Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
6 results
Search Results
Item Human Activity Recognition in Smart Home using Deep Learning Techniques(Institute of Electrical and Electronics Engineers Inc., 2021) Kolkar, R.; Geetha, V.To understand the human activities and anticipate his intentions Human Activity Recognition(HAR) research is rapidly developing in tandem with the widespread availability of sensors. Various applications like elderly care and health monitoring systems in smart homes use smartphones and wearable devices. This paper proposes an effective HAR framework that uses deep learning methodology like Convolution Neural Networks(CNN), variations of LSTM(Long Short term Memory) and Gated Recurrent Units(GRU) Networks to recognize the activities based on smartphone sensors. The hybrid use of CNN-LSTM eliminates the handcrafted feature engineering and uses spatial and temporal data deep. The experiments are carried on UCI HAR and WISDM data sets, and the comparison results are obtained. The result shows a better 96.83 % and 98.00% for the UCI-HAR and WISDM datasets, respectively. © 2021 IEEE.Item IoT-based Human Activity Recognition Models based on CNN, LSTM and GRU(Institute of Electrical and Electronics Engineers Inc., 2022) Kolkar, R.; Singh Tomar, R.P.; Vasantha, G.Smartphones' ability to generate data with their inbuilt sensors has made them used for Human Activity Recognition. The work highlights the importance of Human Activity Recognition (HAR) systems capable of sensing human activities like the inertial motion of a human body. The sensors are worn on a body part and tracked from whole-body motions and monitoring. Real-time signal processing is used to sense human body movements using wearable sensors. The work aims to provide opportunities for promising health applications using IoT. There are many challenges to recognising human activities, including accuracy. This work analyses Human Activity recognition concerning CNN, LSTM, and GRU deep learning models to improve the accuracy of the human activity recognition in the UCI-HAR and WISDM datasets. The comparative analysis shows promising results for Human activity recognition. © 2022 IEEE.Item Comparative Study of Pruning Techniques in Recurrent Neural Networks(Springer Science and Business Media Deutschland GmbH, 2023) Choudhury, S.; Rout, A.K.; Pragnesh, T.; Mohan, B.R.In recent years, there has been a drastic development in the field of neural networks. They have evolved from simple feed-forward neural networks to more complex neural networks such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs are used for tasks such as image recognition where the sequence is not essential, while RNNs are useful when order is important such as machine translation. By increasing the number of layers in the network, we can improve the performance of the neural network (Alford et al. in Pruned and structurally sparse neural networks, 2018 [1]). However, this will also increase the complexity of the network, and also training will require more power and time. By introducing sparsity in the architecture of the neural network, we can tackle this problem. Pruning is one of the processes through which a neural network can be made sparse (Zhu and Gupta in To prune, or not to prune: exploring the efficacy of pruning for model compression, 2017 [2]). Sparse RNNs can be easily implemented on mobile devices and resource-constraint servers (Wen et al. in Learning intrinsic sparse structures within long short-term memory, 2017 [3]). We investigate the following methods to induce sparsity in RNNs: RNN pruning and automated gradual pruning. We also investigate how the pruning techniques impact the model’s performance and provide a detailed comparison between the two techniques. We also experiment by pruning input-to-hidden and hidden-to-hidden weights. Based on the results of pruning experiments, we conclude that it is possible to reduce the complexity of RNNs by more than 80%. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item Optimizing Performance of OpenMP Parallel Applications through Variable Classification(Institute of Electrical and Electronics Engineers Inc., 2024) Kumar, S.; Talib, M.OpenMP provides a versatile framework for parallel computing, allowing developers to transform sequential programs into parallel applications for shared-memory architectures efficiently. One of the central challenges in this transformation lies in accurately identifying appropriate parallel constructs and clauses, which are critical for maximizing performance and ensuring the correctness of the resulting parallel code. A particularly intricate aspect of this process is the classification of variables according to their data-sharing semantics, including first-private, private, last-private, shared, and reduction clauses. Manual classification is laborintensive and significantly susceptible to errors as the program's scale and complexity grow. Although various tools have been developed to assist with variable classification, they often rely on extensive data-dependence analyses and rigid classification schemes, limiting their effectiveness when applied to large-scale programs with complex scoping requirements. This paper presents a novel, cost-effective approach to automate and enhance the accuracy of variable classification in OpenMP parallelization. By reducing the manual effort required and improving the precision of parallel construct insertion, this approach aims to significantly optimize the performance of parallel applications, thereby advancing the utility and accessibility of OpenMP for a wide range of computational tasks. © 2024 IEEE.Item Detection of injections in API requests using recurrent neural networks and transformers(Inderscience Publishers, 2022) Sujan Reddy, A.; Rudra, B.Application programming interfaces (APIs) are playing a vital role in every online business. The objective of this study is to analyse the incoming requests to a target API and flag any malicious activity. This paper proposes a solution based on sequence models and transformers for the identification of whether an API request has SQL injections, code injections, XSS attacks, operating system (OS) command injections, and other types of malicious injections or not. In this paper, we observe that transformers outperform B-RNNs in detecting malicious activity which is present in API requests. We also propose a novel heuristic procedure that minimises the number of false positives. We observe that the RoBERTa transformer outperforms and gives an accuracy of 100% on our dataset. We observe that the heuristic procedure works well in reducing the number of false positives when a large number of false positives exist in the predictions of the models. © © 2022 Inderscience Enterprises Ltd.Item Can News Sentiment Improve Deep Learning Models for Nifty 50 Index Forecasting?(Institute of Electrical and Electronics Engineers Inc., 2025) Kotekar, C.S.; Mohan, R.; Kolukuluri, V.A stock index, such as the Nifty 50, offers diversified exposure and reduces the risk of investing in individual companies. Index price movements are influenced by internal and external factors, including political, economic, and environmental developments, as well as historical trends. The relationship between news sentiment and the Nifty50 return has not been thoroughly studied. This study examines whether financial news sentiment affects index movements and how sentiment can enhance the prediction of next-day returns. Polarity and subjectivity are extracted from financial news using pre-trained transformer models. Deep learning models, including LSTM, GRU, SimpleRNN, and temporal Kolmogorov-Arnold network (TKAN), are trained on return sign, polarity, and subjectivity using a five-day rolling window to forecast the next-day index return sign. Experimental results demonstrate that the proposed approach outperforms baseline methods, achieving a 5.2% improvement in average accuracy. Incorporating polarity and historical return signs enhances performance across all models. By employing a focused feature set, domain-specific sentiment analysis, and a streamlined architecture, the model achieves superior predictive accuracy. Causal analysis and Shapley Additive Explanations (SHAP) reveal that polarity exhibits a causal effect on returns, while subjectivity does not. The study has practical significance, offering day traders and short-term investors timely, data-driven insights to manage risk and make informed investment choices. © 2013 IEEE.