Drift Detector Research Articles

Accelerating deep neural network learning using data stream methodology

This paper introduces a novel machine learning approach, called BBATDD (Boosting-Based Algorithm Trained with Drift Detector), that departs from traditional epoch-based methods by adopting a data stream generation strategy from the original dataset. The proposed method is versatile, applicable to a broad range of deep structures and other machine learning architectures. The key to its effectiveness is the conversion of original data into a streaming sequence, which is then fed into the neural network structure. To monitor the learning process and prevent overfitting, a drift detector, commonly used in stream data mining, is employed. Additionally, two methods, ELB and ENT, are introduced to prioritize problematic data elements during the learning process, enhancing overall model performance. Extensive simulations on benchmark datasets (MNIST and CIFAR) validate that BBATDD achieves higher accuracy with significantly fewer batches, without sacrificing generalization capability. For instance, when applied to the MNIST dataset and processing 10 000 batches while comparing loss function values, our approach demonstrates a remarkable 45.6% improvement.

Dynamic Classification Ensembles for Handling Imbalanced Multiclass Drifted Data Streams

Machine learning models often encounter significant difficulties when dealing with multiclass imbalanced data streams in nonstationary environments. These challenges can lead to biased and unreliable predictions, which ultimately impact the overall performance of the models. To address these issues, we propose an innovative approach that integrates dynamic ensemble selection, an adaptive technique for managing imbalanced multiclass data streams, with a concept drift detector for recognizing stream changes and the K-nearest neighbor (KNN) algorithm to tackle issues related to class overlap. The primary objective was to improve the classification of imbalanced multiclass drifted data streams. The adaptive oversampling method generates synthetic samples to mitigate the issues associated with imbalanced data streams. This method utilizes KNN to ensure that the generated samples do not overlap. To handle incoming data streams, a drift detector assists in deciding whether to retain the existing classifiers or create a new one. Dynamic Ensemble Selection (DES) was utilized to select the most appropriate classifier for incoming data, aiming to optimize the performance of the classification task. The proposed method offers an effective solution for achieving an accurate and resilient classification in the context of imbalanced multiclass drifted data streams. To evaluate the effectiveness of our proposal, we conducted experiments on a variety of datasets, including benchmark datasets, real application stream datasets, and synthetic data streams. The experimental results demonstrate the superiority of our contribution in addressing the challenges posed by imbalanced multiclass drifted data streams.

A rehearsal framework for computational efficiency in online continual learning

In the realm of online continual learning, models are expected to adapt to an ever-changing environment. One of the most persistent hurdles in this adaptation is the mitigation of a phenomenon called "Catastrophic Forgetting" (CF). This critical condition occurs when models trained on non-identically distributed data lose performance in previously learned tasks. Rehearsal methods, leveraging the ability to replay older samples, aim to address this challenge by incorporating a buffer of past training samples. However, the absence of known task boundaries complicates the adaptation of current CF mitigation methods. This paper proposes a method attuned to data stream characteristics and online model performance in a resource-constrained environment. The number of training iterations and learning rate emerges as crucial hyperparameters, impacting the efficacy and efficiency of online continual learning. Up to this point, we propose a combination of Experience Replay methodologies, a Drift Detector, and various training convergence policies, specially tailored for scenarios with unknown task boundaries. Experimental results demonstrate the effectiveness of our approach, maintaining or enhancing performance compared to baseline methods, while significantly improving computational efficiency.

Scattered high-energy synchrotron radiation at theKARA visible-light diagnostic beamline.

To characterize an electron beam, visible synchrotron light is often used and dedicated beamlines at synchrotron sources are becoming a more common feature as instruments and methods for the diagnostics are, along with the accelerators, further developed. At KARA (Karlsruhe Research Accelerator), such a beamline exists and is based on a typical infrared/visible-light configuration. From experience at such beamlines no significant radiation was expected (dose rates larger than 0.5 µSv h-1). This was found not to be the case and a higher dose was measured which fortunately could be shielded to an acceptable level with 0.3 mm of aluminium foil or 2.0 mm of Pyrex glass. The presence of this radiation led to further investigation by both experiment and calculation. A custom setup using a silicon drift detector for energy-dispersive spectroscopy (Ketek GmbH) and attenuation experiments showed the radiation to be predominantly copper K-shell fluorescence and is confirmed by calculation. The measurement of secondary radiation from scattering of synchrotron and other radiation, and its calculation, is important for radiation protection, and, although a lot of experience exists and methods for radiation protection are well established, changes in machine, beamlines and experiments mean a constant appraisal is needed.

