Abstract Medical image analysis and disease diagnosis have significantly improved with the use of AI and Machine Learning algorithms. Automated systems for medical image analysis will help the doctors and radiologists understand the anomaly in a short span of time and with better visualization. Such automated systems will help to reduce the time taken for diagnosis by experts. Recently, Computer Vision is industrialized with the advancements in algorithms and hardware. The proposed study aims to develop a computer vision solution for automatic segmentation and classification of ovarian tumours in discriminating between benign and malignant tumours by image‐to‐image translation approach using Conditional Generative Adversarial Network (cGAN). Our method uses a novel algorithm which segments and classifies the images in a single pipeline which makes the algorithm unique and useful. This research also aims to compare its diagnostic accuracy with that of an expert radiologist. The dataset used by in the present study is formulated with images obtained from a hospital and annotated by doctors from the hospital. The obtained results show the proposed study is promising for ovarian cancer segmentation and classification with an average segmentation score of 0.825 for benign and 0.765 for malignant and classification accuracy of 83% for benign and 79% for malignant, precision score of 85% for benign and 80% for malignant and F1 score of 81% for benign and 80.1% for malignant images. The proposed methodology is evaluated on the existing MRI images to perform segmentation and classification. The results obtained shows that the proposed methodology can perform well on other MRI images. In this study, proposed methodology is convenient as separate segmentation need not be done and is giving good result. The same MRI images are segmented using UNet and classified using RESNET 101 and results are compared with the proposed methodology.

The$(1+1)$quantum evolutionary algorithm [$(1+1)$QEA] uses quantum probability amplification to accelerate the classical$(1+1)$evolutionary algorithm [$(1+1)$EA] for the optimization of pseudo-Boolean functions, which assign real-valued fitness to binary strings. However, determining the optimal mutation rate remains crucial to reducing the required number of function evaluations. To address this, we introduce a dynamic mutation rate strategy for the$(1+1)$QEA, leveraging the guarantees the fixed-point amplitude amplification algorithm gives to adjust the mutation rate dynamically [$(1+1)$QEAdyn]. We derive upper bounds on the number of fitness evaluations the strategy requires to solve well-known benchmark functions. Notably, we close the performance gap of the$(1+1)$QEA on the NEEDLE problem, matching the optimal performance possible in the quantum setting. Nevertheless, our results on ONEMAX and LEADINGONES lag behind known classical results. So, we delve into quantum black-box complexity, exploring the difficulty of problem-solving on a quantum computer without explicit problem knowledge. We introduce a quantum algorithm that solves the$n$-dimensional LEADINGONES problem with at most$n$evaluations. Additionally, we develop a quantum algorithm for the ONEMAX problem, requiring only one function evaluation. Both of these quantum algorithms surpass the capabilities of classical algorithms. Finally, we devise a quantum algorithm for solving the ONEMAX problem using only fitness comparisons and provide computational evidence that it still does so faster than any classical algorithm.

Abstract Ovarian abnormality like polycystic ovarian disease (PCOD) is one of the most common diseases among women worldwide. PCOD not only has an impact on infertility but also hurts the psychological well‐being of women affecting their quality of life. In this study, a two‐class pattern learning problem is designed for the classification of PCOD. In total, 37 clinical parameters and abdominal ultrasound images of women are collected under the proper ethical protocol. Using only clinical data, an accuracy of 93.7% is obtained using Random Forest as the classifier which is further improved to 95.54% by using a Randomized Search CV during Random Forest classification. The ultrasound images are classified using the proposed Attention‐UNet architecture and a mean Dice score of 0.945 is obtained indicating more accurate segmentation. The segmented images are passed through the state‐of‐the‐art EfficientNet B6 for the classification of PCOS and non‐PCOS and recorded an accuracy of 95.47%. Using big data architecture Pyspark, the performance is further enhanced to 96.8% and 96.3% for clinical and ultrasound images respectively along with the reduced computational speed. The results of these classifiers are then used to create metadata and a customized Artificial Neural Network is applied for the final prediction of PCOD and non‐PCOD. From the results, it can be seen that the stacking model outperformed with an accuracy of 98.12% when compared to the single classifier. Our proposed method has very good performance with less computation, contributing a new architecture to evaluate PCOD and hence helping to improve the wellness of women.

