Data-centric Approach Research Articles

Towards a Better Performance in Facial Expression Recognition: A Data-Centric Approach.

Facial expression is the best evidence of our emotions. Its automatic detection and recognition are key for robotics, medicine, healthcare, education, psychology, sociology, marketing, security, entertainment, and many other areas. Experiments in the lab environments achieve high performance. However, in real-world scenarios, it is challenging. Deep learning techniques based on convolutional neural networks (CNNs) have shown great potential. Most of the research is exclusively model-centric, searching for better algorithms to improve recognition. However, progress is insufficient. Despite being the main resource for automatic learning, few works focus on improving the quality of datasets. We propose a novel data-centric method to tackle misclassification, a problem commonly encountered in facial image datasets. The strategy is to progressively refine the dataset by successive training of a CNN model that is fixed. Each training uses the facial images corresponding to the correct predictions of the previous training, allowing the model to capture more distinctive features of each class of facial expression. After the last training, the model performs automatic reclassification of the whole dataset. Unlike other similar work, our method avoids modifying, deleting, or augmenting facial images. Experimental results on three representative datasets proved the effectiveness of the proposed method, improving the validation accuracy by 20.45%, 14.47%, and 39.66%, for FER2013, NHFI, and AffectNet, respectively. The recognition rates on the reclassified versions of these datasets are 86.71%, 70.44%, and 89.17% and become state-of-the-art performance.

MLProbs: A Data-Centric Pipeline for Better Multiple Sequence Alignment.

In this paper, we explore using the data-centric approach to tackle the Multiple Sequence Alignment (MSA) construction problem. Unlike the algorithm-centric approach, which reduces the construction problem to a combinatorial optimization problem based on an abstract mathematical model, the data-centric approach explores using classification models trained from existing benchmark data to guide the construction. We identified two simple classifications to help us choose a better alignment tool and determine whether and how much to carry out realignment. We show that shallow machine-learning algorithms suffice to train sensitive models for these classifications. Based on these models, we implemented a new multiple sequence alignment pipeline, called MLProbs. Compared with 10 other popular alignment tools over four benchmark databases (namely, BAliBASE, OXBench, OXBench-X and SABMark), MLProbs consistently gives the highest TC score. More importantly, MLProbs shows non-trivial improvement for protein families with low similarity; in particular, when evaluated against the 1,356 protein families with similarity ≤ 50%, MLProbs achieves a TC score of 56.93, while the next best three tools are in the range of [55.41, 55.91] (increased by more than 1.8%). We also compared the performance of MLProbs and other MSA tools in two real-life applications - Phylogenetic Tree Construction Analysis and Protein Secondary Structure Prediction - and MLProbs also had the best performance. In our study, we used only shallow machine-learning algorithms to train our models. It would be interesting to study whether deep-learning methods can help make further improvements, so we suggest some possible research directions in the conclusion section.

Smart Detection System for Driver Distraction: Enhanced Support Vector Machine classifier using Analytical Hierarchy Process technique

Driving distraction is the most significant source of motor-vehicle accidents. In 2018 in United States there were 2,800 mortality cases and over 400,000 were injured in distracted driver collision according to Center for Disease Control. Various research techniques have been proposed in literature based on machine learning to complex deep learning to learn the driving pattern and identify the distraction cause to driver. As the model are employed in a real time environment a delay in smallest internal of time can lead to a major accident. The objective of the research is to decrease the computational time of this model without compromising the accuracy, to achieve this, we employed a data-centric approach with enhancement of Support Vector Machine, resulting in significant improvements. In our attempt to increase the accuracy and reduce the training time, we experiment on various feature extraction and dimensionality reduction techniques on Improved SVM. While testing the average time taken by one image to check and respond with the classification whether the driver is distracted or not, our model was able to detect one image in 0.53 seconds which is significantly low.

Phänomenotechnik und Noumenologie

AbstractMy paper examines the potential of Gaston Bachelard’s concept ofphenomenotechniquethrough a critique of its historical reception and an analysis of its epistemological implications. Vis-a-vis constructivist and pragmatist interpretations, I argue that the core issue at stake in this concept is not just to regard scientific knowledge as ‘fabricated’ through technical means, but also to understand the radical difference that lies between the experience of a world given to and measured by scientists, and one that is ontologically expanded by the conceptual creations of science. In his reflections on the ‘phenomenotechnical realisation’ of noumena in microphysics, mathematics and chemistry, Bachelard not only anticipated a new epistemic modality which has become eminent now in computerised, data-centric approaches to nature and culture; he also became one of the first philosophers to explore the human experience of a reality inherently built from, and no longer just described by mathematical tools and models.

