Working as a assistant professor in Computer science engineering department in Netaji Subhas University of Technology, Delhi, India. Research areas are adhoc networks, machine learning and image processing. PhD and masters from NIT Jaipur. Address: Jaipur, Rajasthan, India
IEEE Transactions on Neural Networks and Learning Systems, 2021
Recently introduced deep reinforcement learning (DRL) techniques in discrete-time have resulted i... more Recently introduced deep reinforcement learning (DRL) techniques in discrete-time have resulted in significant advances in online games, robotics, and so on. Inspired from recent developments, we have proposed an approach referred to as Quantile Critic with Spiking Actor and Normalized Ensemble (QC_SANE) for continuous control problems, which uses quantile loss to train critic and a spiking neural network (NN) to train an ensemble of actors. The NN does an internal normalization using a scaled exponential linear unit (SELU) activation function and ensures robustness. The empirical study on multijoint dynamics with contact (MuJoCo)-based environments shows improved training and test results than the state-of-the-art approach: population coded spiking actor network (PopSAN).
The challenges involved in the traditional cloud computing paradigms have prompted the developmen... more The challenges involved in the traditional cloud computing paradigms have prompted the development of architectures for the next generation cloud computing. The new cloud computing architectures can generate and handle huge amount of data, which was not possible to handle with the help of traditional architectures. Deep learning algorithms have the ability to process this huge amount of data and, thus, can now solve the problem of the next generation computing algorithms. Therefore, these days, deep learning has become the state-of-the-art approach for solving various tasks and most importantly in the field of recognition. In this work, recognition of city names is proposed. Recognition of handwritten city names is one of the potential research application areas in the field of postal automation For recognition using a segmentation-free approach (Holistic approach). This proposed work demystifies the role of convolutional neural network (CNN), which is one of the methods of deep lea...
Wireless Sensor Network comprises of hundreds or thousands of sensor nodes to sense the environme... more Wireless Sensor Network comprises of hundreds or thousands of sensor nodes to sense the environment collect data and send these data to a central location known as Base Station or Sink. Sensor Networks are very useful in many application such as military applications, environmental condition detection, whether prediction, measuring temperature, sound, wave, humidity and so on. Each sensor node in Sensor Networks operates on battery. In many scenarios, it is very difficult to recharge or replace battery of sensor nodes. So routing protocols for Wireless Sensor Networks should be as energy efficient as possible. Security is also very important in Wireless Sensor Networks. There may be possible to attack by intruders on routing protocols in Wireless Sensor Networks, and degrade the performance of networks. LEACH (Low Energy Adaptive Clustering Hierarchy) is one of the very energy efficient routing protocols, which uses randomized cluster rotation to distribute energy load among all sen...
The main purpose of this paper is to develop an efficient machine learning model to estimate the ... more The main purpose of this paper is to develop an efficient machine learning model to estimate the electric power load. The developed machine learning model can be used by electric power utilities for proper operation and maintenance of grid and also to trade electricity effectively in energy market. This paper proposes a machine learning model using gated recurrent unit (GRU) and random forest (RF). GRU has been employed to predict the electric power load, whereas RF has been used to reduce the input dimensions of the model. GRU has been estimating the load with good accuracy. RF reduces the input dimensions of the GRU that leads lightweight GRU model. The main benefits of the lightweight GRU models are less computation time and memory space. However, lightweight GRU models will loss small amount of accuracy comparing to the original GRU model. GRU along with RF has been used for the first for short load forecasting. All the machine learning model’s performance has been observed in stochastic environment. Impact of weekends on load forecasting also observed by considering the last 3-week load data.
