Solving Regression Models with First Order Stochastic Based Optimizers
| Autor: | Brian Keegan, Natacha Gueorguieva, Iren Valova |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Artificial neural network
Computer science business.industry Deep learning Computer Science::Neural and Evolutionary Computation Regression analysis Perceptron Convolutional neural network Backpropagation ComputingMethodologies_PATTERNRECOGNITION Stochastic gradient descent Softmax function Artificial intelligence business |
| Zdroj: | NER |
| Popis: | Most of the regression-based Deep Learning (DL) algorithms which were recently proposed are based on Convolutional Neural Networks (CNN) trained by using l 2 loss function. In order to avoid the vulnerability of it to outliers, some authors propose to combine regression methods with a classification function like combination of l 2 loss function and softmax or a strategy based on bounding boxes what improves the performance of the DL neural network.In this research we propose using a combination of first order Stochastic Gradient Descent (SGD) optimizer for Artificial Neural Networks (ANNs) with Deep Learning (DL) and Deep Learning Multilayered Perceptron DLMLP with Back Propagation (BP). We focus on optimization of stochastic objectives, parameters and implementation of different activation functions in solving medical regression problems in high-dimensional space. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|