Active Sampling Strategy Research Articles

Visual object detection model integrating deep active learning

The perception of surrounding objects by vehicle autonomous driving is an important means to ensure traffic safety. Object detection models based on deep learning are widely used, but they require a large amount of labeled data for training. This paper proposes an active visual object detection model that uses Gaussian mixture distribution to estimate the uncertainty of unlabeled images, reducing the dependence of model training on labeled data. First, a mixed density network is used as the detection head, and the image features extracted by the deep neural network are used as input to estimate the probability distribution of the classification and positioning of the target prediction box. Secondly, the classification score value of the target prediction box is mapped to the probability space, and the classification uncertainty of the target is calculated using the edge uncertainty; the positioning uncertainty of the target is measured using the prediction box positioning variance. Finally, the most unstable samples are selected for labeling. Compared with other typical active learning sampling strategies on the VOC dataset, the proposed method achieves the best performance, and the 54%data annotation amount can reach the performance of YOLOX supervised learning 98.8%, saving nearly 45% of the data annotation amount.

Active Sampling: A Machine-Learning-Assisted Framework for Finite Population Inference with Optimal Subsamples

Data subsampling has become widely recognized as a tool to overcome computational and economic bottlenecks in analyzing massive datasets. We contribute to the development of adaptive design for estimation of finite population characteristics, using active learning and adaptive importance sampling. We propose an active sampling strategy that iterates between estimation and data collection with optimal subsamples, guided by machine learning predictions on yet unseen data. The method is illustrated on virtual simulation-based safety assessment of advanced driver assistance systems. Substantial performance improvements are demonstrated compared to traditional sampling methods.

Surrogate based design space exploration and exploitation for an efficient airfoil optimization under uncertainties using transition models

The pursuit of sustainable, zero-emission air travel is heavily dependent on the creation of energy-efficient aircraft. Key strategies for achieving this sustainability in aviation include reducing fuel consumption through low-drag designs harnessing laminar flow. However, designing aircraft with laminar flow characteristics is complex due to their sensitivity to environmental and operational factors. This study tackles the challenge of developing energy-efficient aircraft by using computational fluid dynamics models and sophisticated optimization techniques that account for uncertainty. Our approach demonstrates the effectiveness of surrogate-based optimization and uncertainty quantification in optimizing airfoil drag for a natural laminar airfoil (NLF) design. We use surrogate models, trained with data from detailed airfoil simulations, which include a boundary layer code coupled with a linear stability method and a newly developed transition transport model. Transition location predicted using transition models facilitate an accurate drag prediction used in the optimization process. The accuracy of these surrogate models is enhanced through active sampling strategies. Our robust optimization method considers uncertainties in environmental and operational conditions, offering a deeper insight into their effects on crucial design parameters. Unlike traditional deterministic aerodynamic design optimization, our findings highlight the efficacy and precision of uncertainty-based optimization in achieving robust NLF airfoil designs over large (exploration mode) and small (exploitation mode) design spaces. Investigating design space parameterization based on the size of design variables reveals significant differences in optimal airfoil configurations. The optimized designs we propose favor delayed transition, in contrast to deterministic designs which often result in significant loss of laminarity when facing uncertainties. This study represents a significant advancement in aerospace engineering, providing a practical and effective methodology for creating energy-efficient airfoil designs. The application of these advanced optimization and uncertainty quantification techniques shows great potential for the wider field of aerospace engineering, paving the way for more resilient and robust aircraft designs.

Contrastive-Active Transfer Learning-Based Real-Time Adaptive Assessment Method for Power System Transient Stability.