2021 International Conference on Computer Communication and Informatics (ICCCI), 2021
In a future that is not too distant from today, humanoids are surely going to be an integral part... more In a future that is not too distant from today, humanoids are surely going to be an integral part of both our professional and private lives, assisting us with various tasks. Unlike normal robots that we may encounter in our everyday lives, humanoids are designed in specific manners to give them more human-like capabilities that enable them to perform complex tasks such as climbing a flight of stairs. In this paper, we present a Maze-Solving Algorithm which is a software developed specifically for the humanoid robot, NAO, and gives it the capability to enter and exit a maze autonomously. NAO is a next-gen humanoid bot developed by SoftBank Robotics using the power of AI. The bot is equipped with numerous sensors and cameras. Though various quantitative approaches were considered and experimented with, we stuck onto the one which had the least average time complexity of all after a thorough comparative study. We suggest an approach where the humanoid can detect and localize objects from a distance and take programmable decisions based on them. AI constantly tries to give robots human-thinking capabilities to make their decision-making skills similar to those of humans, if not better than them. This algorithm was developed taking into consideration how a human intellect would react rationally if he is stuck in a maze. The methodology used revolves primarily around the combined use of SONAR(Sound navigation ranging) and tactical sensors, and cameras equipped within the bot. The output values from this hardware were then evaluated to judge the distance from a wall and the reactions from the bot were calculated by the suggested algorithm accordingly.
Autonomous driving problems related to vehicle control using deep reinforcement learning (DRL) te... more Autonomous driving problems related to vehicle control using deep reinforcement learning (DRL) techniques, are still unsolved. DRL approaches have achieved notable results, its dependability on reward functions and defining the type of control actions are dominating factors of the objective, that controls its success. Several DRL approaches applied in the past consider a finite set of available actions to be controlled by the agent hence, it performs sharp actions. While real driving requires precision control capabilities that tend to apply safer and smoother actions. For incorporating such precision control capabilities, this paper considers the driving problem as a continuous control problem. For this, the gym-highway environments are used as these environments are controllable and customizable to simulate diverse driving scenarios. The simulation setup for parking is updated to resemble the complex scenario and for highway driving a novel reward function is designed to handle continuous actions. Dual critic based DRL approaches are applied as these approaches have shown remarkable performance in robotic locomotion control problems. The video results demonstrate the way different policies fulfil the objective.
2019 IEEE 9th International Conference on Advanced Computing (IACC), 2019
Software defined Network is a network defined by software, which is one of the important feature ... more Software defined Network is a network defined by software, which is one of the important feature which makes the legacy old networks to be flexible for dynamic configuration and so can cater to today's dynamic application requirement. It is a programmable network but it is prone to different type of attacks due to its centralized architecture. The author provided a solution to detect and prevent Distributed Denial of service attack in the paper. Mininet [5] which is a popular emulator for Software defined Network is used. We followed the approach in which collection of the traffic statistics from the various switches is done. After collection we calculated the packet rate and bandwidth which shoots up to high values when attack take place. The abrupt increase detects the attack which is then prevented by changing the forwarding logic of the host nodes to drop the packets instead of forwarding. After this, no more packets will be forwarded and then we also delete the forwarding rule in the flow table. Hence, we are finding out the change in packet rate and bandwidth to detect the attack and to prevent the attack we modify the forwarding logic of the switch flow table to drop the packets coming from malicious host instead of forwarding it.
Dataset for real-time face mask detection customised specifically for Indian ethnicity. Classes -... more Dataset for real-time face mask detection customised specifically for Indian ethnicity. Classes - 2 (Class 1 - Without Mask and Class 2 - With Mask) Total images - 1226 Total Annotations - 1226 (Annotations are present in both YOLO and Pascal VOC format ) Types of images included are: Different proportions of face covered with various type of masks like - standard PPE masks, handkerchief masks, dupattas and other type of veils etc.
