Botnets are recognized as one of the most dangerous threats to the Internet infrastructure. They are used for malicious activities such as launching distributed denial of service attacks, sending spam, and leaking personal information. Existing botnet detection methods produce a number of good ideas, but they are far from complete yet, since most of them cannot detect botnets in an early stage of their lifecycle; moreover, they depend on a particular command and control (C&C) protocol. In this paper, we address these issues and propose an online unsupervised method, called BotOnus, for botnet detection that does not require a priori knowledge of botnets. It extracts a set of flow feature vectors from the network traffic at the end of each time period, and then groups them to some flow clusters by a novel online fixed-width clustering algorithm. Flow clusters that have at least two members, and their intra-cluster similarity is above a similarity threshold, are identified as suspicious botnet clusters, and all hosts in such clusters are identi ed as bot infected. We demonstrate the effectiveness of BotOnus to detect various botnets including HTTP-, IRC-, and P2P-based botnets using a testbed network. The results of experiments show that it can successfully detect various botnets with an average detection rate of 94: 33% and an average false alarm rate of 3: 74%.

Embedding learning is an essential issue in Natural Language Processing (NLP) applications. Most existing methods measure the similarity between text chunks in a context using pre-trained word embedding. However, providing labeled data for model training is costly and time-consuming. So, these methods face downward performance when limited amounts of training data are available. This paper presents an unsupervised sentence embedding method that effectively integrates semantic hashing into the Kernel Principal Component Analysis (KPCA) to construct embeddings of lower dimensions that can be applied to any domain. The experiments conducted on benchmark datasets highlighted that the generated embeddings are general-purpose and can capture semantic meanings from both small and large corpora.

Civil structures may experience unexpected loads and consequently damages during their life cycle. Damage identification has been a challenging inverse problem in structural health monitoring. The main difficulty is characterizing the unknown relation between the measurements and damage patterns. Such damage indicators would ideally be able to identify the existence, location, and severity of damages. In order to solve such problems, biologically inspired soft-computing techniques have gained traction. The most widely used soft-computing method, called neural networks is designed such that it can learn from data without a need of feature design process. Damage pattern can be detected using neural network. A deep unsupervised neural network can recognize patterns and extract features from data. In this paper a methodology is described for global and local health condition assessment of structural systems using vibration response of the structure. The model incorporates Fast Fourier Transform and unsupervised deep Boltzmann machine to extract features from the frequency domain of the recorded signals. Restricted boltzmann machine is a shallow neural network with two layer. First layer of restricted boltzmann machine called input layer and second layer of restricted boltzmann machine called hidden layer.Deep Boltzmann machine created by setting some restricted Boltzmann machine sequentional. Hidden layer of each restricted boltzmann machine is input layer of next restricted boltzmann machine. Each layer of restricted Boltzmann machine extract features form input data Recorded data divided to smaller vectors. Fast fourier transformation used to transform divided vectors into frequency domain.  A benefit of the proposed model is that it does not require costly experimental results to be obtained from a scaled version of the structure to simulate different damage states of the structure and only vibration response of the healthy structure is needed to training deep neural network. The input consists of a set of records obtained from the healthy state of the structure and another set of records with unknown health states. The model extracts information from both healthy and unknown sets to determine the health states of the unknown set. The healthy records are low intensity vibrations of the structure at least in one planar direction in the healthy state in the form of time series signals and The unknown records are low intensity vibrations of the structure on unknown state of health. Ambient vibrations can be due to wind, traffic, or human/pedestrian activities. An appropiate health index is defined and calculated for each part of the structure. The value of this index is between 0 and 1. The closer the value is to 1 the healthier the structure. To evaluate the efficiency of the proposed method a building structures with 35 story has been simulated in OPENSEES. Data collection should be selected appropriately to prevent errors. Obtained result demonstrate that proposed method has about 95 percent efficiency to predict damages and their severity. Different damage state put on due to three earthquakes with different severity. Structural health index calculated after each earthquake. Calculated structural health index demonstrate efficieency of proposed method for detecting damages and severity of damages.

