LAL UPENDRA PRATAP SINGH
@soa.ac.in
Assistant Professor, Computer Science and Engineering
Institute of Technical Education and Research, Siksha O' Anusandhan University
RESEARCH INTERESTS
Deep Learning, Transfer Learning, Machine Learning, Computer Vision and Optimization
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Anshul Mishra, Upendra Pratap Singh, and Krishna Pratap Singh
Springer Science and Business Media LLC
Rishi Narang and Upendra Pratap Singh
IEEE
Sequence modelling has shown tremendous potential in solving real-world sequence prediction tasks like speech recognition, time series forecasting, and context identification. However, most of these sequence models are trained on univariate datasets and cannot leverage the information available in a multivariate setting. Moreover, the prediction/decision made by these models is not interpretable; consequently, the end users are unaware of the different steps involved in reaching that prediction/decision and cannot determine if the model aligns with the business and ethical values. This work investigates the performance of different sequence learners trained in a multivariate setting for the sales forecasting task. Specifically, different sequence models, including vanilla LSTM, stacked LSTM, bidirectional LSTM, and convolution neural networkbased-LSTM, have been trained on the Walmart dataset, and a comparative analysis of their performance using mean squared error (MSE) and weighted mean absolute error (WMAE) metric is reported. For training the learners in a multivariate setting, relevant features have been identified using exploratory data analytics. Furthermore, these sequence models are made interpretable using the Local Interpretable Model Agnostic Explanation (LIME) model to explain away the key variables involved in the prediction task. Empirical results obtained on the Walmart sales dataset established that the performance of the stacked LSTM model is superior to other learners. Additionally, the stacked model being the most generalizable, is complemented by the LIME module to explain away its predictions using the relevant features.
Upendra Pratap Singh, Krishna Pratap Singh, and Manoj Thakur
Springer Science and Business Media LLC
Upendra Pratap Singh, Krishna Pratap Singh, and Manoj Thakur
Springer Science and Business Media LLC
Upendra Pratap Singh, Krishna Pratap Singh, and Manoj Thakur
Springer Science and Business Media LLC
Ramji Jaiswal, Upendra Pratap Singh, and Krishna Pratap Singh
IEEE
Rashi Agrawal, Upendra Pratap Singh, and Krishna Pratap Singh
Springer Singapore
Anupam Prakash, Piyush Sharma, Indrajeet Kumar Sinha, and Upendra Pratap Singh
IEEE
Covid-19, caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) virus, has presented tough times for countries all over the world with number of cases and casualties running in millions. While virologists and doctors have spent sleepless nights to come up with a potent vaccine, the work life of government personnel including administrative staffs, hospital employees etc. has not been any easier. Amidst this turmoil, the common question crossing every mind is concerned with the statistics about this infection including expected number of infections, peak prediction etc. We try to answer these questions by analyzing the time series data of Covid-19 infections for certain hard-hit countries and states in India. A series of machine and deep learning models have been built to capture the infection distribution so that these models could predict the fate of this infection in the near future. We also make an attempt to predict the time when active cases would cease to increase.
Upendra Pratap Singh, Swapnil Chavan, Sahil Hindwani, and Krishna Pratap Singh
Springer Singapore
Upendra Pratap Singh, Kaustubh Rakesh, Rishabh, Vipul Kumar, and Krishna Pratap Singh
IEEE
Zero shot learning seeks to learn useful patterns in the source domain and identify novel concepts in the target domain. This transfer learning paradigm has recently gained immense popularity given the inherent limitations in data acquisition and subsequent annotation for a task (or domain). While typical zero shot learning methods utilize all the classes (and their instances) in the source domain in a passive way, we, in our work, actively use only a handful of relevant classes for learning in the source domain. With this intelligent data subset, we jointly learn the source and target domain parameters using coupled semantic autoencoders. This joint learning reduces the projection domain shift problem. We further extend the above model for word embedding based semantic space as well. For classes with no word embedding, we have solved prototype sparsity problem by training a neural network with all classes that has one. This neural network seeks to learn a mapping from attribute space to word embedding space. Experiments on AWA2 and CUB-UCSD datasets confirm the superiority of our hybrid approach over state of art methods by up to 16% and 8% in attribute and word embedding space respectively.