FBDD: feature-based drift detector for batch processing data

FBDD: feature-based drift detector for batch processing data

Enhancing Performances of the VOXES Bragg Spectrometer for XES Investigations

Utilizing a dispersive crystal for X-ray Emission Spectroscopy (XES) significantly enhances the energy resolution when compared with spectroscopy performed with just silicon drift detectors. This high resolution is particularly valuable for studying metals, as it offers essential insights into their electronic structures and chemical environments. Conducting such experiments in the laboratory, as opposed to synchrotron light sources, presents challenges due to the reduced intensities of X-ray tubes and, consequently, low signal rates, with the effect of increasing the acquisition time. In this study, we demonstrate that XES spectra can be acquired within a few hours for a CuNiZn metallic sample alloy while still maintaining a good energy resolution and a large dynamic range. This is achieved with the VOXES spectrometer, developed at INFN National Laboratories of Frascati (LNF), along with a background reduction procedure that enhances the signal from emission lines under study. This study is a showcase for improving the efficiency of XES in tabletop setup experiments.

Assessment of performance rates on the elemental comparison of small and irregular glass fragments using µ-XRF and LIBS

This study describes a systematic assessment of the performance rates when analyzing small and irregular glass fragments using micro-X-ray Fluorescence Spectrometry (µ-XRF) and Laser Induced Breakdown Spectroscopy (LIBS). One hundred glass fragments were collected from the inner and outer panes of a vehicle windshield to assess the false exclusion rates. Additionally, 100 glass fragments originating from different vehicle windshields were used to evaluate the discrimination capabilities. To compare the effects of fragment size on the performance rates, half of the collected fragments were small (longest length between 0.4 mm and < 1 mm, and thickness greater than 0.4 mm for LIBS and 0.1 mm for μ-XRF), and the other half were full-thickness fragments (2 mm and greater). The study shows that precision deteriorates for small/irregular fragments and comparison items must have a similar size, shape, and thickness to minimize error rates. Thus, comparisons between full-thickness and small/irregular fragments should be avoided, regardless of the analytical method. Although this general concept is well known for µ-XRF, this effect was not previously reported as a concern for LIBS. Moreover, this study provides new sampling and comparison recommendations when using modern silicon drift detectors (SDD) and reduced fragment size. Using a 3 s (3 %RSD) comparison interval reduces the false exclusion rates to < 12 % for µ-XRF, and to < 4 % for LIBS when using either a 3 s or 4 s (3 % RSD) criterion. At least 4 known fragments are recommended for full thickness fragments and 6 to 9 known fragments for the small/irregular comparisons.

Comparing the output of measured and GEANT4 simulated X-ray tubes

Monte Carlo simulation codes that model particle interactions with matter, including GEANT4, can be used to support the development of novel X-ray tubes and instruments for applications in industry and research. Based on our knowledge, simulations of the output of transmission-geometry X-ray tubes have not yet been conducted in GEANT4. Simulations of reflection-geometry tubes are scarce, with a limited number of studies investigating the capability and accuracy of GEANT4 in this domain. Therefore, the aim of this work is to benchmark GEANT4 for the characterisation of X-ray tubes with industry applications, using for the first time a range of X-ray tubes geometries and high voltages. In this work, GEANT4 is benchmarked against experimental measurements performed at the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia, with both transmission and reflection X-ray tubes. High voltages between 17 kV and 42.5 kV for tungsten, molybdenum, gold, and silver targets are investigated. The measurements have been performed with a MXR packaged X-ray tube (reflection) and Magnum X-ray tubes (transmission) and a fast silicon drift detector (SDD). Our study shows that GEANT4 predicts the bremsstrahlung continuum output of transmission and reflection X-ray tubes, in agreement with the experimental measurements. The Penelope physics models, distributed with GEANT4, and the GEANT4 X-ray fluorescence data libraries ANSTO HF, based on the Hartree-Fock approach, exhibit the best agreement against experimental measurements. Depending on the high voltage, target and X-ray line agreement was found to vary between 10% and 300%. Differences between GEANT4 and experimental data are more significant for lower voltage X-ray tube measurements and are most likely due to limitations in the ionisation electron cross-section and/or atomic relaxation data libraries used. As next steps, it is recommended to validate GEANT4 against other, independent experimental measurements to corroborate the results of this work. In addition, it would be useful to repeat the same study with other general purpose Monte Carlo codes for radiation physics to compare their capability against GEANT4 in this application domain.

