Output Encoding Research Articles

Dq-YOLOF: An Effective Improvement with Deformable Convolution and Sample Quality Optimization Based on the YOLOF Detector

Single-stage detectors have drawbacks of insufficient accuracy and poor coverage capability. YOLOF (You Only Look One-level Feature) has achieved better performance in this regard, but there is still room for improvement. To enhance the coverage capability for objects of different scales, we propose an improved single-stage object detector: Dq-YOLOF. We have designed an output encoder that employs a series of modules utilizing deformable convolution and SimAM (Simple Attention Module). This module replaces the dilated convolution in YOLOF. This design significantly improves the ability to express details. Simultaneously, we have redefined the sample selection strategy, which optimizes the quality of positive samples based on SimOTA. It can dynamically allocate positive samples according to their quality, reducing computational load and making it more suitable for small objects. Experiments conducted on the COCO 2017 dataset also verify the effectiveness of our method. Dq-YOLOF achieved 38.7 AP, 1.5 AP higher than YOLOF. To confirm performance improvements on small objects, our method was tested on urinary sediment and aerial drone datasets for generalization. Notably, it enhances performance while also lowering computational costs.

Universal representation learning for multivariate time series using the instance-level and cluster-level supervised contrastive learning

AbstractThe multivariate time series classification (MTSC) task aims to predict a class label for a given time series. Recently, modern deep learning-based approaches have achieved promising performance over traditional methods for MTSC tasks. The success of these approaches relies on access to the massive amount of labeled data (i.e., annotating or assigning tags to each sample that shows its corresponding category). However, obtaining a massive amount of labeled data is usually very time-consuming and expensive in many real-world applications such as medicine, because it requires domain experts’ knowledge to annotate data. Insufficient labeled data prevents these models from learning discriminative features, resulting in poor margins that reduce generalization performance. To address this challenge, we propose a novel approach: supervised contrastive learning for time series classification (SupCon-TSC). This approach improves the classification performance by learning the discriminative low-dimensional representations of multivariate time series, and its end-to-end structure allows for interpretable outcomes. It is based on supervised contrastive (SupCon) loss to learn the inherent structure of multivariate time series. First, two separate augmentation families, including strong and weak augmentation methods, are utilized to generate augmented data for the source and target networks, respectively. Second, we propose the instance-level, and cluster-level SupCon learning approaches to capture contextual information to learn the discriminative and universal representation for multivariate time series datasets. In the instance-level SupCon learning approach, for each given anchor instance that comes from the source network, the low-variance output encodings from the target network are sampled as positive and negative instances based on their labels. However, the cluster-level approach is performed between each instance and cluster centers among batches, as opposed to the instance-level approach. The cluster-level SupCon loss attempts to maximize the similarities between each instance and cluster centers among batches. We tested this novel approach on two small cardiopulmonary exercise testing (CPET) datasets and the real-world UEA Multivariate time series archive. The results of the SupCon-TSC model on CPET datasets indicate its capability to learn more discriminative features than existing approaches in situations where the size of the dataset is small. Moreover, the results on the UEA archive show that training a classifier on top of the universal representation features learned by our proposed method outperforms the state-of-the-art approaches.

Theory and practice in secure software development lifecycle: A comprehensive survey

Software development security refers to the practice of integrating security measures and considerations throughout the software development lifecycle to ensure the confidentiality, integrity, and availability of software systems. It involves identifying, mitigating, and eliminating security vulnerabilities and threats that could be exploited by attackers. The goal of this paper is to survey the various concepts and methodologies directed towards software security, and the identification of any missing gaps. Based on the findings, it is noted that the development of secure software requires a proactive and comprehensive approach. It begins with establishing secure design principles and incorporating security requirements from the initial stages of development. Here, secure coding practices, such as input validation, output encoding, and secure authentication and authorization mechanisms, are employed to prevent common security vulnerabilities. In addition, regular security testing, including penetration testing and vulnerability scanning, helps identify and address potential weaknesses in the software. Normally, code reviews and security audits are conducted to ensure adherence to secure coding practices and identify any security flaws. It is important that security training and awareness programs be provided to developers and other stakeholders to foster a security-conscious culture. To minimize potential vulnerabilities, secure configuration management, which involves properly configuring servers, networks, and dependencies may be utilized. On the other hand, regular updates and patching are essential to address known security vulnerabilities in software components. To guide their software development security practices, organizations may follow established security standards and frameworks such as ISO 27001 or NIST Cybersecurity Framework. By prioritizing software development security, organizations can protect sensitive data, prevent unauthorized access, and mitigate the risk of security breaches and incidents. In the long run, this helps build trust with users and stakeholders, enhances the reputation of the software, and reduces the potential impact of security incidents on the organization.

