Papers by Prasad N R
https://www.techrxiv.org/users/575968/articles/619013-probabilistic-representation-additive-shift... more https://www.techrxiv.org/users/575968/articles/619013-probabilistic-representation-additive-shifted-automatic-differentiation
Multiplication of weights may or may not be a practical biological solution (the hypotheses is that biological neurons do not have symmetric feedback and exact 32-bit multiplication). As an approximation of universal approximation theorem, bit-shift (which is a non-differentiable symbol) is proposed. There are two variants of this problem (both of which can be solved using backpropagation): First variant is addition of bit-shift with non-negative float (without subtraction); The second variant is addition and subtraction of integers (without float). So, combining these two, we get additive-shift of float. The accuracy of this version is only about 2% lesser than the original DNN. The idea is to try to avoid clock-cycles and make it easier to do inference almost at the speed of electricity in semiconductor (without memory or with relatively less memory). Another idea is to improve the ML-training and inference process so that they are compute-constrained.
End-to-end explainable Bayes deep-learning research works when there are products of many inputs.... more End-to-end explainable Bayes deep-learning research works when there are products of many inputs. But, multiplication is computationally timeconsuming (and there can be quantisation errors if there are many multiplications on a 32-bit computer). So, a pure additive deep-learningnetwork is being used using logarithms (which can be used as hash-maps or so). A training process and a local approximation theorem are being proposed. The training process includes backpropagation too. Note: The results have not been discussed. I am planning to discuss this using the peer-reviewed version of this.
Explainable AI is an important aspect. But, to the best of knowledge, there isn't any end-to-end ... more Explainable AI is an important aspect. But, to the best of knowledge, there isn't any end-to-end probabilistic explainable algorithms that is easy to work with. So, using Laplace smoothing idea of Bayes probabilities, an end-to-end probabilistic deep-learning algorithm is proposed. Also, there are three initial ideas related to the machine learning training process of this deep-learning.
Blind Descent uses constrained but, guided approach to "learn" the weights; The probability densi... more Blind Descent uses constrained but, guided approach to "learn" the weights; The probability density function is non-zero in the infinite space of the dimension (case in point: Gaussians and normal probability distribution functions). In Blind Descent paper, some of the implicit ideas involving layer by layer training and filter by filter training (with different batch sizes) were proposed as probable greedy solutions. The results of similar experiments are discussed. Octave (and proposed PyTorch vari-ants') source code of the experiments of this paper can be found at https: //github.com/PrasadNR/Attempted-Blind-Constrained-Descent-Experiments-ABCDE-. This is compared against the ABCDE derivatives of the original PyTorch source code of https://github.com/akshat57/Blind-Descent.
We describe an alternative to gradient descent for backpropogation through a neural network, whic... more We describe an alternative to gradient descent for backpropogation through a neural network, which we call Blind Descent. We believe that Blind Descent can be used to augment backpropagation by using it as an initialisation method and can also be used at saturation. Blind Descent, inherently by design, does not face problems like exploding or vanishing gradients.
Springer, 2016
Skin detection is an important part of face localization since the most exposed part of human ski... more Skin detection is an important part of face localization since the most exposed part of human skin is the face. This paper proposes a novel algorithm for face localization via skin detection. The algorithm utilizes a kernel iterative procedure to check for the region of interest in an image where it is likely that the face exists. The algorithm utilizes a swarm of particles in a kernel that randomly check the fitness of the corresponding pixel BGR value which is determined from a skin BGR dataset. Consequently, the kernel which has the best fitness value is chosen as the region of the image where it is likely that the face exists. Following this, we employ an active contour model called Snakes algorithm which further converges on our region of interest and finally a contour of the face region is extracted from the original image.
Elsevier, 2016
In a hybrid road network with multiple paths to same location having prior geographical knowledge... more In a hybrid road network with multiple paths to same location having prior geographical knowledge, successful navigation for mobile robots is one of the main challenges. Path planning is one of the most important issues in the navigation process which enables the selection and identification of a suitable path for the robot to traverse in the workspace area. Path-planning for mapped roads can be considered as the process of navigating a mobile robot around a configured road map, which provides optimized path by considering roughness of roads. In this paper, we propose a novel navigation algorithm for outdoor environments, which permits robots to travel from one static node to another along a planned path. It utilizes Normal probability weight distribution (NPWD) to assign weights between two nodes dynamically. Heuristics based shortest path (HSP) algorithm is employed to solve complex optimization problems concerned with real-world scenarios. The experiments performed on categorized road databases show significant improvement in timings and complexity of system. Our results justify the effectiveness for the implementation of driver-assist system.