We created our own dataset featuring the required military and civilian vehicle classes. The data... more We created our own dataset featuring the required military and civilian vehicle classes. The dataset contains a total of 6772 images of military trucks, military tanks, military aircraft, military helicopters, civilian cars and civilian aircraft. Out of which, 6642 are positive images and 130 are negative images. Positive images are those which contain one or more of the defined objects (i.e. Military Truck, Military Tank, Military Aircraft, Military Helicopter, Civilian Car, Civilian Aircraft). Negative images are those which contain anything else except the defined objects. All the positive images contains a total of 11528 objects. The use of negative images has a specific purpose, which is, to make the model learn about such an environment when there are no detectable objects in the image. The extension of images was converted from .jpeg and .png to .jpg, since, it is difficult to process the models with the different extensions. After the dataset collection and pre-processing, the formation of specific format files is to be carried out for dealing further with the object detection models. The Labelling is done in .txt, .csv, .xml, and tf record formats.
An Electronic Travel Aid (ETA) has become a necessity for visually impaired to provide them prope... more An Electronic Travel Aid (ETA) has become a necessity for visually impaired to provide them proper guidance and assistance in their daily routine. As the number of blind persons are gradually increasing, there is a dire need of an effective and low-cost solution for assisting them in their daily tasks. This paper presents a cane called R-Cane which is an ETA for the visually impaired and is capable of detecting obstacles in front direction using sonar sensor and alerts the user by informing whether the obstacle is within the range of one meter. In R-Cane, tensorflow object-detection API has been used for object recognition. It makes the user aware about the nature of objects by providing them voice-based output through bluetooth earphones. Raspberry Pi has been used for processing and Pi camera has been used to capture frames for object recognition. Further, we have implemented four models based on Single Shot Multibox Detector (SSD) for object detection. The experimental analysis s...
ACM Transactions on Asian and Low-Resource Language Information Processing
Scene text detection is a complicated and one of the most challenging tasks due to different envi... more Scene text detection is a complicated and one of the most challenging tasks due to different environmental restrictions, such as illuminations, lighting conditions, tiny and curved texts, and many more. Most of the works on scene text detection have overlooked the primary goal of increasing model accuracy and efficiency, resulting in heavy-weight models that require more processing resources. A novel lightweight model has been developed in this paper to improve the accuracy and efficiency of scene text detection. The proposed model relies on ResNet50 and MobileNetV2 as backbones with quantization used to make the resulting model lightweight. During quantization, the precision has been changed from float32 to float16 and int8 for making the model lightweight. In terms of inference time and Floating-Point Operations Per Second (FLOPS), the proposed method outperforms the state-of-the-art techniques by around 30-100 times. Here, well-known datasets, i.e. ICDAR2015 and ICDAR2019, have b...
ACM Transactions on Asian and Low-Resource Language Information Processing
The multi-core architecture has revolutionized the parallel computing. Despite this, the modern a... more The multi-core architecture has revolutionized the parallel computing. Despite this, the modern age compilers have a long way to achieve auto-parallelization. Through this paper, we introduce a language that encouraging the auto-parallelization. We are also introducing Front-End for our auto-parallelizing compiler. Later, we examined our compiler employing a different number of core and verify results based on different metrics based on total compilation time, memory utilization, power utilization and CPU utilization. At last, we learned that parallelizing multiple files engage more CPU resources, memory and energy, but it finishes the task at hand in less time. In this paper, we have proposed a loop code generation technique that makes the generation of nested loop IR code faster by dividing the blocks into some extra code blocks using a modular approach. Our TAM compiler technique speedup by 7.506, 5.283 and 2.509 against sequential compilation when we utilized 8, 4 and 2 cores re...
There have been tremendous improvements in deep learning and reinforcement learning techniques. A... more There have been tremendous improvements in deep learning and reinforcement learning techniques. Automating learning and intelligence to the full extent remains a challenge. The amalgamation of Reinforcement Learning and Deep Learning has brought breakthroughs in games and robotics in the past decade. Deep Reinforcement Learning (DRL) involves training the agent with raw input and learning via interaction with the environment. Motivated by recent successes of DRL, we have explored its adaptability to different domains and application areas. This paper also presents a comprehensive survey of the work done in recent years and simulation tools used for DRL. The current focus of researchers is on recording the experience in a better way, and refining the policy for futuristic moves. It is found that even after obtaining good results in Atari, Go, Robotics, multi-agent scenarios, there are challenges such as generalization, satisfying multiple objectives, divergence, learning robust policy. Furthermore, the complex environment and multiple agents are throwing new challenges, which is an open area of research.