Few shot learning algorithms are designed to perform significantly well when the amount of annotated data is scarce. However, recent research shows that these algorithms are highly vulnerable to adversarial examples. Recent works have demonstrated the robustness of few-shot image classifiers. However, there is very little or no attention paid to the robustness of few-shot text classifiers. In this work, we highlight the vulnerability of few-shot text classifiers. We also provide an algorithm for adversarial training of few-shot classifiers that perform well in the presence of adversarial examples. We implemented this algorithm on existing state-of-the-art few-shot classifiers. Experimental results demonstrate that our algorithm resists adversarial attacks and performs better in the presence of adversarial examples.

An estimation-of-distribution algorithm is an evolutionary algorithm that uses a probabilistic model to represent its population. It iteratively draws solutions from its model and, considering their fitnesses, uses these solutions to update the model’s parameters. However, there is a risk of reinforcing random noise from sampling back into the model—a problem termed genetic drift. We propose the competing genes evolutionary algorithm that optimizes a fitness function over a binary search space and avoids this problem by updating only one model parameter at a time. The algorithm uses the model not only to sample solutions but also to select a parameter in each iteration that it pins to a value for the rest of the search process. We obtain upper bounds on the number of fitness evaluations the algorithm needs to solve well-known benchmark functions as a function of its population size and observe better or comparable results to other evolutionary algorithms. We attribute these favorable results to the algorithm’s efficient use of its samples to explore its dwindling search space. We also introduce two variants of the algorithm. The first version eliminates the need to preset the number of solutions to sample per iteration, and the second seeks to escape local optima. We provide evidence that these variants achieve their goals.

Time series classification is an important model in data mining. It involves assigning a class label to a test instance based on the training data with known class labels. Most previous studies developed time series classifiers by disregarding the fuzzy nature of events (i.e., events with similar values may belong to different classes) within the data. Consequently, these studies suffered from performance issues, including decreased accuracy and increased memory, runtime, and energy requirements. With this motivation, this paper proposes a novel fuzzy nearest neighbor classifier for time series data. The basic idea of our classifier is to transform the very large training data into a relatively small representative training data and use it to label a test instance by employing a new fuzzy distance measure known as Ravi. Experimental results on real world benchmark datasets demonstrate that the proposed classifier outperforms the current parameter-free time series classifiers and also the popular deep learning techniques.

In recent times malware attacks on government and private organizations are rising. These attacks are carried out to steal confidential information which leads to loss of privacy, intellectual property issues and loss of revenue. These attacks are sophisticated and described as Advanced Persistent Threats(APT). The payloads used in this type of attacks are polymorphic and metamorphic in nature and contains stealth and root-kit components. As a result the conventional defence mechanisms like rule-based and signature-based methods fail to detect these malware. So modern approaches rely on static and dynamic analysis to detect sophisticated malware. However this process generates huge log files. The domain expert needs to review these logs to classify whether the binary is malicious or benign which is tedious, time consuming and expensive. Our work uses machine learning models trained on the datasets, created using the analysis logs, to overcome these problems. In this paper a number of supervised machine learning models are presented to classify the binary as malicious or benign. In this work we have used automated malware analysis framework to collect run time behavioural artefacts. Static analysis mainly focuses on collecting binary meta information, import functions and opcode sequences. The dataset is created by collecting malware from online sources and benign files from windows operating system and third party software. © 2021 by SCITEPRESS - Science and Technology Publications, Lda.

Modern day cyberattacks are complex in nature. These attacks have adverse effects like loss of privacy, intellectual property and revenue on the victim institutions. These attacks have sophisticated payloads like ransom-ware for money extortion, distributed denial of service(DDOS) malware for service disruptions and advanced persistent threat(APT) malware to posses complete control over the victims computing resources. These malware are metamorphic and polymorphic in nature and contains root-kit components to maintain stealth and hide their malicious activity. So conventional defence mechanisms like rule-based and signature based mechanisms fail to detect these malware. Modern approaches use behavioural analysis(static analysis, dynamic analysis) to identity this kind of malware. However behavioural analysis process is hindered by factors like execution environment detection, code obfuscation, anti virtualization, anti-debugging, analysis environment detection etc. Behavioural analysis also requires domain expert to review the large amount of logs produced by it to decide on the nature of the binary which is complex, time consuming and expensive. To deal with these problems we proposed deep learning methods, where convolutional neural network model is trained on the image representation of the binary to decide the binary nature as malicious or benign. In this work we have encoded the binaries into images in a unique way. Deep convolution neural network is trained on these images to learn the features to identify the binary as malicious or normal. The malware and benign samples for the dataset creation are collected from online sources and windows operating system along with compatible third party application software respectively.