Self-triggered resilient stabilization of linear systems with quantized outputs

This paper studies the problem of stabilizing a self-triggered control system with quantized output. Employing a standard observer-based state feedback control law, a self-triggering mechanism that dictates the next sampling time based on quantized output is co-developed with an output encoding scheme. If, in addition, the transmission protocols at the controller-to-actuator (C–A) and sensor-to-controller (S–C) channels can be adapted, the self-triggered control architecture can be considerably simplified, leveraging a delicate observer-based deadbeat controller to eliminate the need for running the controller in parallel at the encoder side. To account for denial-of-service (DoS) in the S–C channel, the proposed output encoding and self-triggered control schemes are further made resilient. It is shown that a linear time-invariant system can be exponentially stabilized if some conditions on the average DoS duration time are met. There is a trade-off between the maximum inter-sampling time and the resilience against DoS attacks. Finally, a numerical example is presented to demonstrate the practical merits of the proposed self-triggered control schemes and associated theory.

MRI-based brain tumor detection using convolutional deep learning methods and chosen machine learning techniques

BackgroundDetecting brain tumors in their early stages is crucial. Brain tumors are classified by biopsy, which can only be performed through definitive brain surgery. Computational intelligence-oriented techniques can help physicians identify and classify brain tumors. Herein, we proposed two deep learning methods and several machine learning approaches for diagnosing three types of tumor, i.e., glioma, meningioma, and pituitary gland tumors, as well as healthy brains without tumors, using magnetic resonance brain images to enable physicians to detect with high accuracy tumors in early stages.Materials and MethodsA dataset containing 3264 Magnetic Resonance Imaging (MRI) brain images comprising images of glioma, meningioma, pituitary gland tumors, and healthy brains were used in this study. First, preprocessing and augmentation algorithms were applied to MRI brain images. Next, we developed a new 2D Convolutional Neural Network (CNN) and a convolutional auto-encoder network, both of which were already trained by our assigned hyperparameters. Then 2D CNN includes several convolution layers; all layers in this hierarchical network have a 2*2 kernel function. This network consists of eight convolutional and four pooling layers, and after all convolution layers, batch-normalization layers were applied. The modified auto-encoder network includes a convolutional auto-encoder network and a convolutional network for classification that uses the last output encoder layer of the first part. Furthermore, six machine-learning techniques that were applied to classify brain tumors were also compared in this study.ResultsThe training accuracy of the proposed 2D CNN and that of the proposed auto-encoder network were found to be 96.47% and 95.63%, respectively. The average recall values for the 2D CNN and auto-encoder networks were 95% and 94%, respectively. The areas under the ROC curve for both networks were 0.99 or 1. Among applied machine learning methods, Multilayer Perceptron (MLP) (28%) and K-Nearest Neighbors (KNN) (86%) achieved the lowest and highest accuracy rates, respectively. Statistical tests showed a significant difference between the means of the two methods developed in this study and several machine learning methods (p-value < 0.05).ConclusionThe present study shows that the proposed 2D CNN has optimal accuracy in classifying brain tumors. Comparing the performance of various CNNs and machine learning methods in diagnosing three types of brain tumors revealed that the 2D CNN achieved exemplary performance and optimal execution time without latency. This proposed network is less complex than the auto-encoder network and can be employed by radiologists and physicians in clinical systems for brain tumor detection.

Highdicom: a Python Library for Standardized Encoding of Image Annotations and Machine Learning Model Outputs in Pathology and Radiology