Efficiency rectification of predictive machine learning models by data centric optimization

There are two ways to optimize the performance of machine learning models, one is model centric, and another is data centric. Model centric approaches sometimes cannot yield expected results, so researchers and developers are now focusing on data centric optimization. Main steps of data centric approaches are – noise detection and correction, effective feature selection, feature transformation and data augmentation in case of limited datasets. Be it image, audio, text or numerical data, the mentioned approaches are always beneficial irrespective of model architecture. This paper is devoted to developing new preprocessing techniques for numerical data. The developed algorithm is tested with different artificial and real datasets for linear regression, k-means clustering, SVM classifier and deep learning networks. The algorithm is easy to implement and computationally less expensive. Experimental results show that proposed methods work well to improve model efficiency.

Machine-learned control-oriented flow estimation for multi-actuator multi-sensor systems exemplified for the fluidic pinball

We propose the first machine-learned control-oriented flow estimation for multiple-input, multiple-output plants. The starting point is constant actuation with open-loop actuation commands leading to a database with simultaneously recorded actuation commands, sensor signals and flow fields. A key enabler is an estimator input vector comprising sensor signals and actuation commands. The mapping from the sensor signals and actuation commands to the flow fields is realized in an analytically simple, data-centric and general nonlinear approach. The analytically simple estimator generalizes linear stochastic estimation for actuation commands. The data-centric approach yields flow fields from estimator inputs by interpolating from the database – similar to Loiseau, Noack & Brunton (J. Fluid Mech., vol. 844, 2018, pp. 459–490) for unforced flow. The interpolation is performed with$k$nearest neighbours ($k$NN). The general global nonlinear mapping from inputs to flow fields is obtained from a deep neural network (DNN) via an iterative training approach. The estimator comparison is performed for the fluidic pinball plant, which is a multiple-input, multiple-output wake control benchmark (Denget al.,J. Fluid Mech., vol. 884, 2020, A37) featuring rich dynamics under steady controls. We conclude that the machine learning methods clearly outperform the linear model. The performance of$k$NN and DNN estimators are comparable for periodic dynamics. Yet the DNN performs consistently better when the flow is chaotic. Moreover, a thorough comparison regarding the complexity, computational cost and prediction accuracy is presented to demonstrate the relative merits of each estimator. The proposed method can be generalized for closed-loop flow control plants.

Carbonate lithofacies classification in optical microscopy: a data-centric approach using augmentation and GAN synthetic images

Carbonate lithofacies classification in optical microscopy: a data-centric approach using augmentation and GAN synthetic images

Clinical Assessment of Customized 3D-Printed Wrist Orthoses

Plaster casting has been effective in fracture rehabilitation due to its immobility capability. Thermoplastic casting gains some efficiency and convenience from clinical processing but fitting and immobility are two major tradeoffs. Three-dimensional (3D)-printed braces have demonstrated improvements in innovation, customization, and appearance, but one may question their effectiveness in rehabilitation. Via a science-based data-centric approach, we evaluated the feasibility of customizing 3D-printed braces for wrist fractures after surgery. First, we established a framework for determining a proper timeframe for reliable pressure measurement for plaster casts, devised a method to evaluate the immobilization capability of the plaster casts and customized braces, and assessed changes in immobilization capability due to the clearance adjustment of customized braces. Second, we compared the customized wrist braces and plaster casts in immobilization capability and functional effectiveness in clinical trials. Thirty-three patients were recruited (9, control; 24, experimental). Both the Disability of Arm, Shoulder, and Head, and the Manual Ability Measure surveys showed significant differences between the control and experimental groups after brace wearing time frames of 2, 4, 8, and 12 weeks, and 6 and 12 months. We found that customized braces can match plaster casts in immobilization ability and can promote effective functional recovery.

Dealing With Inaccurate Sensor Data in the Context of Mobile Crowdsensing and mHealth.

The technological capabilities and ubiquity of smart mobile devices favor the combined utilization of Ecological Momentary Assessments (EMA) and Mobile Crowdsensing (MCS). In the healthcare domain, this combination particularly enables the collection of ecologically valid and longitudinal data. Furthermore, the context in which these data are collected can be captured through the use of smartphone sensors as well as externally connected sensors. The TrackYourTinnitus (TYT) mobile platform uses these concepts to collect the user's individual subjective perception of tinnitus as well as an objective environmental sound level. However, the sound level data in the TYT database are subject to several possible sensor errors and therefore do not allow a meaningful interpretation in terms of correlation with tinnitus symptoms. To this end, a data-centric approach based on Principal Component Analysis (PCA) is proposed in this paper to cleanse MCS mHealth data sets from erroneous sensor data. To further improve the approach, additional information (i.e., responses to the EMA questionnaire) is considered in the PCA and a prior check for constant values is performed. To demonstrate the practical feasibility of the approach, in addition to TYT data, where it is generally unknown which sensor measurements are actually erroneous, a simulation with generated data was designed and performed to evaluate the performance of the approach with different parameters based on different quality metrics. The results obtained show that the approach is able to detect an average of 29.02% of the errors, with an average false-positive rate of 14.11%, yielding an overall error reduction of 22.74%.