2019 Sixth International Conference on Internet of Things: Systems, Management and Security (IOTSMS)
Detecting a specific person from the crowd using drone along with some resource constraint device... more Detecting a specific person from the crowd using drone along with some resource constraint device is a major concern which we are discussing in the paper. Combining the advanced algorithms and some smart hardware material, we will be finding a way to search for a missing individual in a crowd or at some location. We can also search for a person at a specific location by setting our aerial vehicle to fly autonomously and search for the required person. This will help us to cover areas which cannot be reached by humans easily. The flying robot helps to solve real-time problems and come up with some new and more advanced ways to search for the missing ones with more ease, as advanced technological methods are applied, the probability of getting accurate results increases axiomatically. The drone can fly fully autonomously and search or capture videos/photos of the required location. Location commands could be given using PC, mobile and with the help of IoT, using Raspberry Pi.
2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence)
Software defined networking is going to be an essential part of networking domain which moves the... more Software defined networking is going to be an essential part of networking domain which moves the traditional networking domain to automation network. Data security is going to be an important factor in this new networking architecture. Paper aim to classify the traffic into normal and malicious classes based on features given in dataset by using various deep learning techniques. The classification of traffic into one of the classes after pre-processing of the dataset is done. We got accuracy score of 99.75% by applying Stacked Auto-Encoder Multi-layer Perceptron (SAE-MLP) which is explained in the paper. Thus, the purpose of network traffic classification using deep learning techniques was fulfilled.
Emerging Technologies in Computer Engineering: Microservices in Big Data Analytics
Automatic parallelization is necessary for all system. Every person wants the program to execute ... more Automatic parallelization is necessary for all system. Every person wants the program to execute as soon as possible. Now Days, programmer want to get run faster the sequential program. Automatic parallelization is the greatest challenge in now days. Parallelization implies converting the sequential code to parallel code to getting better utilization of multi-core processor. In parallelization, multi-core use the memory in sharing mode or massage passing. Now day’s programmers don’t want to take extra overheads of parallelization because they want it from the compiler that’s called automatic parallelization. Its main reason is to free the programmers from manual parallelization process. The conversion of a program into parallelize form is very complex work due to program analysis and an unknown value of the variable during compile time. The main reason of conversion is execution time of program due to loops, so the most challenging task is to parallelize the loops and run it on multi-core by breaking the loop iterations. In parallelization process, the compiler must have to check the dependent between loop statements that they are independent of each other. If they are dependent or effect the other statement by running the statement in parallel, so it does not convert it. After checking the dependency, test converts it into parallelization by using OpenMP API. We add some line of OpenMP for enabling parallelization in the loop.
2017 13th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS)
In this article we propose a novel biometric identification method using Footprint. A paper scann... more In this article we propose a novel biometric identification method using Footprint. A paper scanner was used to obtain images to uniquely identify the person. 312 footprint images from 78 persons(2 samples each foot) were analyzed, leading to the conclusion that footprints could also be used to identify human. Physiological and behavioral biometric characteristics make it a great alternative to computational intensive algorithms like fingerprint, palm print, retina or iris scan, and face. Foot biometric is also a great alternative. In spite of having minutia features (considered totally unique and already tested in fingerprint) it also has geometric features like hand geometry which give satisfactory results in recognition.We have computed province, major axis, minor axis, eccentricity in one approach, where a foot is divided into 15 equal sized boxes in another shape-based algorithm. This article also examines the texture features of the foot. It could be applied at those places where people inherently remove their shoes, such as holy places(temples and mosque). They remove shoes at famous monuments such as The Taj Mahal, India from the perspective of hygiene and preservation. Usually, these places are with a strong foot fall and high-risk security due to the chaotic crowd. It could also be employed in newborn authentication and identification. Uniqueness of minutiae foot-print in newborns has been already proved.
