Choummanivong, M Mattis (2019) Auto-ML applied to LSTM Neural Networks using the Minimum Nescience Principle PRE - Research Project, ENSTA.



This report investigates Automated Machine Learning using the Minimum Nescience Prin- ciple applied to univariate time series forecasting using LSTM Neural Networks. Nescience is a philosophical concept meaning lack of knowledge. Researchers at IMDEA Net- works Institute want to transform this concept in a new fitness function that can be used in Automated Machine Learning. The concept of Nescience was originally tuned to classification problems using binary search trees by IMDEA researchers. This report aims to show how, over the course of this internship the Nescience concept was adjusted to regression problems and LSTM Neural Networks. It also compares two Automated Machine Learning methods using the Minimum Nescience Principle: a random search and a grammatical evolution.

Item Type:Thesis (PRE - Research Project)
Subjects:Mathematics and Applications
ID Code:7529
Deposited By:Mattis Choummanivong
Deposited On:05 juill. 2021 11:05
Dernière modification:05 juill. 2021 11:05

Repository Staff Only: item control page