The transient stability assessment based on machine learning faces challenges such as sample data imbalance and poor generalization. To address these problems, this paper proposes an intelligent enhancement method for real-time adaptive assessment of transient stability. In the offline phase, a convolutional neural network (CNN) is used as the base classifier. A model training method based on contrastive learning is introduced, aiming to increase the spatial distance between positive and negative samples in the mapping space. This approach effectively improves the accuracy of the model in recognizing unbalanced samples. In the online phase, when real data with different distribution characteristics from the offline data are encountered, an active transfer strategy is employed to update the model. New system samples are obtained through instance transfer from the original system, and an active sampling strategy considering uncertainty is designed to continuously select high-value samples from the new system for labeling. The model parameters are then updated by fine-tuning. This approach drastically reduces the cost of updating while improving the model's adaptability. Experiments on the IEEE39-node system verify the effectiveness of the proposed method.

Evaluating Solid Phase Adsorption Toxin Tracking (SPATT) for passive monitoring of per- and polyfluoroalkyl substances (PFAS) with Ion Mobility Spectrometry-Mass Spectrometry (IMS-MS)

Detection and monitoring of per- and polyfluoroalkyl substances (PFAS) in aquatic environments has become an increasingly higher priority of regulatory agencies as public concern for human intake of these chemicals continues to grow. While many methods utilize active sampling strategies (“grab samples”) for precise PFAS quantitation, here we evaluate the efficacy of low-cost passive sampling devices (Solid Phase Adsorption Toxin Tracking, or SPATTs) for spatial and temporal PFAS assessment of aquatic systems. For this study, passive samplers were initially deployed in North Carolina along the Cape Fear River during the summer and fall of 2016 and 2017. These were originally intended for the detection of microcystins and monitoring potentially harmful algal blooms, though this period also coincided with occurrences of PFAS discharge from a local fluorochemical manufacturer into the river. Additional samplers were then deployed in 2022 to evaluate changes in PFAS fingerprint and abundances. Assessment of PFAS showed legacy compounds were observed across almost all sampling sites over all 3 years (PFHxS, PFOS, PFHxA, etc.), while emerging replacement PFAS (e.g., Nafion byproducts) were predominantly localized downstream from the manufacturer. Furthermore, samplers deployed downstream from the manufacturer in 2022 noted sharp decreases in observed signal for replacement PFAS in comparison to samplers deployed in 2016 and 2017, indicating mitigation and remediation efforts in the area were able to reduce localized fluorochemical contamination.

Dynamically configured physics-informed neural network in topology optimization applications

The integration of physics-informed neural network (PINN) and topology optimization (TO) is an attractive issue because PINN can avoid the prohibitive data acquisition of solving forward problems compared to traditional machine learning. To enhance the efficiency of this integrated optimization framework, a dynamically configured PINN-based topology optimization (DCPINN-TO) method is proposed in conjunction with an active sampling strategy. The DCPINN comprises two sub-networks with distinct training costs, capable of dynamically adjusting the trainable parameters based on the optimization state of TO. Moreover, the active sampling strategy selectively samples collocations based on the pseudo-densities, which can significantly reduce training costs by decreasing the number of input collocations. Additionally, the Gaussian integral is used to calculate the strain energy of elements to decouple the mapping of the material at the collocations. The proposed method is extended to various scenarios, including those with high resolution, multiple loads, and displacement constraints. Its efficiency and generalization are validated by several illustrative examples. Furthermore, the accuracy of DCPINN versus finite element analysis-based TO (FEA-TO) was also investigated.

Uncertainty Quantification of Data-driven Quality Prediction Model For Realizing the Active Sampling Inspection of Mechanical Properties in Steel Production