IEEE Transactions on Neural Networks and Learning Systems, 2021
Recently introduced deep reinforcement learning (DRL) techniques in discrete-time have resulted i... more Recently introduced deep reinforcement learning (DRL) techniques in discrete-time have resulted in significant advances in online games, robotics, and so on. Inspired from recent developments, we have proposed an approach referred to as Quantile Critic with Spiking Actor and Normalized Ensemble (QC_SANE) for continuous control problems, which uses quantile loss to train critic and a spiking neural network (NN) to train an ensemble of actors. The NN does an internal normalization using a scaled exponential linear unit (SELU) activation function and ensures robustness. The empirical study on multijoint dynamics with contact (MuJoCo)-based environments shows improved training and test results than the state-of-the-art approach: population coded spiking actor network (PopSAN).
The challenges involved in the traditional cloud computing paradigms have prompted the developmen... more The challenges involved in the traditional cloud computing paradigms have prompted the development of architectures for the next generation cloud computing. The new cloud computing architectures can generate and handle huge amount of data, which was not possible to handle with the help of traditional architectures. Deep learning algorithms have the ability to process this huge amount of data and, thus, can now solve the problem of the next generation computing algorithms. Therefore, these days, deep learning has become the state-of-the-art approach for solving various tasks and most importantly in the field of recognition. In this work, recognition of city names is proposed. Recognition of handwritten city names is one of the potential research application areas in the field of postal automation For recognition using a segmentation-free approach (Holistic approach). This proposed work demystifies the role of convolutional neural network (CNN), which is one of the methods of deep lea...
Wireless Sensor Network comprises of hundreds or thousands of sensor nodes to sense the environme... more Wireless Sensor Network comprises of hundreds or thousands of sensor nodes to sense the environment collect data and send these data to a central location known as Base Station or Sink. Sensor Networks are very useful in many application such as military applications, environmental condition detection, whether prediction, measuring temperature, sound, wave, humidity and so on. Each sensor node in Sensor Networks operates on battery. In many scenarios, it is very difficult to recharge or replace battery of sensor nodes. So routing protocols for Wireless Sensor Networks should be as energy efficient as possible. Security is also very important in Wireless Sensor Networks. There may be possible to attack by intruders on routing protocols in Wireless Sensor Networks, and degrade the performance of networks. LEACH (Low Energy Adaptive Clustering Hierarchy) is one of the very energy efficient routing protocols, which uses randomized cluster rotation to distribute energy load among all sen...
The main purpose of this paper is to develop an efficient machine learning model to estimate the ... more The main purpose of this paper is to develop an efficient machine learning model to estimate the electric power load. The developed machine learning model can be used by electric power utilities for proper operation and maintenance of grid and also to trade electricity effectively in energy market. This paper proposes a machine learning model using gated recurrent unit (GRU) and random forest (RF). GRU has been employed to predict the electric power load, whereas RF has been used to reduce the input dimensions of the model. GRU has been estimating the load with good accuracy. RF reduces the input dimensions of the GRU that leads lightweight GRU model. The main benefits of the lightweight GRU models are less computation time and memory space. However, lightweight GRU models will loss small amount of accuracy comparing to the original GRU model. GRU along with RF has been used for the first for short load forecasting. All the machine learning model’s performance has been observed in stochastic environment. Impact of weekends on load forecasting also observed by considering the last 3-week load data.
