Multi-Branch Multi Layer Perceptron: A Solution for Precise Regression using Machine Learning

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216305%3A26220%2F23%3APU148281" target="_blank" >RIV/00216305:26220/23:PU148281 - isvavai.cz</a>

Výsledek na webu

<a href="https://ieeexplore.ieee.org/document/10109076" target="_blank" >https://ieeexplore.ieee.org/document/10109076</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1109/RADIOELEKTRONIKA57919.2023.10109076" target="_blank" >10.1109/RADIOELEKTRONIKA57919.2023.10109076</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Multi-Branch Multi Layer Perceptron: A Solution for Precise Regression using Machine Learning

Popis výsledku v původním jazyce

The problem of simple regression using Multi Layer Perceptron (MLP) has its limitations. The main problem with MLP is that it is difficult to find a perfect architecture to fit all the data. The more complex and multi-dimensional the data we use, the deeper the network has to be, which increases training time as well as optimization and tuning of the network. The solution to fit the data more precisely could be to split the data into groups based on the input variable and use a different model to train and predict data for each of these groups. In most cases, datasets contain a large number of non-linear input features, which makes the method for finding thresholds for each group very difficult. The proposed technique tackles this problem using multiple branches consisting of shallow MLP, one of which acts as a selector (classifier) for the output. The selector's main goal is to generate a weight for the other prediction branches which in other words means, results in one branch being more proficient in predicting certain parts of the feature space while the other branches will be better at predicting completely different parts of the feature space.

Název v anglickém jazyce

Multi-Branch Multi Layer Perceptron: A Solution for Precise Regression using Machine Learning

Popis výsledku anglicky

The problem of simple regression using Multi Layer Perceptron (MLP) has its limitations. The main problem with MLP is that it is difficult to find a perfect architecture to fit all the data. The more complex and multi-dimensional the data we use, the deeper the network has to be, which increases training time as well as optimization and tuning of the network. The solution to fit the data more precisely could be to split the data into groups based on the input variable and use a different model to train and predict data for each of these groups. In most cases, datasets contain a large number of non-linear input features, which makes the method for finding thresholds for each group very difficult. The proposed technique tackles this problem using multiple branches consisting of shallow MLP, one of which acts as a selector (classifier) for the output. The selector's main goal is to generate a weight for the other prediction branches which in other words means, results in one branch being more proficient in predicting certain parts of the feature space while the other branches will be better at predicting completely different parts of the feature space.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

20200 - Electrical engineering, Electronic engineering, Information engineering

Projekt

—

Návaznosti

S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2023

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

RADIOELEKTRONIKA 2023: 2023 33rd International Conference Radioelektronika

ISBN

979-8-3503-9834-2

ISSN

—

e-ISSN

—

Počet stran výsledku

5

Strana od-do

„“-„“

Název nakladatele

IEEE

Místo vydání

NEW YORK

Místo konání akce

Pardubice

Datum konání akce

19. 4. 2023

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

000990505700049