25 research outputs found
Memory-efficient Stochastic methods for Memory-based Transformers
Get PDFTraining Memory-based transformers can require a large amount of memory and
can be quite inefficient. We propose a novel two-phase training mechanism and a
novel regularization technique to improve the training efficiency of
memory-based transformers, which are often used for long-range context
problems. For our experiments, we consider transformer-XL as our baseline model
which is one of memorybased transformer models. We show that our resultant
model, Skip Cross-head TransformerXL, outperforms the baseline on character
level language modeling task with similar parameters and outperforms the
baseline on word level language modelling task with almost 20% fewer
parameters. Our proposed methods do not require any additional memory. We also
demonstrate the effectiveness of our regularization mechanism on BERT which
shows similar performance with reduction in standard deviation of scores of
around 30% on multiple GLUE tasks
Experimental set up of PMU network and application of artificial neural network for PMU generated data analysis
No full textOne of the significant developments in Power systems is the implementation of new devices to ensure stability. The Power system is a dynamic system consists of Generation, Transmission, and Distribution. All these sectors of the Power System need to be monitored with accuracy, other than the system can be unstable within a blink of an eye. The newly developed device, Synchrophasor (PMU-Phasor Measurement Unit) is introduced in Power System for better stability and Wide area measurement (WAM). The high resolution of time-stamped data containing 30 to 60 samples per second provides high accuracy for power grid monitoring. The introduction of PMUs in the power system has increased the large volume of data. These data like the voltage, current, frequency, and phase angle impose an important rule for power quality. Measurements from the above parameters can help for better understanding of disturbances and attacks.
The overall aim of the dissertation is the design, development, and experimental setup of a small-scale PMU network and analyze data obtained from the PMU network. The analysis focuses on the traditional and application of Artificial Intelligence-based Techniques. Methods used in Artificial Intelligence help fast detection and extraction of proper information. For instance, traditional state estimation methods in SCADA (Supervisory Control and Data Acquisition) comprise a small volume of samples 2 to 4 samples per second. This kind of data has the probability to skip large events as the samples are low. But the data generated by PMU is large in volume and it is difficult to analyze the large volume of data with the traditional system. This kind of data is considered Big Data. Extraction of information from a large volume of data is tedious but the contents of the information are worthy. For experimental purposes, the large amount of PMU data from commercial grids are not easily available due to security reasons. At the same time, it is not easy to share total system data from any Opco as it contains network information which can make the system vulnerable, if it is a point of attack for an intruder.
This dissertation proposes the experimental setup of a PMU network to find out different steps of implementing PMUs in a power system. On the other hand, data acquisition is not the final step. Receiving raw data in a PDC (Phasor Data Concentrator) needs to be transformed in a meaningful format for analysis. Another important contribution of this work is to implement a Cyber-Physical Platform to find the characteristic of PMU data when subject to external attack and find the scenario of how the attack propagates.
The main contributions of the dissertation are as follows:
A newly designed Synchrophasor network consisting of seven PMUs in different locations and provides connectivity to a server called PDC (Phasor data concentrator). The connectivity issues comply with IEEE C 37.118 protocol and different IP issues as well as firewall rules to transmit data over a large network. Several barriers need to overcome to make up a smooth data propagation network.
As PMU generates a large volume of data, the data need to be archived and transform into a readable format. Different steps were followed inclusion of application of Artificial Intelligence models to analyze the data. Each dataset needs pre-cleaning and pre-preparation for analysis. All these kinds of analysis were completed with the application of different techniques and programming using R, Python and Matlab.
At the end of the experiment, the PMU based network was also used to develop a Cyber-Physical Platform to find out the scenario after the attack. This is a systematic way to intrude data and find the nature of the trend which will help to discriminate the scenarios of attack and traditional fault.
This dissertation is a combination of the application of different components of the power system as well as the theoretical contribution, application of Microgrid structure to enhance the power system Big data analysis. Last of all an initiative was taken for Cyber physical platform for intrusion detection from PMU data in case of any attack.Embargo status: Restricted to TTU community only. To view, login with your eRaider (top right). Others may request the author grant access exception by clicking on the PDF link to the left
Large-scale generation of human ipsc-derived neural stem cells/early neural progenitor cells and their neuronal differentiation
No full textTo Determine the Prevalence of Cysts and Tumors in Dental College
No full text
This study aimed to ascertain the prevalence of cysts and tumors among patients attending a dental college. Utilizing retrospective data analysis, a significant sample size was examined. Findings reveal critical insights into occurrence rates, contributing to improved diagnostic and therapeutic strategies within the dental community. 