Machine learning (ML) is revolutionizing image-based diagnostics in pathology and radiology. ML models have shown promising results in research settings, but the lack of interoperability between ML systems and enterprise medical imaging systems has been a major barrier for clinical integration and evaluation. The DICOM® standard specifies information object definitions (IODs) and services for the representation and communication of digital images and related information, including image-derived annotations and analysis results. However, the complexity of the standard represents an obstacle for its adoption in the ML community and creates a need for software libraries and tools that simplify working with datasets in DICOM format. Here we present the highdicom library, which provides a high-level application programming interface (API) for the Python programming language that abstracts low-level details of the standard and enables encoding and decoding of image-derived information in DICOM format in a few lines of Python code. The highdicom library leverages NumPy arrays for efficient data representation and ties into the extensive Python ecosystem for image processing and machine learning. Simultaneously, by simplifying creation and parsing of DICOM-compliant files, highdicom achieves interoperability with the medical imaging systems that hold the data used to train and run ML models, and ultimately communicate and store model outputs for clinical use. We demonstrate through experiments with slide microscopy and computed tomography imaging, that, by bridging these two ecosystems, highdicom enables developers and researchers to train and evaluate state-of-the-art ML models in pathology and radiology while remaining compliant with the DICOM standard and interoperable with clinical systems at all stages. To promote standardization of ML research and streamline the ML model development and deployment process, we made the library available free and open-source at https://github.com/herrmannlab/highdicom.

Structural Decomposition in FSM Design: Roots, Evolution, Current State—A Review

The review is devoted to methods of structural decomposition that are used for optimizing characteristics of circuits of finite state machines (FSMs). These methods are connected with the increasing the number of logic levels in resulting FSM circuits. They can be viewed as an alternative to methods of functional decompositions. The roots of these methods are analysed. It is shown that the first methods of structural decomposition have appeared in 1950s together with microprogram control units. The basic methods of structural decomposition are analysed. They are such methods as the replacement of FSM inputs, encoding collections of FSM outputs, and encoding of terms. It is shown that these methods can be used for any element basis. Additionally, the joint application of different methods is shown. The analysis of change in these methods related to the evolution of the logic elements is performed. The application of these methods for optimizing FPGA- based FSMs is shown. Such new methods as twofold state assignment and mixed encoding of outputs are analysed. Some methods are illustrated with examples of FSM synthesis. Additionally, some experimental results are represented. These results prove that the methods of structural decomposition really improve the characteristics of FSM circuits.

Card-Based Cryptography Meets Formal Verification

Card-based cryptography provides simple and practicable protocols for performing secure multi-party computation with just a deck of cards. For the sake of simplicity, this is often done using cards with only two symbols, e.g., clubsuit and heartsuit . Within this paper, we also target the setting where all cards carry distinct symbols, catering for use-cases with commonly available standard decks and a weaker indistinguishability assumption. As of yet, the literature provides for only three protocols and no proofs for non-trivial lower bounds on the number of cards. As such complex proofs (handling very large combinatorial state spaces) tend to be involved and error-prone, we propose using formal verification for finding protocols and proving lower bounds. In this paper, we employ the technique of software bounded model checking (SBMC), which reduces the problem to a bounded state space, which is automatically searched exhaustively using a SAT solver as a backend. Our contribution is threefold: (a) we identify two protocols for converting between different bit encodings with overlapping bases, and then show them to be card-minimal. This completes the picture of tight lower bounds on the number of cards with respect to runtime behavior and shuffle properties of conversion protocols. For computing AND, we show that there is no protocol with finite runtime using four cards with distinguishable symbols and fixed output encoding, and give a four-card protocol with an expected finite runtime using only random cuts. (b) We provide a general translation of proofs for lower bounds to a bounded model checking framework for automatically finding card- and run-minimal (i.e., the protocol has a run of minimal length) protocols and to give additional confidence in lower bounds. We apply this to validate our method and, as an example, confirm our new AND protocol to have its shortest run for protocols using this number of cards. (c) We extend our method to also handle the case of decks on symbols clubsuit and heartsuit , where we show run-minimality for two AND protocols from the literature.

Output Encoding: Sanitizing and Encoding Outputs to Prevent XSS and Other Injection Attacks

Output Encoding: Sanitizing and Encoding Outputs to Prevent XSS and Other Injection Attacks

Crowd Scene Analysis Encounters High Density and Scale Variation.

Crowd scene analysis receives growing attention due to its wide applications. Grasping the accurate crowd location is important for identifying high-risk regions. In this article, we propose a Compressed Sensing based Output Encoding (CSOE) scheme, which casts detecting pixel coordinates of small objects into a task of signal regression in encoding signal space. To prevent gradient vanishing, we derive our own sparse reconstruction backpropagation rule that is adaptive to distinct implementations of sparse reconstruction and makes the whole model end-to-end trainable. With the support of CSOE and the backpropagation rule, the proposed method shows more robustness to deep model training error, which is especially harmful to crowd counting and localization. The proposed method achieves state-of-the-art performance across four mainstream datasets, especially achieves excellent results in highly crowded scenes. A series of analysis and experiments support our claim that regression in CSOE space is better than traditionally detecting coordinates of small objects in pixel space for highly crowded scenes.

