Mohit Agarwal
@bennett.edu.in
Assistant Professor with School of Computer Science and Engineering
Bennett University
RESEARCH INTERESTS
Deep learning
Image Processing
Model Compression
Scopus Publications
Scopus Publications
Mohit Agarwal, Amit Kumar Dwivedi, Dibyanarayan Hazra, Preeti Sharma, Suneet Kumar Gupta, and Deepak Garg
Elsevier BV
Vivek Mehta, Mohit Agarwal, and Rohit Kumar Kaliyar
Springer Science and Business Media LLC
Mohit Agarwal, Suneet K. Gupta, and K. K. Biswas
Springer Science and Business Media LLC
Mohit Agarwal, Garima Jaiswal, Rohit Kumar Kaliyar, Akansha Singh, Krishna Kant Singh, S. S. Askar, and Mohamed Abouhawwash
Institute of Electrical and Electronics Engineers (IEEE)
Federated learning uses the concept of decentralized training of n number of local clients for a small number of epochs say 2-5, and then averaging the learned weights of all local clients, and evaluating on test dataset with the average weights loaded to a global model. The train dataset is split into n clusters and each cluster acts as a distributed data for each local model. Each round of weight averaging and then uploading the average weights on each local client for further training is called communication round and it was observed that similar accuracy can be obtained with a lesser amount of training time. In this paper, instead of averaging the weights, a weighted mean concept was developed where the PSO vector helps to find the weight values for the best accuracy of a global model. It was found that PSO can help in two ways by bettering the accuracy and also reducing the training time. The proposed approach can enhance the performance of pre-trained models like AlexNet, VGG16, InceptionV3, and ResNet50 on CIFAR-10 and CIFAR-100 datasets. The maximum increase was found with VGG16 of around 26.01% for CIFAR-10 and 26.84% for CIFAR-100. Similarly, on the Tomato dataset, AlexNet accuracy can be increased by 28.56%. Multi-modal model accuracy on the fake news dataset was also enhanced by 8.21%.
Salvadi Kasturi, Jagendra Singh, Mohit Agarwal, Ashwini Kumar, Komal Mishra, and Yogita Sharma
IEEE
This research is aimed at discussing the critical meeting point of healthcare and artificial intelligence, including the need to protect the required treatment solution. Investigation into the utility of various deep learning techniques such as VGG 16, VGG 19, Convolutional Neural Networks (CNN), and Recurrent Neural Network (RNN) for the prediction of liver cancer stages and their respective treatments has been completed. The generation of 2340 of medical images from multiple sources, including the internet and records of hospitals, as a dataset was performed in a secure cloud. The security measures have allowed only authorized personnel to access the data and ensured the privacy and secrecy of data. The created dataset was then separated into two parts: training, which consisted of training the model on 70% of images, and testing, which implied that 30% of images were used to test the generated model. For each of the run models, performance evaluation measurements such as accuracy, precision, recall, and F1 score were calculated. The results have shown the highest efficiency of VGG 19 model, which resulted in 98.44% of accuracy. For the generated model, a fictitious ROC curve and specificity rate were also provided to showcase the discrimination abilities of each of the submitted models. The results show that the utility of neural network-based personalized medicine is beneficial for the further progression of the early prediction stage of liver cancer and its treatment. The result showed that the proposed models has successfully contributed to the field of healthcare informatics and further opportunities for artificial intelligence application to increase patient outcomes in the form of personalized medicine that may be performed.
Mohit Agarwal, Rohit Kr Kaliyar, and Vivek Mehta
IEEE
Social media is in wide usage among different sections of society. It helps them to connect with each other and can also be used as a medium to spread fake news. Such misinformation can be very dangerous as it can misguide youth and other age groups to take adverse steps leading to violence and other crimes. Hence efforts are needed to classify social media news as fake or real. Earlier lot of efforts have been made to classify news articles from their text by creating natural language processing deep learning models. Since, these days social media depends a lot on images hence it is needed that both text and associated images must be analyzed to know the fakeness of a news article. Thus multimodal deep learning models have been introduced which can extract features from both images and associated text and then fuse these features to pass to a classification output layer. Using this approach it was found that classification with both data as input an accuracy of 74.4 % was obtained on a publicly available dataset.