2021 International Conference on Computer Communication and Informatics (ICCCI), 2021
In a future that is not too distant from today, humanoids are surely going to be an integral part... more In a future that is not too distant from today, humanoids are surely going to be an integral part of both our professional and private lives, assisting us with various tasks. Unlike normal robots that we may encounter in our everyday lives, humanoids are designed in specific manners to give them more human-like capabilities that enable them to perform complex tasks such as climbing a flight of stairs. In this paper, we present a Maze-Solving Algorithm which is a software developed specifically for the humanoid robot, NAO, and gives it the capability to enter and exit a maze autonomously. NAO is a next-gen humanoid bot developed by SoftBank Robotics using the power of AI. The bot is equipped with numerous sensors and cameras. Though various quantitative approaches were considered and experimented with, we stuck onto the one which had the least average time complexity of all after a thorough comparative study. We suggest an approach where the humanoid can detect and localize objects from a distance and take programmable decisions based on them. AI constantly tries to give robots human-thinking capabilities to make their decision-making skills similar to those of humans, if not better than them. This algorithm was developed taking into consideration how a human intellect would react rationally if he is stuck in a maze. The methodology used revolves primarily around the combined use of SONAR(Sound navigation ranging) and tactical sensors, and cameras equipped within the bot. The output values from this hardware were then evaluated to judge the distance from a wall and the reactions from the bot were calculated by the suggested algorithm accordingly.
Autonomous driving problems related to vehicle control using deep reinforcement learning (DRL) te... more Autonomous driving problems related to vehicle control using deep reinforcement learning (DRL) techniques, are still unsolved. DRL approaches have achieved notable results, its dependability on reward functions and defining the type of control actions are dominating factors of the objective, that controls its success. Several DRL approaches applied in the past consider a finite set of available actions to be controlled by the agent hence, it performs sharp actions. While real driving requires precision control capabilities that tend to apply safer and smoother actions. For incorporating such precision control capabilities, this paper considers the driving problem as a continuous control problem. For this, the gym-highway environments are used as these environments are controllable and customizable to simulate diverse driving scenarios. The simulation setup for parking is updated to resemble the complex scenario and for highway driving a novel reward function is designed to handle continuous actions. Dual critic based DRL approaches are applied as these approaches have shown remarkable performance in robotic locomotion control problems. The video results demonstrate the way different policies fulfil the objective.
2019 IEEE 9th International Conference on Advanced Computing (IACC), 2019
Software defined Network is a network defined by software, which is one of the important feature ... more Software defined Network is a network defined by software, which is one of the important feature which makes the legacy old networks to be flexible for dynamic configuration and so can cater to today's dynamic application requirement. It is a programmable network but it is prone to different type of attacks due to its centralized architecture. The author provided a solution to detect and prevent Distributed Denial of service attack in the paper. Mininet [5] which is a popular emulator for Software defined Network is used. We followed the approach in which collection of the traffic statistics from the various switches is done. After collection we calculated the packet rate and bandwidth which shoots up to high values when attack take place. The abrupt increase detects the attack which is then prevented by changing the forwarding logic of the host nodes to drop the packets instead of forwarding. After this, no more packets will be forwarded and then we also delete the forwarding rule in the flow table. Hence, we are finding out the change in packet rate and bandwidth to detect the attack and to prevent the attack we modify the forwarding logic of the switch flow table to drop the packets coming from malicious host instead of forwarding it.
Dataset for real-time face mask detection customised specifically for Indian ethnicity. Classes -... more Dataset for real-time face mask detection customised specifically for Indian ethnicity. Classes - 2 (Class 1 - Without Mask and Class 2 - With Mask) Total images - 1226 Total Annotations - 1226 (Annotations are present in both YOLO and Pascal VOC format ) Types of images included are: Different proportions of face covered with various type of masks like - standard PPE masks, handkerchief masks, dupattas and other type of veils etc.
We created our own dataset featuring the required military and civilian vehicle classes. The data... more We created our own dataset featuring the required military and civilian vehicle classes. The dataset contains a total of 6772 images of military trucks, military tanks, military aircraft, military helicopters, civilian cars and civilian aircraft. Out of which, 6642 are positive images and 130 are negative images. Positive images are those which contain one or more of the defined objects (i.e. Military Truck, Military Tank, Military Aircraft, Military Helicopter, Civilian Car, Civilian Aircraft). Negative images are those which contain anything else except the defined objects. All the positive images contains a total of 11528 objects. The use of negative images has a specific purpose, which is, to make the model learn about such an environment when there are no detectable objects in the image. The extension of images was converted from .jpeg and .png to .jpg, since, it is difficult to process the models with the different extensions. After the dataset collection and pre-processing, the formation of specific format files is to be carried out for dealing further with the object detection models. The Labelling is done in .txt, .csv, .xml, and tf record formats.
