Abhishek Verma - Academia.edu (original) (raw)
Papers by Abhishek Verma
Convolutional Neural Networks nowadays are of tremendous importance for any image classification ... more Convolutional Neural Networks nowadays are of tremendous importance for any image classification system. One of the most investigated methods to increase the accuracy of CNN is by increasing the depth of CNN. Increasing the depth by stacking more layers also increases the difficulty of training besides making it computationally expensive. Some research found that adding auxiliary forks after intermediate layers increases the accuracy. Specifying which intermediate layer should have the fork just addressed recently. Where a simple rule were used to detect the position of intermediate layers that needs the auxiliary supervision fork. This technique known as convolutional neural networks with deep supervision (CNDS). This technique enhanced the accuracy of classification over the straight forward CNN used on the MIT places dataset and ImageNet. In the other side, Residual Learning is another technique emerged recently to ease the training of very deep CNN. Residual learning framework changed the learning of layers from unreferenced functions to learning residual function with regard to the layer's input. Residual Learning achieved state of arts results on ImageNet 2015 and COCO competitions. In this paper, we study the effect of adding residual Connections to CNDS network. Our experiments results show increasing of accuracy over using CNDS only.
International Journal of Computational Vision and Robotics, 2020
Increasing depth of convolutional neural networks (CNNs) is a highly promising method of increasi... more Increasing depth of convolutional neural networks (CNNs) is a highly promising method of increasing the accuracy of the (CNNs). Increased CNN depth will also result in increased layer count (parameters), leading to a slow backpropagation convergence prone to overfitting. We trained our model (residual-CNDS) to classify very large-scale scene datasets. The outcome result from the two datasets proved our proposed model effectively handled the slow convergence, overfitting, and degradation. Our approach overcomes degradation in the very deep network. We have built two models (residual-CNDS 8), and (residual-CNDS 10). Moreover, we tested our models on two large-scale datasets, and we compared our results with other recently introduced cutting-edge networks in the domain of top-1 and top-5 classification accuracy. As a result, both of models have shown good improvement, which supports the assertion that the addition of residual connections enhances network CNDS accuracy without adding any computation complexity.
International Journal of Computational Vision and Robotics, 2018
Planktons are extremely diverse groups of organisms that exist in large water columns. They are s... more Planktons are extremely diverse groups of organisms that exist in large water columns. They are sources of food for fishes and many other marine life animals. The plankton distribution is essential for the survival of many ocean lives and plays a critical role in marine ecosystem. In recent years, intelligent image classification systems were developed to study plankton distribution through classification of the plankton images taken by underwater imaging devices. Due to the significant differences in both shapes and sizes of the plankton population, accurate classification poses a daunting challenge. The mixed quality of the collected images adds more difficulty to the task. In this paper, we present an intelligent machine learning system built on convolutional neural networks (CNN) for plankton image classification. Unlike most of the existing image classification algorithms, CNN based systems do not depend on features engineering and they can be efficiently extended to encompass new classes. The experimental results on SIPPER image datasets show that the proposed system achieves higher accuracy compared with the state-of-the-art approaches. The new system is also capable of learning a much larger number of plankton classes.
Medical Journal of Australia, 2009
Warfarin and heparin are the traditional mainstay anticoagulant therapies for treating thromboemb... more Warfarin and heparin are the traditional mainstay anticoagulant therapies for treating thromboembolic disease. These drugs, with a documented history of utility, also have inherent difficulties in usage; in particular, the complicated monitoring and numerous drug-drug interactions of warfarin, and the need for parenteral administration of heparins. New agents have recently emerged that target specific elements of the clotting pathway. Rivaroxaban, which inhibits activated factor X (Xa), is currently in clinical trials and is the most advanced factor Xa inhibitor. The drug offers once-daily oral dosing, with no need for injections, dose titration, or frequent blood tests to monitor the international normalised ratio. It has a rapid onset of action and, although there is no specific antidote, it has a short plasma elimination half-life (about 5-9 hours). Evidence from recently published large-scale phase III clinical trials shows rivaroxaban to be superior to enoxaparin for prophylaxis of venous thromboembolism after major orthopaedic surgery. Studies have shown rivaroxaban to have a sound safety profile, with an incidence of bleeding similar to enoxaparin in phase III clinical trials. Few side effects and drug-drug interactions between rivaroxaban and common medications have been found thus far, although some interactions with potent cytochrome P450 3A4 inhibitors have been observed. It is hoped that rivaroxaban may be used as a first-line anticoagulant for prophylaxis of venous thromboembolic disease in postsurgical patients.