Optimization of Output Spike Train Encoding for a Spiking Neuron Based on its Spatio–Temporal Input Pattern

A common learning task for a spiking neuron is to map a spatio–temporal input pattern to a target output spike train. There is no prescribed method for selection of the target output spike train. However, the precise spiking pattern of the target output spike train (output encoding) can affect the learning performance of the spiking neuron. Therefore, systematic methods of finding the optimum spiking pattern for a target output spike train that can be learned by spiking neurons are needed. Here, a method is proposed to adaptively adjust an initial suboptimal output encoding during different learning epochs to find the optimal output encoding. A time varying value of a local event called a spike trace is used to calculate the amount of a required adjustment. The remote supervised method (ReSuMe) learning algorithm is used to train the weights, and the proposed method is used for finding optimized output encoding (optimized desired spikes). Experimental results show that optimizing the output encoding during the learning phase increases the accuracy. The proposed method was applied to find optimized output encoding in classification tasks and the results revealed improvements up to 16.5% in accuracy compared to when using the non-adapted method. It also increases the accuracy in a classification task from 90% to 100%.

Ensemble of Random Binary Output Encoding for Adversarial Robustness

Despite the excellent classification performance, recent research has revealed that the Convolutional Neural Network (CNN) could be readily deceived by only the small adversarial perturbation. Its imperceptible to human eyes and transferability from one model to another actually threaten the security of a CNN-based system. In this paper, we propose to create multiple and independent random binary codes per input class and train ensemble of homogeneous CNN classifiers with these codes to improve the adversarial robustness of the networks. The proposed ensemble structure consists of replicas of the same learning architecture, but each network is trained with different random target outputs. The network model is simultaneously trained with their own unique binary codes, and optimized through a single and common objective function in an end-to-end manner. It is demonstrated with experimental results that assigning different encoded labels for each classifier in ensemble leverages the diversity and eventually improves the classification performance on adversarial attacks. We also conduct several performance analysis to understand how the different aspects can contribute to the robustness of the proposed algorithm. The proposed algorithm provides significantly improved classification accuracies as compared to the recent relevant studies, verified with various network architectures, datasets, and adversarial attacks.

Secure Programming with SAS: Mitigating Risks and Protecting Data Integrity

This article examines the significance of safe programming with SAS (Statistical Analysis System) for risk mitigation and data integrity protection in data-driven environments. The study intends to investigate data protection strategies to improve application security, look at best practices for secure SAS coding, and identify prevalent security threats related to SAS applications. The review process is secondary data, with insights gleaned from online resources, industry reports, conference papers, and academic journals. Important discoveries show that SAS programs frequently have vulnerabilities, including SQL injection, cross-site scripting (XSS), and unsafe data handling. The best practices that have been established encompass data encryption, secure access controls, output encoding, parameterized queries, and input validation. The policy implications emphasize the significance of legislative frameworks for data protection and encouraging instruction in secure programming practices. This report emphasizes how important it is for businesses to use SAS for secure programming to protect sensitive data, abide by data protection laws, and defend against cyberattacks.

A wearable mobility device for the blind using retina-inspired dynamic vision sensors.

Proposed is a prototype of a wearable mobility device which aims to assist the blind with navigation and object avoidance via auditory-vision-substitution. The described system uses two dynamic vision sensors and event-based information processing techniques to extract depth information. The 3D visual input is then processed using three different strategies, and converted to a 3D output sound using an individualized head-related transfer function. The performance of the device with different processing strategies is evaluated via initial tests with ten subjects. The outcome of these tests demonstrate promising performance of the system after only very short training times of a few minutes due to the minimal encoding of outputs from the vision sensors which are translated into simple sound patterns easily interpretable for the user. The envisioned system will allow for efficient real-time algorithms on a hands-free and lightweight device with exceptional battery life-time.