Anode capacitance measurement of silicon drift detectors in operating conditions

We describe a technique for measuring with high accuracy and precision the capacitance of the anode of a semiconductor drift detector (SDD) fully biased in operating conditions and connected to the front-end electronics. The proposed method does not require the use of a probe station, making it suitable for low-cost and easy-to-setup application environments. We have performed measurements on four SDDs with different geometries and production processes, observing a total capacitance ranging from 20 fF to 75 fF, depending on the bias conditions. Different components of the anode capacitance with respect to the front cathode and the adjacent drift cathodes have been measured and commented as well. Precision in the capacitance measurements better than 0.1 fF has been obtained. These results can be effectively used for the optimization of the front-end electronics for SDDs to achieve minimum electronic noise.

Design and 3D Electrical Simulations for a Controllable Equal-Gap Large-Area Silicon Drift Detector.

In this study, a controllable equal-gap large-area silicon drift detector (L-SDD) is designed. The surface leakage current is reduced by reducing the SiO2-Si interface through the new controllable equal-gap design. The design of the equal gap also solves the problem whereby the gap widens due to the larger detector size in the previous SDD design, which leads to a large invalid area of the detector. In this paper, a spiral hexagonal equal-gap L-SDD of 1 cm radius is selected for design calculation, and we implement 3D modeling and simulation of the device. The simulation results show that the internal potential gradient distribution of the L-SDD is uniform and forms a drift electric field, with the direction of electron drift pointing towards the collecting anode. The L-SDD has an excellent electron drift channel inside, and this article also analyzes the electrical performance of the drift channel to verify the correctness of the design method of the L-SDD.

A comprehensive analysis of concept drift locality in data streams

Adapting to drifting data streams is a significant challenge in online learning. Concept drift must be detected for effective model adaptation to evolving data properties. Concept drift can impact the data distribution entirely or partially, which makes it difficult for drift detectors to accurately identify the concept drift. Despite the numerous concept drift detectors in the literature, standardized procedures and benchmarks for comprehensive evaluation considering the locality of the drift are lacking. We present a novel categorization of concept drift based on its locality and scale. A systematic approach leads to a test bed of 2760 data stream benchmarks, reflecting various difficulty levels following our proposed categorization. We conduct a comparative assessment of 9 state-of-the-art drift detectors across diverse difficulties, highlighting their strengths and weaknesses for future research. We examine how drift locality influences the classifier performance and propose strategies for different drift categories to minimize the recovery time. Lastly, we provide lessons learned and recommendations for future concept drift research. Our benchmark data streams and experiments are publicly available at https://github.com/gabrieljaguiar/locality-concept-drift.

On the forensic relevance of tattoos: distinguishing black inks with energy dispersive spectroscopy and backscattered scanning electron microscopy.

This study aimed to analyze black tattoo inks by means of energy dispersive spectroscopy and backscattered scanning electron microscopy. The sample consisted of five types of commercial tattoo pigments of the black colour (Easy Glow™, Electric Ink™, Iron Works™, Master Ink™, and Viper™). An Energy Dispersive Spectroscopy (EDS) detector (Silicon Drift Detector - SDD - type) attached to a Scanning Electron Microscope (SEM) device (Tescan Vega3 LMU, Libusina, Czech Republic) was used. X-ray characteristic signs were detected for each tattoo ink in an interval between 0 and 2.5keV. The electron acceleration potential in the microscope was 15keV. Two regions were analyzed for each sample (n = 10). On each region, a micrography of backscattered electrons (BSE) was obtained. Means and standard deviations (SD) of the weight percentages (Wt%) were calculated. C and O were predominant, with a mean O/C ratio between 2.69 and 2.74 Wt%. Electric Ink and Master Ink were the most similar pigments, while Easy Glow was the most distinctive - with agglomerates of Al that had a concentration 25 × higher than other specimens. Other compounds detected in the sample were Cl and Cu. EDS and SEM were efficient to distinguish black tattoo inks. These are our preliminary outcomes on the use of EDS and SEM to analyze black tattoo inks. Thus, careful interpretation is necessary to avoid rash applications in human identification practice.

A versatile beamline for soft x-ray reflectivity, absorption, and fluorescence measurements at Indus-2 synchrotron source.