What the PHIRI federated research infrastructure has achieved so far?

Abstract PHIRI infrastructure follows a federated approach that is governed following the European Interoperability Framework. The vision of PHIRI is to create an infrastructure for individual level data processing following the privacy-by-design principle in a data-centric approach. As a basis to legal interoperability and compliance with the GDPR, the queries or algorithms are moved to the data instead of moving the data. So far, the PHIRI technological developments have focused on a client-server architecture. In this architecture a Coordinator Hub, the server, is in charge of orchestrating the deployment of the data-centric analysis solutions, in the form of R and Python scripts, that will be later executed in the partner nodes (data hubs), the clients. To perform the orchestration the Coordinator Hub encapsulates the scripts in software containers, using Docker images; all the outputs are published in Zenodo. The software containers are then deployed manually from Zenodo in the partner nodes and executed by its IT specialists using their own individual level data - the software containers have represented the technical interoperability layer. The data used on each partner has been previously adapted to a common data model (CDM) and the quality of the dataset has been assessed against the data model by each partner -this has represented the semantic interoperability layer. Finally, the outputs of the analysis's execution are aggregated data that are sent back to the Coordinator Hub to perform a comparative analysis. This stepwise approach has been tested in various research questions promoted by a leading researcher and agreed by the partner nodes who act as data hubs. A help-desk services and a developer's forum and a help-desk service have been set up to ease the implementation and deployment of the research queries - these both have represented the organisational interoperability layer

The ASR Post-Processor Performance Challenges of BackTranScription (BTS): Data-Centric and Model-Centric Approaches

Training an automatic speech recognition (ASR) post-processor based on sequence-to-sequence (S2S) requires a parallel pair (e.g., speech recognition result and human post-edited sentence) to construct the dataset, which demands a great amount of human labor. BackTransScription (BTS) proposes a data-building method to mitigate the limitations of the existing S2S based ASR post-processors, which can automatically generate vast amounts of training datasets, reducing time and cost in data construction. Despite the emergence of this novel approach, the BTS-based ASR post-processor still has research challenges and is mostly untested in diverse approaches. In this study, we highlight these challenges through detailed experiments by analyzing the data-centric approach (i.e., controlling the amount of data without model alteration) and the model-centric approach (i.e., model modification). In other words, we attempt to point out problems with the current trend of research pursuing a model-centric approach and alert against ignoring the importance of the data. Our experiment results show that the data-centric approach outperformed the model-centric approach by +11.69, +17.64, and +19.02 in the F1-score, BLEU, and GLEU tests.

DATA-CENTRIC DEEP LEARNING METHOD FOR PULMONARY NODULE DETECTION

Lung cancer is one of the most serious cancer-related diseases in Vietnam and all over the world. Early detection of lung nodules can help to increase the survival rate of lung cancer patients. Computer-aided diagnosis (CAD) systems are proposed in the literature for early detection of lung nodules. However, most of the current CAD systems are based on the building of high-quality machine learning models for a fixed dataset rather than taking into account the dataset properties which are very important for the lung cancer diagnosis. In this paper, we follow the direction of data-centric approach for lung nodule detection by proposing a data-centric method to improve detection performance of lung nodules on CT scans. Our method takes into account the dataset-specific features (nodule sizes and aspect ratios) to train detection models as well as add more training data from local Vietnamese hospital. We experiment our method on the three widely used object detection networks (Faster R-CNN, YOLOv3 and RetinaNet). The experimental results show that our proposed method improves detection sensitivity of these object detection models up to 4.24%.