Pre-production quality defect inspection is a crucial step in industrial manufacturing, and many traditional inspection strategies suffer from inefficiency issues. This is especially true for tasks such as mechanical performance testing of steel products, which involve time-consuming processes like offline sampling, specimen preparation, and testing. The inspection volume significantly impacts the production cycle, inventory, yield, and labor costs. Constructing a data-driven model for predicting product quality and implementing proactive sampling inspection based on the prediction results is an appealing solution. However, the prediction uncertainty of data-driven models poses a challenging problem that needs to be addressed. This paper proposes an active quality inspection approach for steel products based on the uncertainty quantification in the predictive model for mechanical performance. The objective is to reduce both the sampling frequency and the omission rate on the production site. First, an ensemble model based on improved lower and upper bound estimation is established for interval prediction of mechanical performance. The uncertainty of the specific value prediction model is quantitatively estimated using interval probability distributions. Then, a predictive model for the mechanical performance failure probability is built based on the prediction interval size and probability distribution. By determining an appropriate probability threshold, the trade-off between prediction accuracy and defect detection accuracy (recall rate) is balanced, enabling the establishment of an active sampling strategy. Finally, this functionality is integrated into the manufacturing execution system of a steel factory, realizing a mechanical performance inspection approach based on proactive sampling. The proposed approach is validated using real production datasets. When the probability threshold is set to 30%, the prediction accuracy and recall rate for failure mechanical performance samples are 75% and 100%, respectively. Meanwhile, the sampling rate is only 5.33%, while controlling the risk of omission. This represents a 50% reduction in sampling rate compared to the inspection rules commonly used in actual production. The overall efficiency of product quality inspection is improved, and inspection costs are reduced.

An edge-aware graph autoencoder trained on scale-imbalanced data for traveling salesman problems

In recent years, there has been a notable surge in research on machine learning techniques for combinatorial optimization. It has been shown that learning-based methods outperform traditional heuristics and mathematical solvers on the Traveling Salesman Problem (TSP) in terms of both performance and computational efficiency. However, most learning-based TSP solvers are primarily designed for fixed-scale TSP instances, and also require a large number of training samples to achieve optimal performance. To fill this gap, this work proposes a data-driven graph representation learning method for solving TSPs with various numbers of cities. Specifically, we formulate the TSP as a link prediction task and propose an edge-aware graph autoencoder (EdgeGAE) model that can solve TSPs by learning from various-scale samples with an imbalanced distribution. A residual gated encoder is trained to learn latent edge embeddings, followed by an edge-centered decoder to output link predictions in an end-to-end manner. Furthermore, we introduce an active sampling strategy into the training process to improve the model’s generalization capability in large-scale scenarios. To investigate the model’s practical applicability, we generate a scale-imbalanced dataset comprising 50,000 TSP instances ranging from 50 to 500 cities. The experimental results demonstrate that the proposed edge-aware graph autoencoder model achieves a highly competitive performance among state-of-the-art graph learning-based approaches in solving TSPs with various scales, implying its remarkable potential in dealing with practical optimization challenges.

A patch distribution-based active learning method for multiple instance Alzheimer's disease diagnosis

Medical data, particularly the complex brain imaging structures, acquisition presents significant difficulties and high diagnostic expenses, resulting in a scarcity of the trainable samples in the real-world scenarios. To overcome this limitation, we present an active learning-based sampling strategy that selects the most informative samples from the unlabeled candidate sample pool for expert annotation, leading to high classification performance with a reduced number of training samples. This study adopts a patch-level perspective and introduces a multi-instance learning framework for Alzheimer's Disease diagnosis. Initially, a patch pre-selection module is designed to identify pathology-prone regions while excluding background areas and irrelevant information. Subsequently, an inner-patch local attention mechanism block and an outer-patch global attention mechanism block are developed to enhance the extraction of discriminative local and global information by the network model. Finally, an active learning sampling strategy is devised to minimize the costs associated with data acquisition and expert annotation in medical domain. The effectiveness of the proposed network framework and active learning strategy was validated through four sets of control experiments on the ADNI dataset.

An active inference model of hierarchical action understanding, learning and imitation.

We advance a novel active inference model of the cognitive processing that underlies the acquisition of a hierarchical action repertoire and its use for observation, understanding and imitation. We illustrate the model in four simulations of a tennis learner who observes a teacher performing tennis shots, forms hierarchical representations of the observed actions, and imitates them. Our simulations show that the agent's oculomotor activity implements an active information sampling strategy that permits inferring the kinematic aspects of the observed movement, which lie at the lowest level of the action hierarchy. In turn, this low-level kinematic inference supports higher-level inferences about deeper aspects of the observed actions: proximal goals and intentions. Finally, the inferred action representations can steer imitative responses, but interfere with the execution of different actions. Our simulations show that hierarchical active inference provides a unified account of action observation, understanding, learning and imitation and helps explain the neurobiological underpinnings of visuomotor cognition, including the multiple routes for action understanding in the dorsal and ventral streams and mirror mechanisms.

