DA Techniques Research Articles

ERKT-Net: Implementing Efficient and Robust Knowledge Distillation for Remote Sensing Image Classification

The classification of Remote Sensing Images (RSIs) poses a significant challenge due to the presence of clustered ground objects and noisy backgrounds. While many approaches rely on scaling models to enhance accuracy, the deployment of RSI classifiers often requires substantial computational and storage resources, thus necessitating the use of lightweight algorithms. In this paper, we present an efficient and robust knowledge transfer network named ERKT-Net, which is designed to provide a lightweight yet accurate Convolutional Neural Network (CNN) classifier. This method utilizes innovative yet simple concepts to better accommodate the inherent nature of RSIs, thereby significantly improving the efficiency and robustness of traditional Knowledge Distillation (KD) techniques developed on ImageNet-1K. We evaluated ERKT-Net on three benchmark RSI datasets and found that it demonstrated superior accuracy and a very compact volume compared to 40 other advanced methods published between 2020 and 2023. On the most challenging NWPU45 dataset, ERKT-Net outperformed other KD-based methods with a maximum Overall Accuracy (OA) value of 22.4%. Using the same criterion, it also surpassed the first-ranked multi-model method with a minimum OA value of 0.7 but presented at least an 82% reduction in parameters. Furthermore, ablation experiments indicated that our training approach has significantly improved the efficiency and robustness of classic DA techniques. Notably, it can reduce the time expenditure in the distillation phase by at least 80%, with a slight sacrifice in accuracy. This study confirmed that a logit-based KD technique can be more efficient and effective in developing lightweight yet accurate classifiers, especially when the method is tailored to the inherent characteristics of RSIs.

Natural Language Processing Approaches to Text Data Augmentation: A Computational Linguistic Analysis

In the context of Natural Language Processing (NLP) tasks, problems such as insufficient or skewed data are frequently encountered. One practical solution to this problem is to generate additional textual data. Text Data Augmentation (TDA) refers to small changes made to accessible text at the character, word, or sentence level to generate synthetic data that is subsequently inserted into data loaders to train the model. By producing synthetic data, models can learn from a larger range of instances and, hence, enhance their resilience and generalization skills. Despite the fact that the entire NLP community has extensively studied many NLP DA approaches, recent research on the subject suggests that the relationship between the several DA techniques now in use is not entirely known in practice. Therefore, this study applies and extends the advances of TDA to encounter and cover varied tools on multiple settings or contexts. To carry out a thorough practical implementation of NLP DA approaches, comparing the way they perform and highlighting some of the significant similarities and differences in these various scenarios, this work depends on different tools of easy data augmentation and neural-based augmentation. This study suggests that some typical DA techniques might not be suitable in some circumstances or text environments. Specifically, according to the initial results, the context and word count of a text may have a significant impact on the quality of the synthetic data.

AUDITING WITH AI: A THEORETICAL FRAMEWORK FOR APPLYING MACHINE LEARNING ACROSS THE INTERNAL AUDIT LIFECYCLE

Artificial Intelligence (hereinafter AI), and specifically, Machine Learning (hereinafter, ML), has shown tremendous potential to revolutionize the internal audit (hereinafter, IA) profession, from enabling audit coverage of entire test populations, to introducing objectivity in the analysis of key areas. However, prior literature shows that the multiplicity of innovation options can be overwhelming. This paper aims to offer a theoretical framework that would enable audit practitioners, within both industry and professional services, to consider how ML capabilities can be harnessed to their fullest potential across the internal audit lifecycle, from audit planning to reporting. The paper discusses how DA and ML capabilities relate to the internal audit function’s (hereinafter, IAF) remit, drawing from extant literature. In doing so, the paper identifies the most specific options available to IAFs to drive innovation across each segment of the audit lifecycle, leveraging various DA and ML techniques, and supports the assertion that auditors require a continuous innovation mind-set to be effective change agents. The paper also draws out the requirement for effective guardrails, especially with emerging technology, such as Generative AI. Finally, the paper discusses how the value arising from these efforts can be measured by IAFs.