The task of language independent acoustic unit modeling in unlabeled raw speech (zero-resource setting) has gained significant interest over the recent years. The main challenge here is the extraction of acoustic representations that elicit good similarity between the same words or linguistic tokens spoken by different speakers and to derive these representations in a language independent manner. In this paper, we explore the use of Hidden Markov Model (HMM) based posteriograms for unsupervised acoustic unit modeling. The states of the HMM (which represent the language independent acoustic units) are initialized using a Gaussian mixture model (GMM) - Universal Background Model (UBM). The trained HMM is subsequently used to generate a temporally contiguous state alignment which are then modeled in a hybrid deep neural network (DNN) model. For the purpose of testing, we use the frame level HMM state posteriors obtained from the DNN as features for the ZeroSpeech challenge task. The minimal pair ABX error rate is measured for both the within and across speaker pairs. With several experiments on multiple languages in the ZeroSpeech corpus, we show that the proposed HMM based posterior features provides significant improvements over the baseline system using MFCC features (average relative improvements of 25% for within speaker pairs and 40% for across speaker pairs). Furthermore, the experiments where the target language is not seen training illustrate the proposed modeling approach is capable of learning global language independent representations.

Classification is one of the most popular techniques in the data mining area. In supervised learning, a new pattern is assigned a class label based on a training set whose class labels are already known. This paper proposes a novel classification algorithm for time series data. In our algorithm, we use four parameters and based on their significance on different benchmark datasets, we have assigned the weights using simulated annealing process. We have taken the combination of these parameters as a performance metric to find the accuracy and time complexity. We have experimented with 6 benchmark datasets and results shows that our novel algorithm is computationally fast and accurate in several cases when compared with 1NN classifier.

A novel pattern is an observation which is different as compared to the rest of the data. The task of novelty detection is to build a model which identifies novel patterns from a data set. This model has to be built in such a way that if a pattern is distant from the given training data, it should be classified as a novel pattern otherwise it should be classified into any one of the given classes. In this paper, we present two such new models, based on Probabilistic Neural Network for novelty detection. In the first model, we generate negative examples around the target class data and then train the classifier with these negative examples. In the second model, which is an incremental model, we present a new method to find optimal threshold for each class and if output value for a test pattern being assigned to a target class is less than the threshold of the target class, then we classify that pattern as a novel pattern. We show how decision boundaries are created when we add novelty detection mechanism and when we do not add novelty detection to our model. We show a comparative performance of both approaches.

The paper describes an algorithm for multi-label classification. Since a pattern can belong to more than one class, the task of classifying a test pattern is a challenging one. We propose a new algorithm to carry out multi-label classification which works for discrete data.We have implemented the algorithm and presented the results for different multi-label data sets. The results have been compared with the algorithm multi-label KNN or ML-KNN and found to give good results.

The problem of classification of time series data is an interesting problem in the field of data mining. Even though several algorithms have been proposed for the problem of time series classification we have developed an innovative algorithm which is computationally fast and accurate in several cases when compared with 1NN classifier. In our method we are calculating the fuzzy membership of each test pattern to be classified to each class. We have experimented with 6 benchmark datasets and compared our method with 1NN classifier.

Time series classification deals with the problem of classification of data that is multivariate in nature. This means that one or more of the attributes is in the form of a sequence. The notion of similarity or distance, used in time series data, is significant and affects the accuracy, time, and space complexity of the classification algorithm. There exist numerous similarity measures for time series data, but each of them has its own disadvantages. Instead of relying upon a single similarity measure, our aim is to find the near optimal solution to the classification problem by combining different similarity measures. In this work, we use genetic algorithms to combine the similarity measures so as to get the best performance. The weightage given to different similarity measures evolves over a number of generations so as to get the best combination. We test our approach on a number of benchmark time series datasets and present promising results.