Similarity of materials and data-quality assessment by fingerprinting

Identifying similar materials (i.e., those sharing a certain property or feature) requires interoperable data of high quality. It also requires means to measure similarity. We demonstrate how a spectral fingerprint as a descriptor, combined with a similarity metric, can be used for establishing quantitative relationships between materials data, thereby serving multiple purposes. This concerns, for instance, the identification of materials exhibiting electronic properties similar to a chosen one. The same approach can be used for assessing uncertainty in data that potentially come from different sources. Selected examples show how to quantify differences between measured optical spectra or the impact of methodology and computational parameters on calculated properties, like the density of states or excitonic spectra. Moreover, combining the same fingerprint with a clustering approach allows us to explore materials spaces in view of finding (un)expected trends or patterns. In all cases, we provide physical reasoning behind the findings of the automatized assessment of data.Impact statementTo predict novel materials with desired properties, data-centric approaches are in the process of becoming an additional fundament of materials research. Prerequisite for their success are well-curated data. Ideally, one can make use of multiple data collections. Bringing data from different sources together, poses challenges on their interoperability which are routed in two out of the 4V of Big Data. These are the uncertainty of data quality (veracity) and the heterogeneity in form and meaning of the data (variety). To overcome this barrier, universal and interpretable measures must be established, which quantify differences between data that are supposed to have the same meaning. Here, we show how a spectral fingerprint in combination with a similarity metric can be used for assessing spectral properties of materials. Our approach allows for tracing back in computed as well as measured data, differences stemming from various aspects. It thus paves the way for automatized data-quality assessment toward interoperability. Based on this, in turn, materials exhibiting similar features can be identified.Graphical abstract

DIGITAL TRANSFORMATION THROUGH DATA-DRIVEN AND DATA-CENTRIC APPROACHES WITH ARTIFICIAL INTELLIGENCE

Data-driven approach emphasises on extracting information and insights from data into deducing the next course of action. Whilst data-centric approach focuses on exploiting domain knowledge based on existing data and their relationships into producing near accurate output. They serve as the key to the shipping industry’s transformation which defines innovations and the use of digital technologies to deliver processes, product quality, and organisational improvements. Digital transformation plays an important role in ensuring the industry remains competitive and efficient. This work reviews the potentials, and challenges of adopting the approaches. Case studies are explored at the UTM Marine Technology centre on the design, simulation and analysis tools, intelligent towing tank, AI and IoT technologies applications. They present significant impact on the existing processes and productivities with focus on the data collection, modelling, validation, and optimization. Therefore, the findings open a new venture into expanding the use cases, technologies, and processes improvements.

Defense against adversarial attacks on deep convolutional neural networks through nonlocal denoising

<p><span lang="EN-US">Despite substantial advances in network architecture performance, the susceptibility of adversarial attacks makes deep learning challenging to implement in safety-critical applications. This paper proposes a data-centric approach to addressing this problem. A nonlocal denoising method with different luminance values has been used to generate adversarial examples from the Modified National Institute of Standards and Technology database (MNIST) and Canadian Institute for Advanced Research (CIFAR-10) data sets. Under perturbation, the method provided absolute accuracy improvements of up to 9.3% in the MNIST data set and 13% in the CIFAR-10 data set. Training using transformed images with higher luminance values increases the robustness of the classifier. We have shown that transfer learning is disadvantageous for adversarial machine learning. The results indicate that simple adversarial examples can improve resilience and make deep learning easier to apply in various applications.</span></p>

Task reformulation and data-centric approach for Twitter medication name extraction.

Automatically extracting medication names from tweets is challenging in the real world. There are many tweets; however, only a small proportion mentions medications. Thus, datasets are usually highly imbalanced. Moreover, the length of tweets is very short, which makes it hard to recognize medication names from the limited context. This paper proposes a data-centric approach for extracting medications in the BioCreative VII Track 3 (Automatic Extraction of Medication Names in Tweets). Our approach formulates the sequence labeling problem as text entailment and question–answer tasks. As a result, without using the dictionary and ensemble method, our single model achieved a Strict F1 of 0.77 (the official baseline system is 0.758, and the average performance of participants is 0.696). Moreover, combining the dictionary filtering and ensemble method achieved a Strict F1 of 0.804 and had the highest performance for all participants. Furthermore, domain-specific and task-specific pretrained language models, as well as data-centric approaches, are proposed for further improvements. Database URL https://competitions.codalab.org/competitions/23925 and https://biocreative.bioinformatics.udel.edu/tasks/biocreative-vii/track-3/