Using Data Assimilation to Improve Turbulence Modeling for Inclined Jets in Crossflow

Abstract Data assimilation (DA) integrating limited experimental data and computational fluid dynamics is applied to improve the prediction accuracy of flow and mixing behavior in inclined jet-in-crossflow (JICF). The ensemble Kalman filter (EnKF) approach is used as the DA technique, and the Reynolds-averaged Navier–Stokes (RANS) modeling serves as the prediction framework. The flow field and scalar mixing characteristics of a cylinder-inclined JICF and a sand dune (SD)-inspired inclined JICF are studied at various velocity ratios (VR = 0.4, 0.8, and 1.2). First, the Spalart–Allmaras (SA) model and the standard k-ɛ model are investigated based on the cylinder configuration at VR = 1.2. An optimized set of model constants are determined for each model using the EnKF-based data assimilation. The SA model shows remarkable improvement and better prediction in flow separation than the standard k-ɛ model after DA. Further exploration demonstrates that this set of the SA model constants can be extended to other VRs and even the SD-inspired configuration, mainly due to the correction of the predicted flow separation in inclined JICF. Finally, an investigation of the concentration field also shows satisfying improvement, resulting from a more appropriate turbulent Schmidt number. The optimized model constants, the revealed extensibility, and the uncovered mechanism of using the EnKF-based DA to improve the simulation of JICF could facilitate the design of related applications such as gas turbine film cooling.

Designing FIR Filter using Distributed Arithmetic

Abstract: Distributed arithmetic technique is employed in this paper to develop the FIR filter. To eliminate undesired signal and noise, FIR filters are frequently employed as digital filters in digital signal processing. This method also calls for shift registers and an accumulator and stores the FIR filter coefficients in a look-up table (LUT).By using this method, shift and accumulate can take the place of the multipliers in the FIR filter. The size of the LUT can be reduced for larger filter coefficients by dividing it into any number of LUTs, which are made up of taps that are FIR filter coefficients. FIR filter is created in MATLAB using the DA technique, and manual computation yields the same result

A comparative study of Fourier transform and CycleGAN as domain adaptation techniques for weed segmentation

Automatic weed identification is becoming increasingly important in the Precision Agriculture field as a fundamental capability for targeted spraying or mechanical weed destruction. Targeted weed elimination reduces herbicides' use and thus lowers the environmental impact of treatments. Convolutional Neural Networks are one of the most successful techniques to automatically detect weeds on RGB images. Such models require a high amount of labeled data to obtain satisfying detection performance. The agricultural context presents a high degree of variability, and it is thus unfeasible to expect a representative dataset for each specific condition that can appear in the fields. Domain Adaptation techniques are exploited to maintain high detection performance in different field conditions, lowering the need for labeled data. This study presents a comparison of the two main style transfer techniques for performing domain adaptation, that is, the Fourier Transform and the CycleGAN architecture. We used these techniques to reduce the domain gap in two use cases: one with images collected by different robots with different cameras and another with images collected by the same platform in different years. We show how, in the first case, the CycleGAN architecture attains satisfying performance and beats the simpler Fourier Transform. Instead, in the second case, all the tested DA techniques struggle to reach baseline performance. We also show how introducing a loss based on phase discrepancy in the CycleGAN architecture stabilizes the training and improves the performance. Moreover, we release a new dataset of labeled agricultural images and the code of our experiments for the reproducibility of the results and comparison with future works.

Diffeomorphic transforms for data augmentation of highly variable shape and texture objects