Background: Accurate bone age assessment is essential for determining the actual degree of development and indicating a disorder in growth. While clinical bone age assessment techniques are time-consuming and prone to inter/intra-observer variability, deep learning-based methods are used for automated bone age estimation.Objective: The current study aimed to develop an unsupervised pre-training approach for automatic bone age estimation, addressing the challenge of limited labeled data and unique features of radiographic images of hand bones. Bone age estimation is complex and usually requires more labeling data. On the other hand, there is no model trained with hand radiographic images, reused for bone age estimation.Material and Methods: In this fundamental-applied research, the collection of Radiological Society of North America (RSNA) X-ray image collection is used to evaluate the efficiency of the proposed bone age estimation method. An autoencoder is trained to reconstruct the original hand radiography images. Then, a model based on the trained encoder produces the final estimation of bone age.Results: Experimental results on the Radiological Society of North America (RSNA) X-ray image collection achieve a Mean Absolute Error (MAE) of 9.3 months, which is comparable to state-of-the-art methods. Conclusion: This study presents an approach to estimating bone age on hand radiographs utilizing unsupervised pre-training with an autoencoder and also highlights the significance of autoencoders and unsupervised learning as efficient substitutes for conventional techniques.

Hyperspectral sensors have high capability in identifying objects by acquiring a large number of adjacent electromagnetic bands. Although This large number of bands makes it possible to approximate the more precise spectral curve of the material, it also brings some challenges. The difficulty in data transfer, the weak performance of conventional statistical classifications due to the limited number of training data, and the high processing time are the most important ones. Hence, different methods of dimensionality reduction are proposed for hyperspectral images. In the following article, an unsupervised feature extraction method is proposed based on the bands clustering technique. In the proposed method, after the prior image clustering and forming the prototype space with the aid of the clusters’ averages, the bands are clustered using the K-medoids clustering algorithm. In each cluster, four types of central tendency measures, mean, geometric mean, harmonic mean, and median are used to extract the final features. The experiments are conducted on the three real hyperspectral images with medium and high spatial resolution. Final results indicate that the classification results of the proposed method can reach (72. 12) which is 7% higher than the other four competitive methods, principal component analysis (PCA) (64. 39), wavelet (64. 58), feature selection method based on bands clustering based on variance (65. 30) and non-parametric weighted features extraction (NWFE) (64. 12).

With the rapid increase in the use of search engines, the need for developing more effective information retrieval and ranking methods has become critical. One of the key challenges in information retrieval is predicting query performance, which involves estimating how well a search engine can fulfill a user's information need. Accurate prediction of query performance allows search engines to take adaptive actions, such as query reformulation or ranking adjustment, to enhance retrieval effectiveness. Query Performance Prediction (QPP) methods fall into two main categories: pre-retrieval prediction and post-retrieval prediction. Pre-retrieval predictors estimate query difficulty before the retrieval process, relying on linguistic and statistical query features rather than retrieved documents. In contrast, post-retrieval prediction methods assess query performance based on the ranking list and document collection, providing deeper insights into retrieval effectiveness. In this study, we propose a novel unsupervised post-retrieval QPP method that evaluates query performance by analyzing the clustering behavior of retrieved documents. Our method defines five new metrics—CC, DCIC, DCNIC, DCNICR, and CCR— to measure the distribution and coherence of retrieved documents. These metrics help assess query difficulty by capturing how documents group into clusters, identifying outlier documents that do not fit well into clusters, and evaluating the overall structure of retrieved results. By leveraging these metrics, our approach provides a more fine-grained estimation of query performance without requiring human-labeled data. To evaluate the effectiveness of the proposed method, we conduct experiments on three datasets: TREC DL 2019, TREC DL 2020, and DL-Hard. The results demonstrate that our approach improves Spearman's correlation coefficient by 0.009 and 0.163 on the TREC DL 2019 and DL-Hard datasets, respectively. Additionally, it increases Pearson’s correlation coefficient by 0.037 on the TREC DL 2020 dataset compared to state-of-the-art unsupervised QPP methods. These improvements indicate that clustering-based QPP methods can effectively capture query difficulty and retrieval quality without the need for external supervision.