Characterization, modeling and vibration control of a flexible joint for a robotic system

This paper presents the experimental characterization and vibration control of a flexible robotic system. For this work, a test bench was built to characterize the harmonic drive (HD) and joint components, while control algorithms were designed and compared to minimize vibration. Encoder accuracy was critical since the difference in the measurements between two encoders was used to evaluate the vibrational behavior of the test set-up. Therefore, a laser tracker was used to characterize the error of the output encoder. Real-time compensation using this technique achieved an angular position accuracy of 50 µrad. Four rosette strain gauges were fixed to the HD’s flexible spline to determine its torsion. To reduce torque ripple, a real-time correcting function was applied. It was thus possible to reduce the error to 0.3% of the full-scale error. Two vibration control strategies were developed, namely, singular perturbation and feed-forward control. Simulation results showed that both control strategies greatly reduced vibration response compared to a common rigid control. However, test results showed that good vibration control could only be achieved with the feed-forward approach: the singular perturbation technique generated too much torque ripple to the motor. A feed-forward controller can quickly stabilize the link, achieving the same settling time as with the rigid control algorithm.

Theoretical Correlations and Measured Correlations: Relating Recognition and Recall in Four Distributed Memory Models.

This article addresses the relation between item recognition and associative (cued) recall. Going beyond measures of performance on each task, the analysis focuses on the degree to which the contingency between successful recognition and successful recall of a studied item reflects the commonality of memory processes underlying the recognition and recall tasks. Specifically, 4 classes of distributed memory models are assessed for their ability to account for the relatively invariant correlation (approximately .5) between successive recognition and recall. Basic versions of each model either under- or overpredict the intertask correlation. Introducing variability in goodness-of-encoding and response criteria, as well as output encoding, enabled all 4 models to reproduce the moderate intertask correlation and the increase in correlation observed in 2 mixed-list experiments. This model-based analysis provides a general theoretical framework for interpreting contingencies between successive memory tests.

Weak repetitions in Sturmian strings

In this paper, we analyze the weak repetitions in Sturmian strings and show that an optimally efficient algorithm to compute the weak repetitions in Sturmian strings according to the output encodings defined in the literature is quadratic in the string length. Finally, we present an encoding that leads to a linear-time algorithm.

Low-area edge sampler using the Chinese remainder theorem

We present a method of reducing the area of edge sampling circuits that employ the matched delay technique. The method uses the Chinese remainder theorem to decompose a long delay line of length L into n small and separate lines operating in parallel and having a total length proportional to n /sup n//spl radic/L. The area reduction is important because it allows lower power dissipation. It also yields higher measurement resolution because delay element variation along the line, which is a function of chip area, is reduced. We also present the design of an ASIC, called the CRT Sampler, using a 2 /spl mu/m, double metal, single poly CMOS process. We show its layout and give its performance results. We discuss several important design issues, including delay element implementation, maximum measurable delay, output encoding and turn-off delay effects.

A hyper optimal encoding scheme for self-checking circuits

A typical self-checking circuit has an unordered code encoded output. The optimal scheme needs [log(r+1)] check bits, where r is the number of unique weights in all output patterns. A hyper optimal scheme for self-checking output encoding is proposed in this paper where the number of check bits will be further reduced in some cases. Two algorithms are presented to search for the hidden m-out-of-n code words. The hidden m-out-of-n code words are found when all unique output patterns, specified by the circuit specification, in the n selected output bits have exactly m 1s. The output bits that belong to the hidden m-out-of-n code words are then excluded from further encoding. Typically, the number of added check bits of the proposed technique ranges from 0 to [log(p+1)], where p/spl les/r. When hidden m-out-of-n code words exist, applying the proposed scheme usually results in significant hardware cost and delay time reduction. In the five MCNC FSM benchmark circuits that have been identified with hidden m-out-of-n code words, 10% to 41% hardware reductions are exhibited compared to the theoretically optimal separable code encoding scheme. In addition, 7% to 45% reductions in checking delays are demonstrated for the same circuits compared to the separable code encoding scheme.

A review and analysis of backpropagation neural networks for classification of remotely-sensed multi-spectral imagery

Abstract A literature survey and analysis of the use of neural networks for the classification of remotely-sensed multi-spectral imagery is presented. As part of a brief mathematical review, the backpropagation algorithm, which is the most common method of training multi-layer networks, is discussed with an emphasis on its application to pattern recognition. The analysis is divided into five aspects of neural network classification: (1) input data preprocessing, structure, and encoding, (2) output encoding and extraction of classes, (3) network architecture, (4) training algorithms, and (5) comparisons to conventional classifiers. The advantages of the neural network method over traditional classifiers are its nonparametric nature, arbitrary decision boundary capabilities, easy adaptation to different types of data and input structures, fuzzy output values that can enhance classification, and good generalization for use with multiple images. The disadvantages of the method are slow training time, inconsis...