Background and objective: Training a deep convolutional neural network (CNN) for automatic image classification requires a large database with images of labeled samples. However, in some applications such as biology and medicine only a few experts can correctly categorize each sample. Experts are able to identify small changes in shape and texture which go unnoticed by untrained people, as well as distinguish between objects in the same class that present drastically different shapes and textures. This means that currently available databases are too small and not suitable to train deep learning models from scratch. To deal with this problem, data augmentation techniques are commonly used to increase the dataset size. However, typical data augmentation methods introduce artifacts or apply distortions to the original image, which instead of creating new realistic samples, obtain basic spatial variations of the original ones.Methods: We propose a novel data augmentation procedure which generates new realistic samples, by combining two samples that belong to the same class. Although the idea behind the method described in this paper is to mimic the variations that diatoms experience in different stages of their life cycle, it has also been demonstrated in glomeruli and pollen identification problems. This new data augmentation procedure is based on morphing and image registration methods that perform diffeomorphic transformations.Results: The proposed technique achieves an increase in accuracy over existing techniques of 0.47%, 1.47%, and 0.23% for diatom, glomeruli and pollen problems respectively.Conclusions: For the Diatom dataset, the method is able to simulate the shape changes in different diatom life cycle stages, and thus, images generated resemble newly acquired samples with intermediate shapes. In fact, the other methods compared obtained worse results than those which were not using data augmentation. For the Glomeruli dataset, the method is able to add new samples with different shapes and degrees of sclerosis (through different textures). This is the case where our proposed DA method is more beneficial, when objects highly differ in both shape and texture. Finally, for the Pollen dataset, since there are only small variations between samples in a few classes and this dataset has other features such as noise which are likely to benefit other existing DA techniques, the method still shows an improvement of the results.

Optimisation of user-selectable volumetric modulated arc therapy (VMAT) planning parameters: VMAT arcs for prostate and head-and-neck cancers.

To evaluate different VMAT planning techniques for a multi-leaf collimator (MLC)providing maximum aperture of 21 × 16 cm². A VMAT planning study of nine prostate and nine head-and-neck cancer patients was carried out. The patients were previously treated with Intensity Modulated Radiotherapy (IMRT) technique during 2014-15, at radiation oncology SanBortolo Hospital, Vicenza, Italy. Three VMAT plans for each of prostate and head-and-neck cancer patient were optimized using Pinnacle treatment planning system for 6MV photon beam from ElektaSynergyS®Linac system. Single arc, dual arc and combined two independent-single-arcs were optimized for collimator angle 45°. VMAT treatment planning comparison was made among single-arc dual-arc and combined two independent-single-arcs. The student's t-test (two-sided) with p ≤ 0.05 was applied for significant difference between dose volume indices of plans. All VMAT techniques produced clinically acceptable plans for prostate, while for head-and-neck, DA and ISAs significantly improved target coverage compared to SA. Single arc is sufficient for prostate, while inefficient in case of head-and-neck dose-planning. In spite of having different VMAT optimization approach dual arc and two combined independent-single-arcs provided very similar target coverage, while dual arc improved sparing of organs-at-risk. Similar treatment delivery times were observed for DA and ISAs VMAT techniques. Single arc is sufficient for prostate, while inefficient in case of head-and-neck dose-planning. Dual arc and two combined independent-single-arcs provided similar PTV coverage, while DA provided better sparing of organs at risk.

Development of Experimental Data Based Model Using DA Technique with ANN for I.C. Engine Operating Performance Analysis

Development of Experimental Data Based Model Using DA Technique with ANN for I.C. Engine Operating Performance Analysis

Satellite radiance data assimilation for binary tropical cyclone cases over the western North Pacific

AbstractA total of three binary tropical cyclone (TC) cases over the Western North Pacific are selected to investigate the effects of satellite radiance data assimilation on analyses and forecasts of binary TCs. Two parallel cycling experiments with a 6 h interval are performed for each binary TC case, and the difference between the two experiments is whether satellite radiance observations are assimilated. Satellite radiance observations are assimilated using the Weather Research and Forecasting Data Assimilation (WRFDA)'s three‐dimensional variational (3D‐Var) system, which includes the observation operator, quality control procedures, and bias correction algorithm for radiance observations. On average, radiance assimilation results in slight improvements of environmental fields and track forecasts of binary TC cases, but the detailed effects vary with the case. When there is no direct interaction between binary TCs, radiance assimilation leads to better depictions of environmental fields, and finally it results in improved track forecasts. However, positive effects of radiance assimilation on track forecasts can be reduced when there exists a direct interaction between binary TCs and intensities/structures of binary TCs are not represented well. An initialization method (e.g., dynamic initialization) combined with radiance assimilation and/or more advanced DA techniques (e.g., hybrid method) can be considered to overcome these limitations.