Lopamudra Mohanty, Ashish Kumar, Vivek Mehta, Mohit Agarwal, and Jasjit S. Suri
Springer Science and Business Media LLC
Vibhuti Dabas, Garima Jaiswal, Mohit Agarwal, Ritu Rani, and Arun Sharma
Springer Science and Business Media LLC
Arun Kumar, Mohit Agarwal, and Mohd Aquib
Springer Nature Switzerland
Mohit Agarwal, Amit Kumar Dwivedi, Suneet Kr. Gupta, Mohammad Najafzadeh, and Mani Jindal
Springer Nature Switzerland
Venkateshh Moningi, Mohit Agarwal, Mainak Biswas, and Jasjit S Suri
IOP Publishing
Sanjeev Kumar, Suneet Kumar Gupta, Umesh Gupta, and Mohit Agarwal
Springer Science and Business Media LLC
Mohit Agarwal, Suneet K. Gupta, Mainak Biswas, and Deepak Garg
Springer Science and Business Media LLC
Mohit Agarwal, Suneet Kr. Gupta, and K. K. Biswas
Springer Science and Business Media LLC
Mohit Agarwal, Suneet K. Gupta, and K. K. Biswas
Springer Science and Business Media LLC
Dibyanarayan Hazra, Suneet Kumar Gupta, Umesh Gupta, and Mohit Agarwal
ACM
Each cuisine required tomato in their kitchen for various food items and this makes tomato most popular crop worldwide and India is in second rank in terms production of tomato. Now a days, production of tomato goes down because of various diseases and to treat these diseases farmer needs to have extensive prior knowledge about the pathogen and along with various factor which promote the disease in the tomato. Due to lack of knowledge, the disease spreads rapidly and destroys all crops. To fill this gap, deep learning (DL) has been playing an important role, and there is much research on DL, how it can be used in medical industry and the agriculture industry for the identification of disease using images. There is a limitation for DL model that it does not work well with small dataset and huge amount of samples are required to train the model. Moreover, the data are not shared openly for security or for any other reason. Therefore, to overcome this challenge a Federated Learning (FL) based approach has been presented in the article. In FL, a deep learning model is shared with organizations which having the data and train the model. After training, the model information is shared with a centralized server which designs a generalized model. After getting the generalized model, it is shared with all other sites. The process is repeated until a generalized model is not designed and well-suited with all the sites. In our study, we tested our model on a tomato leaf disease data set using FL methodology with 10 clients and achieved the best precision with 88. 01%.
Suneet Kumar Gupta, Dibyanarayan Hazra, Mohit Agarwal, Simar Preet Singh, Rahul Dass, and Deepika Pantola
IEEE
In the last decades, the domain of IoT has been explored by research community due to its vast real time applications. A combination of deep learning and IoT is well accepted worldwide as using deep learning, IoT devices can be easily converted into intelligence devices. Moreover, these devices are capable enough to take the decision based on real-time data. However, deployment of deep learning model is not so easy in IoT devices as these devices are constraint with limited computational power and storage space. Generally, deep learning architectures are large in terms of storage space, and due to the complication of model, it required resources to generate the output. To overcome the storage space and the large resource barrier, we proposed the method based on the particle swarm optimization technique for compression of the UNet architecture for its easy deployment on IoT devices for semantic segmentation usages. In this paper, all the intermediate steps involved for this compression of UNet using PSO is well explained with suitable examples. Experimentally, it has been proven that the proposed algorithm compresses the UNet architecture in the chest radiograph data set by 77% after 0. 68% decrease in accuracy with an improvement in the inference time by 2.23X.