Practical uncertainty quantification for space-dependent inverse heat conduction problem via ensemble physics-informed neural networks

Inverse heat conduction problems (IHCPs) are problems of estimating unknown quantities of interest (QoIs) of the heat conduction with given temperature observations. The challenge of IHCPs is that it is usually ill-posed since the observations are noisy, and the estimations of QoIs are generally not unique or unstable, especially when there are unknown spatially varying QoIs. In this study, an ensemble physics-informed neural network (E-PINN) is proposed to handle function estimation and uncertainty quantification of space-dependent IHCPs. The distinctive characteristics of E-PINN are ensemble learning and adversarial training (AT). Compared with other data-driven UQ approaches, the suggested method is more than straightforward to implement and also achieves high-quality uncertainty estimates of the QoI. Furthermore, an adaptive active sampling (AS) strategy based on the uncertainty estimates from E-PINNs is also proposed to improve the accuracy of material field inversion problems. Finally, the proposed method is validated through several numerical experiments of IHCPs.

Active Sensing in Bees Through Antennal Movements Is Independent of Odor Molecule.

When sampling odors, many insects are moving their antennae in a complex but repeatable fashion. Previous studies with bees have tracked antennal movements in only two dimensions, with a low sampling rate and with relatively few odorants. A detailed characterization of the multimodal antennal movement patterns as function of olfactory stimuli is thus wanted. The aim of this study is to test for a relationship between the scanning movements and the properties of the odor molecule. We tracked several key locations on the antennae of bumblebees at high frequency and in three dimensions while stimulating the insect with puffs of 11 common odorants released in a low-speed continuous flow. Water and paraffin were used as negative controls. Movement analysis was done with the neural network Deeplabcut. Bees use a stereotypical oscillating motion of their antennae when smelling odors, similar across all bees, independently of the identity of the odors and hence their diffusivity and vapor pressure. The variability in the movement amplitude among odors is as large as between individuals. The main type of oscillation at low frequencies and large amplitude is triggered by the presence of an odor and is in line with previous work, as is the speed of movement. The second oscillation mode at higher frequencies and smaller amplitudes is constantly present. Antennae are quickly deployed when a stimulus is perceived, decorrelate their movement trajectories rapidly, and oscillate vertically with a large amplitude and laterally with a smaller one. The cone of airspace thus sampled was identified through the 3D understanding of the motion patterns. The amplitude and speed of antennal scanning movements seem to be function of the internal state of the animal, rather than determined by the odorant. Still, bees display an active olfactory sampling strategy. First, they deploy their antennae when perceiving an odor. Second, fast vertical scanning movements further increase the odorant capture rate. Finally, lateral movements might enhance the likelihood to locate the source of odor, similarly to the lateral scanning movement of insects at odor plume boundaries.

Active federated transfer algorithm based on broad learning for fault diagnosis

Federated learning (FL) guaranteeing data privacy is of great interest in decentralized fault diagnosis. However, limited research attention has been paid to the dynamic domain-shift issue due to varying working conditions. This paper proposes an active federated transfer algorithm based on broad learning to address the domain shift issue in FL. First, a central server dispatches a global model to the source clients for collaborative modeling. Subsequently, the global model is initialized with a federated averaging strategy. Next, the initialized global model is used to annotate emerging signals from the target clients based on an active sampling strategy proposed. Finally, an asynchronous update scheme is designed to adapt the global model to the target domain. The performance of the AFTBL algorithm is validated with three datasets, including 24 centralized- and decentralized-modeling tasks. The computational results indicate that the proposed algorithm is more accurate and efficient than the prevalent algorithms.

