M. S. Ramaiah Institute of Technology

About: M. S. Ramaiah Institute of Technology is a based out in . It is known for research contribution in the topics: Feature extraction & Photoluminescence. The organization has 2853 authors who have published 2434 publications receiving 23507 citations.

Statistical Analysis of Spectral Entropy Features for the Detection of Alcoholics Based on Elecroencephalogram (EEG) Signals

Abstract: EEG happens to be an important tool for brain study providing a non- invasive and cost effective method to detect the effects of alcohol on the human brain. This paper highlights the importance of statistical analysis to determine the usefulness of spectral entropy features for discriminating alcoholics from healthy subjects. The open source EEG database consisting of 10 alcoholic and 10 control subjects recordings under visual stimulus is considered for the study. The EEG signal is preprocessed to remove eye blink artefact using independent component analysis (ICA) and the gamma sub band is extracted by using an elliptic band pass filter to obtain the visually evoked pattern (VEP) signal. The spectral entropy (SEN) coefficients are computed on all the 61 VEP signals of each subject. To obtain a statistical measure of SEN coefficients suitability for classifying the alcoholic EEG, ANOVA tests are performed. Results show that the test exhibits interesting observations in the form of p-value 0.05 for the remaining channels. This study may help in identifying those significant channels (p<0.05) which contribute to the classification of both the groups.

Kalman filter based multiple sensor data fusion in systems with time delayed state

Abstract: Obtaining accurate data in any system is a challenging problem. Multi-sensor data fusion is a widely used technique to improve the accuracy. In this paper, measurement level fusion, covariance union fusion, and state vector fusion based on Kalman filters for systems with delayed states is presented. The obtained Kalman filter based data fusion algorithms for time delayed systems are then modified to handle the missing measurements which is a common problem in wireless sensor networks. The simulation results of the data fusion algorithms obtained using MATLAB are compared. The performances of the fusion filters are found to be similar and the choice of a fusion algorithm would only depend on the particular application.

Design of optimized reversible Binary and BCD adders

Abstract: Reversible logic has emerged as a possible low cost alternative to conventional logic in terms of speed, power consumption and computing capability. An adder block is a very basic and essential component for any processor and optimized design of these adders' results in efficient processors. In this work we propose optimized Binary adders and BCD adders. The adders designed in this work are optimized for Quantum cost, Delay and Area. A modified BCD adder is also proposed which removes redundancy in the circuit and acts as most efficient BCD adder. Here we explore the use of Negative control lines for detecting overflow logic of BCD adder which considerably reduces Quantum cost, delay and gate count which result in high speed BCD adder with optimized area which give way to lot of scope in the field of reversible computing in near future.

A computationally intelligent agent for detecting fake news using generative adversarial networks

Abstract: The explosion of smartphones has led to major changes in the perception of news, including the use of social media to propagate fake news and system-generated content without proper validation. The existing solutions to this problem employ the use of technologies like machine learning. The identification of unverified articles is a classification problem where, given a document, the system classifies it as “fake” or “valid.” This process involves the collection of large amounts of text corpus of both valid and fake news articles. The issue with these existing systems is the validity of the data aggregated from different sources. This can lead to the problem of human bias in labeling the articles collected. As an initial step to reduce this bias, this project proposes a fake news detection framework, which uses a generative modeling technique. In this technique, a generator–discriminator (G–D) setup is employed. The G–D model is extended from the SeqGAN model. The generator generates new data instances, while the discriminator evaluates them for authenticity. When the models train competitively, the generator becomes better at creating synthetic samples and the discriminator gets better at identifying the synthetic samples. Thus, a data set is synthesized by merging the real articles with the articles generated by the generator, which is trained using the G–D setup. This data set is then used to train the agent (classifier) to identify the articles as “fake” or “valid.”

Mechanism of cube grain nucleation during recrystallization of deformed commercial purity aluminium

Abstract: Cube texture is a sharp recrystallization texture component infcc metals like aluminium, copper, etc. It is described by an ideal orientation i.e. (100) (100). The subject of cube texture nucleation i.e. cube grain nucleation, from the deformed state of aluminium and copper is of scientific curiosity with concurrent technological implications. There are essentially two models currently in dispute over the mechanism of cube grain nucleation i.e. the differential stored energy model founded on the hypothesis proposed by Ridha and Hutchinson and the micro-growth selection model of Dugganet al. In this paper, calculations are made on the proposal of Ridha and Hutchinson model and the results are obtained in favour of the differential stored energy model. It is also shown that there is no need for the micro-growth model.