Search

With the burden of deaths and disability caused by neurological disorders like Alzheimer's disease, Parkinson’s disease, and epilepsy on the rise, speeding up the novel neurotherapeutic drug discovery process becomes crucial. Determining whether a drug is blood-brain barrier permeable or not is a prerequisite for discovering all central nervous system (CNS) drugs. Accurate and high throughput screening of clinical CNS drugs candidates based on their blood–brain barrier (BBB) permeability therefore becomes important, given the high failure rate of drug candidates due to their inability to penetrate the BBB. Molecules being richly annotated molecular graphs with atom, bond, and substructure features make graph representation learning for blood–brain barrier permeability worth studying. In this work, we holistically benchmark the performance and generalizability of 4 graph neural network models - Message Passing Neural Network (MPNN), Graph Convolutional Network (GCN), Graph Attention Network (GAT), and Graph Isomorphism Network (GIN) to quantitatively and qualitatively predict the BBB permeability of molecules.

This article describes how sequential data modeling is a relevant task in Cybersecurity. Sequences are attributed temporal characteristics either explicitly or implicitly. Recurrent neural networks (RNNs) are a subset of artificial neural networks (ANNs) which have appeared as a powerful, principle approach to learn dynamic temporal behaviors in an arbitrary length of large-scale sequence data. Furthermore, stacked recurrent neural networks (S-RNNs) have the potential to learn complex temporal behaviors quickly, including sparse representations. To leverage this, the authors model network traffic as a time series, particularly transmission control protocol / internet protocol (TCP/IP) packets in a predefined time range with a supervised learning method, using millions of known good and bad network connections. To find out the best architecture, the authors complete a comprehensive review of various RNN architectures with its network parameters and network structures. Ideally, as a test bed, they use the existing benchmark Defense Advanced Research Projects Agency / Knowledge Discovery and Data Mining (DARPA) / (KDD) Cup ‘99' intrusion detection (ID) contest data set to show the efficacy of these various RNN architectures. All the experiments of deep learning architectures are run up to 1000 epochs with a learning rate in the range [0.01-0.5] on a GPU-enabled TensorFlow and experiments of traditional machine learning algorithms are done using Scikit-learn. Experiments of families of RNN architecture achieved a low false positive rate in comparison to the traditional machine learning classifiers. The primary reason is that RNN architectures are able to store information for long-term dependencies over time-lags and to adjust with successive connection sequence information. In addition, the effectiveness of RNN architectures are shown for the UNSW-NB15 data set.

Describes a simple algorithm which has been developed for determining exact moments of top event failure probability from the moments of the basic events in a fault tree. This method requires neither Taylor series expansion of top event failure probability function nor its partial derivates to find these moments. This method has subsequently extended to systems with multistate components. Describes a general algorithm for the purpose. Provides several illustrations to highlight its usefulness.

“Parallel Implementation of MPEG-4 Encoding over a Cluster of Wrokstations” is a research project and a project proposal to reduce the delay involved in the MPEG-4 encoding process. As a case study we take Amrita Vishwa Vidyapeetham which is a multicampus Deemed University, one of its own kinds in India. All the campuses are connected through satellite provided by ISRO (Indian Space Research Organization). A number of e-learning classes, guest lectures and meetings are conducted across the satellite network with the video being compressed in MPEG-4 standard. The project optimizes the system by parallelizing the encoding process. For this a cluster of machines is created and the video frames are distributed among these nodes. This follows the SPMD {Single Program Multiple Data) model. Dedicated nodes within the cluster perform the encoding process. while there is a node that distributes the video to be encoded over these nodes. Another node collects the encoded video and places them back in the original sequence. The two nodes can be utilized for the encoding process also.

A faulty gear affects the functionality of the machine. Thus, it is necessary to diagnose the faults at an initial stage so as to reduce the losses incurred. Vibration signals contain dynamic information about the health condition of a rotating machine. Many researches are based on FFT and have its own drawback with non-stationary signals like the ones from gears. Hence, there is a need for development of new methodologies to infer diagnostic information from such signals. The vibrations produced by gears from good and simulated faulty conditions can be used to detect the faults in these gears. In the present study, variational mode decomposition (VMD) was used as a new signal pre-processing technique along with decision trees which have provided good classification performance. The technique decomposes the signals into various modes by identifying a compact frequency support around its central frequency, such that adding all the modes reconstructs the original signal. Descriptive statistical features were extracted from VMD processed signals. J48 decision tree algorithm was used to identify the useful features and the selected features were classified using the decision trees namely, J48, random tree and decision stump algorithm. The performance analyses of various algorithms are discussed in detail.