2015 54th IEEE Conference on Decision and Control (CDC), 2015
This paper investigates the application of human motion guidance principles to the development of... more This paper investigates the application of human motion guidance principles to the development of an autonomous guidance process. Primarily, it extends previous work, which identifies the organization of human guidance performance given by trajectory data. First, the paper describes interaction patterns (IP), subgoals, and partitions and defines these elements in terms of control theoretical concepts. Second, IPs are applied to form a computational solution to a constrained optimal control problem and simulation results demonstrate that this approach produces feasible solutions. Finally, first-person human experimental results are provided to validate theoretical concepts.
Scaling effects-i.e., the relationship between the aircrafts maneuvering capabilities and the env... more Scaling effects-i.e., the relationship between the aircrafts maneuvering capabilities and the environment scale characteristics-are expected to play an important role in how the performance of a guidance system plays out. These effects are most significant for unmanned aerial vehicles (UAVs) since these vehicles can be built in a much broader range of scales than manned vehicles. This paper investigates how scaling effects should be accounted for in the evaluation of guidance performance. It proposes a non-dimensional quantity call the agility scale ratio (ASR). This quantity is similar to the Reynolds number or the Fraude number and captures the relationship between vehicle scale and maneuverability and the scale characteristics of the environment. Investigating guidance performance from a non-dimensional perspective can also provide important insights into how agile vehicles interact with complex envi-ronments. The paper presents results from examples motivated by recent work in g...
Aerospace Engineering and Mechanics, University of Minnesota Minneapolis, Minnesota, USA This pap... more Aerospace Engineering and Mechanics, University of Minnesota Minneapolis, Minnesota, USA This paper describes a guidance algorithm for autonomous operation in partially known environments. The emphasis of the paper is enabling learning within a receding horizon trajectory optimization framework. The information acquired from an exteroceptive sensor is assimilated into a spatial value function. This setup has the advantage that the system learns information directly relevant to optimal guidance and control behavior and enables efficient trajectory-planning in unknown or partially known environments. The system's performance is demonstrated using successive runs in high-fidelity indoor simulations.
Convolutional Neural Networks nowadays are of tremendous importance for any image classification ... more Convolutional Neural Networks nowadays are of tremendous importance for any image classification system. One of the most investigated methods to increase the accuracy of CNN is by increasing the depth of CNN. Increasing the depth by stacking more layers also increases the difficulty of training besides making it computationally expensive. Some research found that adding auxiliary forks after intermediate layers increases the accuracy. Specifying which intermediate layer should have the fork just addressed recently. Where a simple rule were used to detect the position of intermediate layers that needs the auxiliary supervision fork. This technique known as convolutional neural networks with deep supervision (CNDS). This technique enhanced the accuracy of classification over the straight forward CNN used on the MIT places dataset and ImageNet. In the other side, Residual Learning is another technique emerged recently to ease the training of very deep CNN. Residual learning framework changed the learning of layers from unreferenced functions to learning residual function with regard to the layer's input. Residual Learning achieved state of arts results on ImageNet 2015 and COCO competitions. In this paper, we study the effect of adding residual Connections to CNDS network. Our experiments results show increasing of accuracy over using CNDS only.
International Journal of Computational Vision and Robotics, 2020
Increasing depth of convolutional neural networks (CNNs) is a highly promising method of increasi... more Increasing depth of convolutional neural networks (CNNs) is a highly promising method of increasing the accuracy of the (CNNs). Increased CNN depth will also result in increased layer count (parameters), leading to a slow backpropagation convergence prone to overfitting. We trained our model (residual-CNDS) to classify very large-scale scene datasets. The outcome result from the two datasets proved our proposed model effectively handled the slow convergence, overfitting, and degradation. Our approach overcomes degradation in the very deep network. We have built two models (residual-CNDS 8), and (residual-CNDS 10). Moreover, we tested our models on two large-scale datasets, and we compared our results with other recently introduced cutting-edge networks in the domain of top-1 and top-5 classification accuracy. As a result, both of models have shown good improvement, which supports the assertion that the addition of residual connections enhances network CNDS accuracy without adding any computation complexity.