A versatile beamline for performing reflectivity, fluorescence, and absorption experiments in the soft x-ray region of 100-1500eV is commissioned on a bending magnet port of the Indus-2 synchrotron source. A high vacuum 2-axis reflectometer with x, y, and z sample scanning stages is installed. This reflectometer is used to measure the reflectivity of large samples up to 300mm in length and 5kg in weight. This feature is useful for characterizing x-ray optical elements, such as mirrors, gratings, and multilayers. A flange mounted silicon drift detector is installed in the downstream of the reflectometer for soft x-ray fluorescence measurements. The soft x-ray absorption measurements are carried out in the total electron yield and partial fluorescence yield modes. Integration of three different experimental techniques in the experimental station makes the beamline versatile for materials science applications as it provides structural, chemical, and electronic state information by performing the required experiments in an identical environment. The beamline uses a varied line spacing plane grating monochromator and gives a high flux (∼109 to 1011 photon/s) with a moderate resolution (λ/Δλ ~1000-5000). A three-mirror-based higher harmonic setup is incorporated to get rid of harmonics and to get a high spectral purity monochromatic beam with less than 0.1% harmonic content. In the present article, the beamline optical scheme, mechanical configuration, and details of the experimental setups are presented, along with a few representative results of each experimental mode.

The XGIS instrument on-board THESEUS: detector principle and read-out electronics

The functionality and experimental performance characterization of the latest four channel version of ORION ASIC, a very low noise multichip read out and processing electronics customized for the X and Gamma Imaging Spectrometer (XGIS) instrument onboard the Transient High-Energy Sky and Early Universe Surveyor (THESEUS) mission, is presented. XGIS is a set of two coded-masked wide field deep sky cameras using monolithic SDDs (Silicon Drift Detectors) and CsI:Tl (Thallium doped-Cesium Iodide) scintillator-based X-γ ray detectors. This paper highlights the design, working principle and the expected performances of the XGIS, on a small scale 2×2 prototype. Furthermore, the evolution timeline of different versions of ORION with detailed performance observations and analysis for spectroscopic resolution, electronic noise and the operational linear energy ranges of both X and the γ processors of the four-pixel ASIC version bonded to a 2×2 SDD array are emphasized. Each 2×2 SDD array element is electrically and dimensionally equivalent to single elements of the THESEUS 8×8 SDD array.

LSTMDD: an optimized LSTM-based drift detector for concept drift in dynamic cloud computing.

This study aims to investigate the problem of concept drift in cloud computing and emphasizes the importance of early detection for enabling optimum resource utilization and offering an effective solution. The analysis includes synthetic and real-world cloud datasets, stressing the need for appropriate drift detectors tailored to the cloud domain. A modified version of Long Short-Term Memory (LSTM) called the LSTM Drift Detector (LSTMDD) is proposed and compared with other top drift detection techniques using prediction error as the primary evaluation metric. LSTMDD is optimized to improve performance in detecting anomalies in non-Gaussian distributed cloud environments. The experiments show that LSTMDD outperforms other methods for gradual and sudden drift in the cloud domain. The findings suggest that machine learning techniques such as LSTMDD could be a promising approach to addressing the problem of concept drift in cloud computing, leading to more efficient resource allocation and improved performance.

Investigation of the chemical composition of the domestic apple tree fruits by x-ray fluorescence analysis

Introduction. Apple fruits are widely used in the food industry and as non-pharmacopoeial medicinal raw materials. The value of this plant is due to the presence of a large number of biologically active compounds, namely: phenols, flavonoids and tannins. Despite the abundance of information on the chemical composition of this raw material, the study of the elemental composition of apple fruits remains relevant for potential use in the treatment of element deficiency conditions. Objective: to study the content of chemical elements using X-ray fluorescence analysis in three different parts of apple fruits – in the exocarp, mesocarp and endocarp for four varieties, as well as to identify the nitrate content and measure acidity for these apple varieties. Material and methods. Apple tree fruits for research were harvested in the Moscow region. After preliminary ashing of the samples, the content of chemical elements was determined by X-ray fluorescence analysis with a silicon drift detector. pH and nitrate ion concentrations were measured using an EXPERT-001 pH meter. Results. Using X-ray fluorescence analysis (XFA), the elemental composition of four varieties of domestic apple fruits was studied for various parts of the fruit. An increased content of lead was noted in the exocarp of apples of the Spartan variety. A study of the content of nitrate ions and pH was also carried out. Conclusion. The chemical composition of domestic apple fruits was studied using X-ray fluorescence. The content of chemical elements was different in different parts of the fruit. The lowest content of chemical elements was determined in the mesocarp of the fruit. It is also worth noting the high lead content in the exocarp of Spartan apples.