Rohit Kumar Kaliyar, Anurag Goswami, Ujali Sharma, Kanika Kanojia, and Mohit Agrawal
IEEE
The phenomenal rise of social media platforms like Twitter, Facebook, Instagram, and Reddit has led to the blending of native languages or regional tongues with English for the purpose of improving communication in linguistically open geographic regions around the world. There are many ways in which Holocaust denial can lead to an increase in violence, from direct assault to purging out of compassion. Online, people are very hostile to one another. Distinguishing between language that incites hatred and language that is disparaging is a fundamental challenge in the categorization and tracking of extremely toxic lexical features. Our research focuses on identifying harmful tweets composed in Hinglish, a fusion of Hindi and the Roman alphabet. We propose a system in this paper for classifying tweets as either abusive, neutral, or offensive. The help of Hindi-English offensive tweet dataset is comprised of tweets written in the code-transferred language of Hindi and is further subdivided into three groups: neutral, abusive, and hateful. We studied the abusive and hate speech dataset with transfer learning and pre-trained the proposed model on Hinglish-processed English tweets. With our proposed model, we were able to improve accuracy to 98.54 percent.
Mohit Agarwal, Lokesh Kumar Sharma, Suneet Kumar Gupta, Deepak Garg, and Mani Jindal
Springer Nature Switzerland
Mohit Agarwal, Simar Preet Singh, Rohit Kaliyar, Suneet Kumar Gupta, Deepak Garg, and Mani Jindal
Springer Nature Switzerland
Mohit Agarwal, Rohit Kr. Kaliyar, and Suneet Kr. Gupta
IEEE
Medical diagnosis has been widely enhanced by the deep learning methods using medical images such as X-rays, CT scans and MRI scans. The physical diagnosis by viewing the images can vary from one doctor to another. The deep learning based methods are found to produce more accurate results. This article proposes usage of transfer learning based pre-trained models such as VGG19, MobileNet, AlexNet, etc. Several traditional machine learning methods such as Logistic Regression, k-Nearest Neighbours (k-NN), Decision Trees (DT), Random Forest (RF), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), Naive Bayes have also been used to show different computer based methods for medical diagnosis. With the advent of robot based devices in various medical fields a need is created to deploy these models on low memory devices. Hence the pre-trained models which need more than 100 MBs space are compressed using Differential Evolution algorithm to reduce the space need to few KBs with similar accuracy.
Sanagala S. Skandha, Mohit Agarwal, Kumar Utkarsh, Suneet K. Gupta, Vijaya K. Koppula, and Jasjit S. Suri
Springer Science and Business Media LLC
Mohit Agarwal, Sushant Agarwal, Luca Saba, Gian Luca Chabert, Suneet Gupta, Alessandro Carriero, Alessio Pasche, Pietro Danna, Armin Mehmedovic, Gavino Faa,et al.
Elsevier BV
Isha Puthige, Tanveer Hussain, Suneet Gupta, and Mohit Agarwal
Elsevier BV
Hikmat Yar, Tanveer Hussain, Mohit Agarwal, Zulfiqar Ahmad Khan, Suneet Kumar Gupta, and Sung Wook Baik
Institute of Electrical and Electronics Engineers (IEEE)
Vision-based fire detection systems have been significantly improved by deep models; however, higher numbers of false alarms and a slow inference speed still hinder their practical applicability in real-world scenarios. For a balanced trade-off between computational cost and accuracy, we introduce dual fire attention network (DFAN) to achieve effective yet efficient fire detection. The first attention mechanism highlights the most important channels from the features of an existing backbone model, yielding significantly emphasized feature maps. Then, a modified spatial attention mechanism is employed to capture spatial details and enhance the discrimination potential of fire and non-fire objects. We further optimize the DFAN for real-world applications by discarding a significant number of extra parameters using a meta-heuristic approach, which yields around 50% higher FPS values. Finally, we contribute a medium-scale challenging fire classification dataset by considering extremely diverse, highly similar fire/non-fire images and imbalanced classes, among many other complexities. The proposed dataset advances the traditional fire detection datasets by considering multiple classes to answer the following question: what is on fire? We perform experiments on four widely used fire detection datasets, and the DFAN provides the best results compared to 21 state-of-the-art methods. Consequently, our research provides a baseline for fire detection over edge devices with higher accuracy and better FPS values, and the proposed dataset extension provides indoor fire classes and a greater number of outdoor fire classes; these contributions can be used in significant future research. Our codes and dataset will be publicly available at https://github.com/tanveer-hussain/DFAN.