with other state-of-the-art methods. The source code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/lms-07/CVSSN</uri>

Neighborhood-Adaptive Multi-Cluster Ranking for Deep Metric Learning

Deep metric learning methods generally concentrate on designing distance-based losses to learn sample embeddings, which tacitly presuppose the neighborhood structure around each sample (e.g., hypersphere for Euclidean distance). However, this supposition is overly optimistic: 1) visual data is often located on low-dimensional manifolds curved in high-dimensional space, and all regions of the manifold may hardly share the same local structures in the input space; 2) it is unlikely that the local structure in the output embedding space is as homogeneous as assumed due to the non-linearity of neural networks. Hence, simply characterizing sample embeddings while ignoring the respective neighborhood structures leads to limitations. To address this problem, this paper presents a Neighborhood-Adaptive Multi-cluster Ranking (NAMR) framework by leveraging the heterogeneity of local structures. Specifically, considering that indexing algorithms are usually required in large-scale retrieval, NAMR characterizes an image from two kinds of embeddings (i.e., sample embedding and structure embedding). The sample embedding can be trained using any distance-based loss, while the structure embedding representing the neighborhood structure can be jointly learned with the sample embedding in a self-supervised multi-cluster ranking manner. In this way, existing indexing algorithms can seamlessly support large-scale retrieval employing NAMR embeddings without any modifications. We evaluate the proposed model on five standard benchmarks, consistently and explicitly improving four baselines (especially the simplest triplet loss) and achieving state-of-the-art performance.