Interactive active learning for fairness with partial group label

The rapid development of AI technologies has found numerous applications across various domains in human society. Ensuring fairness and preventing discrimination are critical considerations in the development of AI models. However, incomplete information often hinders the complete collection of sensitive attributes in real-world applications, primarily due to the high cost and potential privacy violations associated with such data collection. Label reconstruction through building another learner on sensitive attributes is a common approach to address this issue. However, existing methods focus solely on improving the prediction accuracy of the sensitive learner as a separate model, while ignoring the disparity between its accuracy and the fairness of the base model. To bridge this gap, this paper proposes an interactive learning framework that aims to optimize the sensitive learner while considering the fairness of the base learner. Furthermore, a new active sampling strategy is developed to select the most valuable data for the sensitive learner regarding the fairness of the base model. The effectiveness of our proposed method in improving model fairness is demonstrated through comprehensive evaluations conducted on various datasets and fairness criteria.

3D Single-pixel imaging with active sampling patterns and learning based reconstruction

Single-pixel imaging (SPI) is significant for applications constrained by transmission bandwidth or lighting band, where 3D SPI can be further realized through capturing signals carrying depth. The sampling strategy and the reconstruction algorithm are the key issues of SPI. Traditionally, random patterns are often adopted for sampling, but this blindly passive strategy requires a high sampling rate, and even so, it is difficult to develop a reconstruction algorithm that can maintain high accuracy and robustness. In this paper, an active and flexible sampling strategy is proposed to make up for the shortcoming of the single-pixel detector (SPD) that lacks spatial resolution capability. The sampling rate and sampling order can be controlled actively and flexibly by designing the sampling patterns according to the needs. Furthermore, two approaches of deep learning are proposed to improve the quality of reconstruction, where a large amount of training data for 3D SPI are generated by a virtual system with graphic software. The ability of deep learning to reconstruct desired information under low sampling rates is analyzed. Abundant experiments verify that our method improves the precision of SPI even if the sampling rate is very low, which has the potential to be extended in similar systems or applied in practice.

An Efficient Framework for Constructing Speech Emotion Corpus Based on Integrated Active Learning Strategies

Speech emotion recognition has been developed rapidly in recent decades because of the appearance of machine learning. Nevertheless, lack of corpus remains a significant issue. For actual speech emotion corpus construction, many professional actors are required to perform voices with various emotions in specific scenes. In the process of data labelling, since the number of samples of different emotion categories is extremely imbalanced, it is difficult to efficiently label the samples. Hence, we proposed an integrated active learning sampling strategy and designed an efficient framework for constructing speech emotion corpora in order to address the problems presented above. Comparing experiments with other active learning algorithms on 13 datasets, our method was shown to improve sampling efficiency. In addition, it is able to select small category samples to be labelled with preference in imbalanced datasets. During the actual corpus construction experiments, our method can prioritize selecting small class emotion samples. As even when the amount of labelled data is less than 50%, the accuracy rate still can reach 90%. This greatly enhances the efficiency of constructing the speech emotion corpus and fills in the gaps.

Active learning-assisted semi-supervised learning for fault detection and diagnostics with imbalanced dataset

Data-driven Fault Detection and Diagnostics (FDD) methods often assume that sufficient labeled samples are class-balanced and faulty classes in testing are precedent or seen previously during model training. When monitoring a large fleet of assets at scale, these assumptions may be violated: (I) only a limited number of samples can be manually labeled due to constraints of time and/or cost; (II) most of the samples collected in the engineering systems are under normal conditions, leading to a highly imbalanced class distribution and a biased prediction model. This work presents a robust and cost-effective FDD framework that integrates active learning and semi-supervised learning methods to detect both known and unknown failure modes iteratively. This framework allows to strategically select the samples to be annotated from a fully unlabeled dataset, while labeling cost is minimal. Specifically, a novel graph-based semi-supervised classifier with adaptive graph construction is developed to predict labels with imbalanced data and detect novel classes. We designed a multi-criteria active learning sampling strategy to select the most informative samples from unlabeled data in order to query minimal number of labels for classification. We tested the framework and algorithms in three synthetic datasets and one real-world dataset of vehicle air intake systems, and demonstrated the superior performance compared to the state-of-the-art methods for fleet-level FDD.