A data-centric weak supervised learning for highway traffic incident detection

Using the data from loop detector sensors for near-real-time detection of traffic incidents on highways is crucial to averting major traffic congestion. While recent supervised machine learning methods offer solutions to incident detection by leveraging human-labeled incident data, the false alarm rate is often too high to be used in practice. Specifically, the inconsistency in the human labeling of the incidents significantly affects the performance of supervised learning models. To that end, we focus on a data-centric approach to improve the accuracy and reduce the false alarm rate of traffic incident detection on highways. We develop a weak supervised learning workflow to generate high-quality training labels for the incident data without the ground truth labels, and we use those generated labels in the supervised learning setup for final detection. This approach comprises three stages. First, we introduce a data preprocessing and curation pipeline that processes traffic sensor data to generate high-quality training data through leveraging labeling functions, which can be domain knowledge-related or simple heuristic rules. Second, we evaluate the training data generated by weak supervision using three supervised learning models—random forest, k-nearest neighbors, and a support vector machine ensemble—and long short-term memory classifiers. The results show that the accuracy of all of the models improves significantly after using the training data generated by weak supervision. Third, we develop an online real-time incident detection approach that leverages the model ensemble and the uncertainty quantification while detecting incidents. Overall, we show that our proposed weak supervised learning workflow achieves a high incident detection rate (0.90) and low false alarm rate (0.08).

Enhancing convolutional neural network predictions of electrocardiograms with left ventricular dysfunction using a novel sub-waveform representation

BackgroundElectrocardiogram (ECG) deep learning (DL) has promise to improve the outcomes of patients with cardiovascular abnormalities. In ECG DL, researchers often use convolutional neural networks (CNNs) and traditionally use the full duration of raw ECG waveforms that create redundancies in feature learning and result in inaccurate predictions with large uncertainties.ObjectiveFor enhancing these predictions, we introduced a sub-waveform representation that leverages the rhythmic pattern of ECG waveforms (data-centric approach) rather than changing the CNN architecture (model-centric approach).ResultsWe applied the proposed representation to a population with 92,446 patients to identify left ventricular dysfunction. We found that the sub-waveform representation increases the performance metrics compared to the full-waveform representation. We observed a 2% increase for area under the receiver operating characteristic curve and 10% increase for area under the precision-recall curve. We also carefully examined three reliability components of explainability, interpretability, and fairness. We provided an explanation for enhancements obtained by heartbeat alignment mechanism. By developing a new scoring system, we interpreted the clinical relevance of ECG features and showed that sub-waveform representation further pushes the scores towards clinical predictions. Finally, we showed that the new representation significantly reduces prediction uncertainties within subgroups that contributes to individual fairness.ConclusionWe expect that this added control over the granularity of ECG data will improve the DL modeling for new artificial intelligence technologies in the cardiovascular space.

Bagging and Multilayer Perceptron Hybrid Intelligence Models Predicting the Swelling Potential of Soil

Seasonal variations of the moisture content of fine-grained soils may result in the accumulation of significant volumetric strains, which may affect the stability of geotechnical infrastructure. Predicting the swelling potential of soil may therefore be crucial to geotechnical infrastructure integrity. This research presents a series of machine learning models namely Gaussian Process (GP), Multilayer Perceptron (MLP), and Bagging-MLP neural network models that developed for the prediction of the soil’s swelling potential. A data driven approach based was based on site specific data from residual soils in the Mong Cai-Van Don expressway in Vietnam. The analysis involved simple soil index indicators i.e the particle size distribution, Atterberg limits, optimum dry density, in order to determne the swelling potential of the soil. The experimental database was then used to train and develop Bagging and Multilayer Perceptron Hybrid Intelligence models for the prediction of the soil’s swelling potential. The results show that the model performance in this area of geotechnical engineering performed with the highest prediction accuracy obtained using the Bagging-MLP model. The results of the study show promising steps towards a data centric approach in order to support geotechnical design.

Energy Saving with Zero Hot Spots: A Novel Power Control Approach for Sustainable and Stable Data Centers

Data centers with high energy consumption have become a threat to urban sustainability on electric energy. In contrast, hot spots in a data center are another threat to server stability, which leads to unsafe data storage and service provisioning to urban lives. However, state-of-the-art works cannot ensure sustainability and stability together because they fail to consider them holistically. For example, some existing works eliminate the hot spots by increasing cooling power, which results in lower sustainability. In contrast, others reduce energy consumption by saving the cooling power, which harms stability. Therefore, to balance the hot spot elimination and energy saving through power control remains challenging, this paper proposes a novel power control approach for energy saving with zero hot spots in data centers. Power control works when hot spots appear, or consumed energy is excess. Specifically, we formulated a total consumption minimization problem to characterize and analyze the optimal set points for power control, where the number of hot spots is zero and the energy consumption is low. Adding the interactional penalty models can determine the power control approach when the objective function obtains the optimal solution. We propose a Modified Differential Evolution algorithm (MDE) to solve the function quickly and accurately. It adopts adaptive parameters to reduce the computing time. Meanwhile, it avoids optimal local solutions by changing mutation operations. Further, simulation experiments using our optimal solution demonstrate that energy consumption saves about 13% on average with zero hot spots, compared with three typical approaches.