Anupam Agrawal, Manisha Malik, and Lal Upendra Pratap Singh
IEEE
Color blindness, also called Color Vision Deficiency or CVD affects males more than it affects females round the globe. According to Apollo Hospital, CVD is very common with more than 10 million cases per year in India. Inherited color blindness cannot be treated. If any other condition is the cause of color blindness then the same could be compensated to help the color blind person. This paper presents a toolkit called CBAT that finds out if a person is suffering from color blindness or not using the standard Ishihara test. In case, color blindness is diagnosed, the same is compensated using LMS based Daltonization algorithm. The three types of color blindness considered in this paper are Protanopia, Deuteranopia and Tritanopia. The software tool was tested on large number of students present in our campus some of whom were actually diagnosed with color blindness in their past. The results obtained using our tool were fairly satisfactory.
Lal Upendra Pratap Singh and Anupam Agrawal
IEEE
The need for establishing authenticity of digital images is becoming inevitably important given the ease with which images may be tempered. Moreover, the images are tampered with such expertise that it is impossible, at least visually, to figure out if they are tampered or not. In recent years, copy-move forgery has emerged as one of the most researched topics in the field of image forensics. The common detection techniques used are either block based (for smoothed regions) or keypoint based (for non-smoothed regions), each having its own share of pros and cons. While a block based approach provides higher accuracy, it is computationally taxing and fails to handle geometric transformations. Similarly, a Keypoint based approach fails to deal with smoothed regions. Our hybrid approach handles these limitations in an intelligent and adaptive way. The image fed to our software tool is adaptively segmented into semantically meaningful non-overlapped regions/segments using Simple Linear Iterative Clustering (SLIC) algorithm. Feature points as keypoints are extracted using the Scale Invariant Feature Transform (SIFT) algorithm. Given these keypoints, a segment is classified as either smoothed or non-smoothed depending upon a predetermined threshold. Once done with this classification, the proposed hybrid approach engages one of the aforementioned detection strategies to detect image forgery. Accordingly, a block based approach is implemented using Zernike moments with the matching algorithm based on Euclidean distances, while a keypoint based approach is implemented using SIFT features in tandem with the FLANN matching algorithm for the detection of matching pairs. The proposed hybrid approach has been compared with the individual approaches for an optimal value of threshold. Experimental results on different types of original as well as forged images establish that our proposed approach is able to detect image forgery in smoothed and non-smoothed images with reasonable accuracy and recall.
Lalaram Arya, Upendra Pratap Singh, Anupam Shukla, and Ritu Tiwari
IEEE
Recent advancements and efficient training procedures in deep neural networks (DNNs) have significantly outperformed the hidden Markov model-Gaussian mixture model (HMM-GMM). The performance of DNNs can further be improved should it be given better phonetic context information. This is manifested by state specific vectors (SSV) of subspace Gaussian mixture model (SGMM). In this paper, we use the state specific vectors of SGMM as features to provide additional phonetic context information to the DNN framework. The state specific vectors are aligned with each observation vector of the training data to form the state specific vector (SSV) feature set. The combination of linear discriminant analysis (LDA) feature sets and state specific feature sets are then used as input features to train the DNN framework. Relative improvement of up to 4.13% is obtained on Hindi database using DNN trained with a combination of state specific feature sets and LDA feature sets compared to the DNN trained only with LDA feature sets. Since state specific vectors provide extra information about the phonetic context, they show improved results when combined with DNN framework. In this paper, we also investigate the performance of speech recognition on different training data selection strategies. The idea is to implement an approach that maximizes the information content in the training corpus. The experiments in this paper are carried on the training data set having maximum information content.