An Electronic Travel Aid (ETA) has become a necessity for visually impaired to provide them prope... more An Electronic Travel Aid (ETA) has become a necessity for visually impaired to provide them proper guidance and assistance in their daily routine. As the number of blind persons are gradually increasing, there is a dire need of an effective and low-cost solution for assisting them in their daily tasks. This paper presents a cane called R-Cane which is an ETA for the visually impaired and is capable of detecting obstacles in front direction using sonar sensor and alerts the user by informing whether the obstacle is within the range of one meter. In R-Cane, tensorflow object-detection API has been used for object recognition. It makes the user aware about the nature of objects by providing them voice-based output through bluetooth earphones. Raspberry Pi has been used for processing and Pi camera has been used to capture frames for object recognition. Further, we have implemented four models based on Single Shot Multibox Detector (SSD) for object detection. The experimental analysis s...
ACM Transactions on Asian and Low-Resource Language Information Processing
Scene text detection is a complicated and one of the most challenging tasks due to different envi... more Scene text detection is a complicated and one of the most challenging tasks due to different environmental restrictions, such as illuminations, lighting conditions, tiny and curved texts, and many more. Most of the works on scene text detection have overlooked the primary goal of increasing model accuracy and efficiency, resulting in heavy-weight models that require more processing resources. A novel lightweight model has been developed in this paper to improve the accuracy and efficiency of scene text detection. The proposed model relies on ResNet50 and MobileNetV2 as backbones with quantization used to make the resulting model lightweight. During quantization, the precision has been changed from float32 to float16 and int8 for making the model lightweight. In terms of inference time and Floating-Point Operations Per Second (FLOPS), the proposed method outperforms the state-of-the-art techniques by around 30-100 times. Here, well-known datasets, i.e. ICDAR2015 and ICDAR2019, have b...
ACM Transactions on Asian and Low-Resource Language Information Processing
The multi-core architecture has revolutionized the parallel computing. Despite this, the modern a... more The multi-core architecture has revolutionized the parallel computing. Despite this, the modern age compilers have a long way to achieve auto-parallelization. Through this paper, we introduce a language that encouraging the auto-parallelization. We are also introducing Front-End for our auto-parallelizing compiler. Later, we examined our compiler employing a different number of core and verify results based on different metrics based on total compilation time, memory utilization, power utilization and CPU utilization. At last, we learned that parallelizing multiple files engage more CPU resources, memory and energy, but it finishes the task at hand in less time. In this paper, we have proposed a loop code generation technique that makes the generation of nested loop IR code faster by dividing the blocks into some extra code blocks using a modular approach. Our TAM compiler technique speedup by 7.506, 5.283 and 2.509 against sequential compilation when we utilized 8, 4 and 2 cores re...
There have been tremendous improvements in deep learning and reinforcement learning techniques. A... more There have been tremendous improvements in deep learning and reinforcement learning techniques. Automating learning and intelligence to the full extent remains a challenge. The amalgamation of Reinforcement Learning and Deep Learning has brought breakthroughs in games and robotics in the past decade. Deep Reinforcement Learning (DRL) involves training the agent with raw input and learning via interaction with the environment. Motivated by recent successes of DRL, we have explored its adaptability to different domains and application areas. This paper also presents a comprehensive survey of the work done in recent years and simulation tools used for DRL. The current focus of researchers is on recording the experience in a better way, and refining the policy for futuristic moves. It is found that even after obtaining good results in Atari, Go, Robotics, multi-agent scenarios, there are challenges such as generalization, satisfying multiple objectives, divergence, learning robust policy. Furthermore, the complex environment and multiple agents are throwing new challenges, which is an open area of research.