Environmental monitoring study of pesticide contamination in Denmark through honey bee colonies using APIStrip-based sampling

Due to their extensive use in both agricultural and non-agricultural applications, pesticides are a major source of environmental contamination. Honey bee colonies are proven sentinels of these and other contaminants, as they come into contact with them during their foraging activities. However, active sampling strategies involve a negative impact on these organisms and, in most cases, the need of analyzing multiple heterogeneous matrices. Conversely, the APIStrip-based passive sampling is innocuous for the bees and allows for long-term monitorings using the same colony. The versatility of the sorbent Tenax, included in the APIStrip composition, ensures that comprehensive information regarding the contaminants inside the beehive will be obtained in one single matrix. In the present study, 180 APIStrips were placed in nine apiaries distributed in Denmark throughout a six-month sampling period (10 subsequent samplings, April to September 2020). Seventy-five pesticide residues were detected (out of a 428-pesticide scope), boscalid and azoxystrobin being the most frequently detected compounds. There were significant variations in the findings of the sampling sites in terms of number of detections, pesticide diversity and average concentration. A relative indicator of the potential risk of pesticide exposure for the honey bees was calculated for each sampling site. The evolution of pesticide detections over the sampling periods, as well as the individual tendencies of selected pesticides, is herein described. The findings of this large-scale monitoring were compared to the ones obtained in a previous Danish, APIStrip-based pilot monitoring program in 2019. Samples of honey and wax were also analyzed and compared to the APIStrip findings.

Dissipation and cross-contamination of miticides in apiculture. Evaluation by APIStrip-based sampling

The active substances coumaphos, tau-fluvalinate and amitraz are among the most commonly employed synthetic miticides to control varroa infestations in apiculture. These compounds can persist inside the beehive matrices and can be detected long time after their application. The present study describes the application of a new passive sampling methodology to assess the dissipation of these miticides as well as the cross-contamination in neighboring beehives. The APIStrips are a recently developed sampling device based on the sorbent Tenax, which shows a remarkable versatility for the sorption of molecules onto its surface. This avoids the need of actively sampling apicultural matrices such as living bees, wax or reserves (honey and pollen), therefore allowing to obtain representative information of the contamination in the beehive environment in one single matrix. The results show that the amitraz-based treatments have the fastest dissipation rate (half-life of 11–14 days), whereas tau-fluvalinate and coumaphos remain inside the beehive environment for longer time periods, with a half-life up to 39 days. In the present study, tau-fluvalinate originated an intense cross-contamination, as opposed to coumaphos and amitraz. This study also demonstrates the contribution of drifting forager bees in the pesticide cross-contamination phenomena. Moreover, the sampling of adult living bees has been compared to the APIStrip-based sampling, and the experimental results show that the latter is more effective and consistent than traditional active sampling strategies. The active substances included in this study do not migrate to the honey from the treated colonies in significant amounts.

Active Sampling Based Polynomial-Chaos–Kriging Model for Orbital Uncertainty Propagation

Propagating uncertainties usually requires repeated evaluation of the subject model for a large number of different input parameters. Surrogate models have been widely used to avoid this computationally intensive uncertainty propagation by replacing the original model by an easy-to-evaluate function model. In this Paper, the polynomial chaos based Kriging is used as such a surrogate model for orbital uncertainty propagation. The polynomial chaos represents the global trend of the uncertainty distribution, while the Kriging describes the local variations. Such a combination can provide a more precise surrogate model than individual polynomial chaos or ordinary Kriging representation. To further enhance the accuracy, a new active sampling strategy is proposed to incrementally build and improve the polynomial chaos based Kriging model. This new modeling scheme only requires a small number of sampling points while achieving close performance to the Monte Carlo based propagation. It is also more accurate than the random sampling based Kriging model. Three orbital uncertainty propagation examples are used to demonstrate the effectiveness of the proposed surrogate model.