Springer, Singapore, 2019
TraQuad is an autonomous tracking quadcopter capable of tracking any moving (or static) object li... more TraQuad is an autonomous tracking quadcopter capable of tracking any moving (or static) object like cars, humans, other drones or any other object on-the-go. This article describes the applications and advantages of TraQuad and the reduction in cost (to about 250$) that has been achieved so far using the hardware and software capabilities and our custom algorithms wherever needed. This description is backed by strong data and the research analyses which have been drawn out of extant information or conducted on own when necessary. This also describes the development of completely autonomous (even GPS is optional) low-cost drone which can act as a major platform for further developments in automation, transportation, reconnaissance and more. We describe our ROS Gazebo simulator and our STATUS algorithms which form the core of our development of our object tracking drone for generic purposes.
Conference Presentations by Prasad N R
Elsevier, 2016
In a hybrid road network with multiple paths to same location having prior geographical knowledge... more In a hybrid road network with multiple paths to same location having prior geographical knowledge, successful navigation for mobile robots is one of the main challenges. Path planning is one of the most important issues in the navigation process which enables the selection and identification of a suitable path for the robot to traverse in the workspace area. Path-planning for mapped roads can be considered as the process of navigating a mobile robot around a configured road map, which provides optimized path by considering roughness of roads. In this paper, we propose a novel navigation algorithm for outdoor environments, which permits robots to travel from one static node to another along a planned path. It utilizes Normal probability weight distribution (NPWD) to assign weights between two nodes dynamically. Heuristics based shortest path (HSP) algorithm is employed to solve complex optimization problems concerned with real-world scenarios. The experiments performed on categorized road databases show significant improvement in timings and complexity of system. Our results justify the effectiveness for the implementation of driver-assist system.
Springer, Singapore, 2019
TraQuad is an autonomous tracking quadcopter capable of tracking any moving (or static) object li... more TraQuad is an autonomous tracking quadcopter capable of tracking any moving (or static) object like cars, humans, other drones or any other object on-the-go. This article describes the applications and advantages of TraQuad and the reduction in cost (to about 250$) that has been achieved so far using the hardware and software capabilities and our custom algorithms wherever needed. This description is backed by strong data and the research analyses which have been drawn out of extant information or conducted on own when necessary. This also describes the development of completely autonomous (even GPS is optional) low-cost drone which can act as a major platform for further developments in automation, transportation, reconnaissance and more. We describe our ROS Gazebo simulator and our STATUS algorithms which form the core of our development of our object tracking drone for generic purposes.
Uploads
Papers by Prasad N R
Multiplication of weights may or may not be a practical biological solution (the hypotheses is that biological neurons do not have symmetric feedback and exact 32-bit multiplication). As an approximation of universal approximation theorem, bit-shift (which is a non-differentiable symbol) is proposed. There are two variants of this problem (both of which can be solved using backpropagation): First variant is addition of bit-shift with non-negative float (without subtraction); The second variant is addition and subtraction of integers (without float). So, combining these two, we get additive-shift of float. The accuracy of this version is only about 2% lesser than the original DNN. The idea is to try to avoid clock-cycles and make it easier to do inference almost at the speed of electricity in semiconductor (without memory or with relatively less memory). Another idea is to improve the ML-training and inference process so that they are compute-constrained.
Patent Journal: http://www.ipindia.nic.in/writereaddata/Portal/IPOJournal/1_4789_1/Part-1.pdf
Application number: 201841035016
Journal number: 38/2019
Journal date: 20/09/2019
Conference Presentations by Prasad N R
Multiplication of weights may or may not be a practical biological solution (the hypotheses is that biological neurons do not have symmetric feedback and exact 32-bit multiplication). As an approximation of universal approximation theorem, bit-shift (which is a non-differentiable symbol) is proposed. There are two variants of this problem (both of which can be solved using backpropagation): First variant is addition of bit-shift with non-negative float (without subtraction); The second variant is addition and subtraction of integers (without float). So, combining these two, we get additive-shift of float. The accuracy of this version is only about 2% lesser than the original DNN. The idea is to try to avoid clock-cycles and make it easier to do inference almost at the speed of electricity in semiconductor (without memory or with relatively less memory). Another idea is to improve the ML-training and inference process so that they are compute-constrained.
Patent Journal: http://www.ipindia.nic.in/writereaddata/Portal/IPOJournal/1_4789_1/Part-1.pdf
Application number: 201841035016
Journal number: 38/2019
Journal date: 20/09/2019