2019 Sixth International Conference on Internet of Things: Systems, Management and Security (IOTSMS)
Detecting a specific person from the crowd using drone along with some resource constraint device... more Detecting a specific person from the crowd using drone along with some resource constraint device is a major concern which we are discussing in the paper. Combining the advanced algorithms and some smart hardware material, we will be finding a way to search for a missing individual in a crowd or at some location. We can also search for a person at a specific location by setting our aerial vehicle to fly autonomously and search for the required person. This will help us to cover areas which cannot be reached by humans easily. The flying robot helps to solve real-time problems and come up with some new and more advanced ways to search for the missing ones with more ease, as advanced technological methods are applied, the probability of getting accurate results increases axiomatically. The drone can fly fully autonomously and search or capture videos/photos of the required location. Location commands could be given using PC, mobile and with the help of IoT, using Raspberry Pi.
2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence)
Software defined networking is going to be an essential part of networking domain which moves the... more Software defined networking is going to be an essential part of networking domain which moves the traditional networking domain to automation network. Data security is going to be an important factor in this new networking architecture. Paper aim to classify the traffic into normal and malicious classes based on features given in dataset by using various deep learning techniques. The classification of traffic into one of the classes after pre-processing of the dataset is done. We got accuracy score of 99.75% by applying Stacked Auto-Encoder Multi-layer Perceptron (SAE-MLP) which is explained in the paper. Thus, the purpose of network traffic classification using deep learning techniques was fulfilled.
Emerging Technologies in Computer Engineering: Microservices in Big Data Analytics
Automatic parallelization is necessary for all system. Every person wants the program to execute ... more Automatic parallelization is necessary for all system. Every person wants the program to execute as soon as possible. Now Days, programmer want to get run faster the sequential program. Automatic parallelization is the greatest challenge in now days. Parallelization implies converting the sequential code to parallel code to getting better utilization of multi-core processor. In parallelization, multi-core use the memory in sharing mode or massage passing. Now day’s programmers don’t want to take extra overheads of parallelization because they want it from the compiler that’s called automatic parallelization. Its main reason is to free the programmers from manual parallelization process. The conversion of a program into parallelize form is very complex work due to program analysis and an unknown value of the variable during compile time. The main reason of conversion is execution time of program due to loops, so the most challenging task is to parallelize the loops and run it on multi-core by breaking the loop iterations. In parallelization process, the compiler must have to check the dependent between loop statements that they are independent of each other. If they are dependent or effect the other statement by running the statement in parallel, so it does not convert it. After checking the dependency, test converts it into parallelization by using OpenMP API. We add some line of OpenMP for enabling parallelization in the loop.
2017 13th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS)
In this article we propose a novel biometric identification method using Footprint. A paper scann... more In this article we propose a novel biometric identification method using Footprint. A paper scanner was used to obtain images to uniquely identify the person. 312 footprint images from 78 persons(2 samples each foot) were analyzed, leading to the conclusion that footprints could also be used to identify human. Physiological and behavioral biometric characteristics make it a great alternative to computational intensive algorithms like fingerprint, palm print, retina or iris scan, and face. Foot biometric is also a great alternative. In spite of having minutia features (considered totally unique and already tested in fingerprint) it also has geometric features like hand geometry which give satisfactory results in recognition.We have computed province, major axis, minor axis, eccentricity in one approach, where a foot is divided into 15 equal sized boxes in another shape-based algorithm. This article also examines the texture features of the foot. It could be applied at those places where people inherently remove their shoes, such as holy places(temples and mosque). They remove shoes at famous monuments such as The Taj Mahal, India from the perspective of hygiene and preservation. Usually, these places are with a strong foot fall and high-risk security due to the chaotic crowd. It could also be employed in newborn authentication and identification. Uniqueness of minutiae foot-print in newborns has been already proved.
Uploads
Papers by Gaurav Singal