Machine Learning: Hardware Optimized for Machine Learning Computations Using Mixed-signal Inputs and Reconfigurable Parameters
Abstract
The aim of this project is to develop customizable hardware that can perform Machine Learning tasks. Machine Learning is the science of leveraging advance statistics and data mining to "teach" computers how to recognize patterns and perform tasks without direct human instructions. Circuits optimized for machine learning using mixed-signal inputs will be able to act as a dedicated hardware for performing conventional computational tasks required in learning systems; improving both efficiency and power consumption.
The hardware will be comprised of an array of neurons, an activation-function block at the output of every neuron, and a back propagation protocol for every layer. The use of both digital and analog inputs will provide us with a means for not only faster computations, but also more intuitive results. The project will focus on answering the question: If and how we can implement an efficient circuit that uses both analog and digital inputs to train a device to learn patterns from data.
Subject
Mixed-signal circuitsMachine learning
Neural network
Machine-learning hardware
Analog computation
Citation
Moronfoye, Oluwaseyi (2019). Machine Learning: Hardware Optimized for Machine Learning Computations Using Mixed-signal Inputs and Reconfigurable Parameters. Undergraduate Research Scholars Program. Available electronically from https : / /hdl .handle .net /1969 .1 /175388.