Online Machine Learning from Non-stationary Data Streams in the Presence of Concept Drift and Class Imbalance: A Systematic Review

In IoT environment applications generate continuous non-stationary data streams with in-built problems of concept drift and class imbalance which cause classifier performance degradation. The imbalanced data affects the classifier during concept detection and concept adaptation. In general, for concept detection, a separate mechanism is added in parallel with the classifier to detect the concept drift called a drift detector. For concept adaptation, the classifier updates itself or trains a new classifier to replace the older one. In case, the data stream faces a class imbalance issue, the classifier may not properly adapt to the latest concept. In this survey, we study how the existing work addresses the issues of class imbalance and concept drift while learning from nonstationarydata streams. We further highlight the limitation of existing work and challenges caused by other factors of class imbalance alongwith concept drift in data stream classification. Results of our survey found that, out of 1110 studies, by using our inclusion and exclusion criteria, we were able to narrow the pool of articles down to 35 that directly addressed our study objectives. The study found that issues such as multiple concept drift types, dynamic class imbalance ratio, and multi-class imbalance in presence of concept drift are still open for further research. We also observed that, while major research efforts have been dedicated to resolving concept drift and class imbalance, not much attention has been given to with-in-class imbalance, rear examples, and borderline instances when they exist with concept drift in multi-class data. This paper concludes with some suggested future directions.

BenchMFC: A benchmark dataset for trustworthy malware family classification under concept drift

Concept drift poses a critical challenge in deploying machine learning models to mitigate practical malware threats. It refers to the phenomenon that the distribution of test data changes over time, gradually deviating from the original training data and degrading model performance. A promising direction for addressing concept drift is to detect drift samples and then retrain the model. However, this field currently lacks a unified, well-curated, and comprehensive benchmark, which often leads to unfair comparisons and inconclusive outcomes. To improve the evaluation and advance further, this paper presents a new Benchmark dataset for trustworthy Malware Family Classification (BenchMFC), which includes 223 K samples of 526 families that evolve over years. BenchMFC provides clear family, packer, and timestamp tags for each sample, it thus can support research on three types of malware concept drift: 1) unseen families, 2) packed families, and 3) evolved families. To collect unpacked family samples from large-scale candidates, we introduce a novel crowdsourcing malware annotation pipeline, which unifies packing detection and family annotation as a consensus inference problem to prevent costly packing detection. Moreover, we provide two case studies to illustrate the application of BenchMFC in 1) concept drift detection and 2) model retraining. The first case demonstrates the impact of three types of malware concept drift and compares nine notable concept drift detectors. The results show that existing detectors have their own advantages in dealing with different types of malware concept drift, and there is still room for improvement in malware concept drift detection. The second case explores how static feature-based machine learning operates on packed samples when retraining a model. The experiments illustrate that packers do preserve some kind of signals that appear to be “effective” for machine learning models, but the robustness of these signals requires further research. BenchMFC has been released to the community at https://github.com/crowdma/benchmfc.

Confirming X-ray parametric down conversion by time–energy correlation

We present measurements of X-ray Parametric Down Conversion at the Advanced Photon Source synchrotron facility. Using an incoming pump beam at 22 keV, we observe the simultaneous, elastic emission of down-converted photon pairs generated in a diamond crystal. The pairs are detected using high count rate silicon drift detectors with low noise. Production by down-conversion is confirmed by measuring time–energy correlations in the detector signal, where photon pairs within an energy window ranging from 10 to 12 keV are only observed at short time differences. By systematically varying the crystal misalignment and detector positions, we obtain results that are consistent with the constant total of the down-converted signal. Our maximum rate of observed pairs was 130/h, corresponding to a conversion efficiency for the down-conversion process of 5.3±0.5×10−13.

Templated Synthesis of CsPbBr3 Nanowire Arrays Toward Low Current Drift and Stable X-ray Detector

All-inorganic halide perovskite CsPbBr3 with superior environmental stability, remarkable radiation hardness and great defect tolerance has emerged as a competitive material in radiation detection. However, the operation stability of CsPbBr3...