International Journal of Computational Vision and Robotics, 2018
Planktons are extremely diverse groups of organisms that exist in large water columns. They are s... more Planktons are extremely diverse groups of organisms that exist in large water columns. They are sources of food for fishes and many other marine life animals. The plankton distribution is essential for the survival of many ocean lives and plays a critical role in marine ecosystem. In recent years, intelligent image classification systems were developed to study plankton distribution through classification of the plankton images taken by underwater imaging devices. Due to the significant differences in both shapes and sizes of the plankton population, accurate classification poses a daunting challenge. The mixed quality of the collected images adds more difficulty to the task. In this paper, we present an intelligent machine learning system built on convolutional neural networks (CNN) for plankton image classification. Unlike most of the existing image classification algorithms, CNN based systems do not depend on features engineering and they can be efficiently extended to encompass new classes. The experimental results on SIPPER image datasets show that the proposed system achieves higher accuracy compared with the state-of-the-art approaches. The new system is also capable of learning a much larger number of plankton classes.
Medical Journal of Australia, 2009
Warfarin and heparin are the traditional mainstay anticoagulant therapies for treating thromboemb... more Warfarin and heparin are the traditional mainstay anticoagulant therapies for treating thromboembolic disease. These drugs, with a documented history of utility, also have inherent difficulties in usage; in particular, the complicated monitoring and numerous drug-drug interactions of warfarin, and the need for parenteral administration of heparins. New agents have recently emerged that target specific elements of the clotting pathway. Rivaroxaban, which inhibits activated factor X (Xa), is currently in clinical trials and is the most advanced factor Xa inhibitor. The drug offers once-daily oral dosing, with no need for injections, dose titration, or frequent blood tests to monitor the international normalised ratio. It has a rapid onset of action and, although there is no specific antidote, it has a short plasma elimination half-life (about 5-9 hours). Evidence from recently published large-scale phase III clinical trials shows rivaroxaban to be superior to enoxaparin for prophylaxis of venous thromboembolism after major orthopaedic surgery. Studies have shown rivaroxaban to have a sound safety profile, with an incidence of bleeding similar to enoxaparin in phase III clinical trials. Few side effects and drug-drug interactions between rivaroxaban and common medications have been found thus far, although some interactions with potent cytochrome P450 3A4 inhibitors have been observed. It is hoped that rivaroxaban may be used as a first-line anticoagulant for prophylaxis of venous thromboembolic disease in postsurgical patients.
2015 54th IEEE Conference on Decision and Control (CDC), 2015
This paper investigates the application of human motion guidance principles to the development of... more This paper investigates the application of human motion guidance principles to the development of an autonomous guidance process. Primarily, it extends previous work, which identifies the organization of human guidance performance given by trajectory data. First, the paper describes interaction patterns (IP), subgoals, and partitions and defines these elements in terms of control theoretical concepts. Second, IPs are applied to form a computational solution to a constrained optimal control problem and simulation results demonstrate that this approach produces feasible solutions. Finally, first-person human experimental results are provided to validate theoretical concepts.
Scaling effects-i.e., the relationship between the aircrafts maneuvering capabilities and the env... more Scaling effects-i.e., the relationship between the aircrafts maneuvering capabilities and the environment scale characteristics-are expected to play an important role in how the performance of a guidance system plays out. These effects are most significant for unmanned aerial vehicles (UAVs) since these vehicles can be built in a much broader range of scales than manned vehicles. This paper investigates how scaling effects should be accounted for in the evaluation of guidance performance. It proposes a non-dimensional quantity call the agility scale ratio (ASR). This quantity is similar to the Reynolds number or the Fraude number and captures the relationship between vehicle scale and maneuverability and the scale characteristics of the environment. Investigating guidance performance from a non-dimensional perspective can also provide important insights into how agile vehicles interact with complex envi-ronments. The paper presents results from examples motivated by recent work in g...
Aerospace Engineering and Mechanics, University of Minnesota Minneapolis, Minnesota, USA This pap... more Aerospace Engineering and Mechanics, University of Minnesota Minneapolis, Minnesota, USA This paper describes a guidance algorithm for autonomous operation in partially known environments. The emphasis of the paper is enabling learning within a receding horizon trajectory optimization framework. The information acquired from an exteroceptive sensor is assimilated into a spatial value function. This setup has the advantage that the system learns information directly relevant to optimal guidance and control behavior and enables efficient trajectory-planning in unknown or partially known environments. The system's performance is demonstrated using successive runs in high-fidelity indoor simulations.