SU-E-I-53: Comparison of Kerma-Area-Product Between the Micro-Angiographic Fluoroscope (MAF) and a Flat Panel Detector (FPD) as Used in Neuro-Endovascular Procedures

Purpose: To determine the reduction of integral dose to the patient when using the micro-angiographic fluoroscope (MAF) compared to when using the standard flat-panel detector (FPD) for the techniques used during neurointerventional procedures. Methods: The MAF is a small field-of-view, high resolution x-ray detector which captures 1024 × 1024 pixels with an effective pixel size of 35μm and is capable of real-time imaging up to 30 frames per second. The MAF was used in neuro-interventions during those parts of the procedure when high resolution was needed and the FPD was used otherwise. The technique parameters were recorded when each detector was used and the kerma-area-product (KAP) per image frame was determined. KAP values were calculated for seven neuro interventions using premeasured calibration files of output as a function of kVp and beam filtration and included the attenuation of the patient table for the frontal projections to be more representative of integral patient dose. The air kerma at the patient entrance was multiplied by the beam area at that point to obtain the KAP values. The ranges of KAP values per frame were determined for the range of technique parameters used during the clinical procedures. To appreciate the benefit of the higher MAF resolution in the region of interventional activity, DA technique parameters were generally used with the MAF. Results: The lowest and highest values of KAP per frame for the MAF in DA mode were 4 and 50 times lower, respectively, compared to those of the FPD in pulsed fluoroscopy mode. Conclusion: The MAF was used in those parts of the clinical procedures when high resolution and image quality was essential. The integral patient dose as represented by the KAP value was substantially lower when using the MAF than when using the FPD due to the much smaller volume of tissue irradiated. This research was supported in part by Toshiba Medical Systems Corporation and NIH Grant R01EB002873.

Distributed arithmetic based non recursive filter for high throughput and low power applications

The proposed work demonstrates a design of adaptive nonrecursive filter design based on distributed arithmetic technique. First, the efficient pipelined distributed arithmetic technique has been proposed here, it achieves low power and reduced switching activity than the conventional one. Second, to attain high throughput the parallel processing based distributed technique is designed. Analysis has been made between the first and second model experimental result shows pipelined DA technique increasing the speed with reduced switching activity and to achieve better area, parallel processing technique will be the efficient one. The proposed design is simulated in modelsim 6.5b by using VHDL, synthesized in Xilinx ISE 14.2 and implemented in FPGA Spartan-3E.

Investigating VMAT planning technique to reduce rectal and bladder dose in prostate cancer treatment plans

Background: RapidArc is a volumetric modulated arc therapy (VMAT) technique that can deliver conformal dose distribution to the target while minimizing dose to critical structures. The main purpose of this study was to compare dosimetric quality of full double arc (full DA), full single arc (full SA), and partial double arc (partial DA) techniques in RapidArc planning of prostate cancer. Materials and Methods: Twelve cases of prostate cancer involving seminal vesicles were selected for this retrospective study. For each case, RapidArc plans were created using full DA (two full arcs), full SA (one full arc), and partial DA (two partial arcs with anterior and posterior avoidance sectors) techniques. For planning target volume (PTV), the maximum and mean doses, conformity, and inhomogeneity indices were evaluated. For bladder and rectum, volumes that received 70, 50, 40, and 20 Gy (V 70Gy , V 50Gy , V 40Gy and V 20Gy , respectively), and mean dose were compared. For femoral heads, V 40Gy , V 20Gy , and mean dose were evaluated. Additionally, an integral dose and monitor units (MUs) were compared for each treatment plan. Results: In comparison to full DA and full SA techniques, the partial DA technique was better in sparing of rectum and bladder but delivered higher femoral head dose, which was nonetheless within the planning criteria. No clear dosimetric differences were found between full DA and partial DA plans for dose conformity and target homogeneity. The number of MUs and integral dose were largest with the partial DA technique and lowest with the full SA technique. Conclusion: The partial DA technique provides an alternative RapidArc planning approach for low risk prostate cancer.

Pain related to maggot debridement therapy

To summarise our experience with maggot debridement therapy (MDT) in relation to pain observed in patients treated in 16 departments and units of the Hadassah Hospital in Jerusalem, Israel. A secondary analysis of data from 435 patients. Maggots were either placed directly on the wound using a cage-like dressing (DA), or they were applied to the wound encased in a previously prepared hermetic tea-bag like pouch (TBA). During each treatment, patients were asked whether they felt any difference in the level of pain before and during MDT treatment. Overall, 165 patients (38%) reported increased pain during MDT. Seventeen patients (41%), who were treated with the TBA technique, and 148 patients (38%), who were treated with the DA technique, reported increased pain and were treated with analgesics before or during MDT. In five patients, the treatment had to be discontinued due to uncontrolled pain during MDT. Pain control measures were undertaken in patients who were already in intense pain prior to the initiation of MDT and in those patients who reported intense pain during DA treatment. The measures included shorter periods of application of maggots, use of the TBA rather than DA technique, and applying relatively small maggots as well as a smaller number of maggots during an MDT session. In patients who continued reporting intense pain, systemic analgesic medications and peripheral nerve blocks were used. Since a full debridement requires an average of 2-3 maggot cycles, which last 3-5 days, and since a large percentage of patients treated with MDT complain of pain that may last throughout the therapy period, it is deemed worthwhile and even essential to titrate analgesics as needed and be prepared to treat patients even with potent analgesics, such as opioids. In cases when pain is intense and uncontrolled with systemic medications, peripheral nerve blocks should be considered. There were no external sources of funding for this study. The authors have no conflicts of interest to declare.

Reversing the Perspective on the Applications of Decision Analysis

This paper looks into the strengths, weaknesses, opportunities, and threats (SWOT) of the field of the applications based on DA techniques; focusing especially on MAVT. The need for best practices and bias-resistant analysis procedures is pointed out. In most application papers there are no reports on the verification or testing of the procedures used. The Internet provides great opportunities in the delivery of DA software and e-learning material. Public sites such as Decisionarium ( www.decisionarium.hut.fi ) provide encouragement and a low entry level to try value tree analysis in new applications. There are also great opportunities in reaching out the other MCDA communities such as AHP practitioners. Today the need and demand for decision support is perhaps growing most rapidly in environmental problems. A clear trend is also the integration of DA into GIS and other models. Many of the applications are now published in non-OR/MS subject area journals.

Reversing the Perspective on the Applications of Decision Analysis (Comment on Keefer et al. 2004)

This paper looks into the strengths, weaknesses, opportunities, and threats (SWOT) of the field of the applications based on DA techniques; focusing especially on MAVT. The need for best practices and bias-resistant analysis procedures is pointed out. In most application papers there are no reports on the verification or testing of the procedures used. The Internet provides great opportunities in the delivery of DA software and e-learning material. Public sites such as Decisionarium (www.decisionarium.hut.fi) provide encouragement and a low entry level to try value tree analysis in new applications. There are also great opportunities in reaching out the other MCDA communities such as AHP practitioners. Today the need and demand for decision support is perhaps growing most rapidly in environmental problems. A clear trend is also the integration of DA into GIS and other models. Many of the applications are now published in non-OR/MS subject area journals.

The American Association for Vascular Surgery: Advocate for independence

The American Association for Vascular Surgery: Advocate for independence

The American Association for Vascular Surgery: Advocate for independence

The American Association for Vascular Surgery: Advocate for independence

A survey of DA techniques for PLD and FPGA based systems

Programmable logic devices (PLDs) are gaining in acceptance, of late, for designing systems of all complexities ranging from glue logic to special purpose parallel machines. Higher densities and integration levels are made possible by the new breed of complex PLDs and FPGAs. The added complexities of these devices make automatic computer aided tools indispensable for achieving good performance and a high usable gate-count. In this article, we attempt to present in an unified manner, the different tools and their underlying algorithms using an example of a vending machine controller as an illustrative example. Topics covered include logic synthesis for PLDs and FPGAs along with an in-depth survey of important technology mapping, partitioning and place and route algorithms for different FPGA architectures.