About the work
This Master's thesis focuses on a detailed review of various neural models, with special attention to transformer-based architectures applied to time series forecasting. While these models have demonstrated their effectiveness in fields such as natural language processing and artificial vision, their performance in time series prediction tends to be less notable. Within the framework of this project, the intent is to delve into the current state of the discipline by meticulously investigating preprocessing strategies and the various proposed architectures, both from the perspective of computational efficiency and in terms of temporal modeling. Finally, and with the knowledge acquired from this review of the state of the art, a customized neural architecture will be designed and evaluated with the aim of forecasting a time series selected from one of the benchmarks most frequently used in the discipline for the evaluation and comparison of neural models.
Creativity declaration
100% human created
Declaration Date:
Jun 24, 2023, 8:55 AM
Identification level:
Low
Print work information
Work information
Title Neural Models for Time Series Forecasting
This Master's thesis focuses on a detailed review of various neural models, with special attention to transformer-based architectures applied to time series forecasting. While these models have demonstrated their effectiveness in fields such as natural language processing and artificial vision, their performance in time series prediction tends to be less notable. Within the framework of this project, the intent is to delve into the current state of the discipline by meticulously investigating preprocessing strategies and the various proposed architectures, both from the perspective of computational efficiency and in terms of temporal modeling. Finally, and with the knowledge acquired from this review of the state of the art, a customized neural architecture will be designed and evaluated with the aim of forecasting a time series selected from one of the benchmarks most frequently used in the discipline for the evaluation and comparison of neural models.
Work type Research papers, Thesis, Lecture notes
Tags timestamp encodings, causal convolutions, tensorflow.keras, temporal complexity, positional encodings, multi-head attention, neural networks, transformers based models, time series, recurrent neural networks, sarima, dlinear, informer, fedformer, ast
-------------------------
Registry info in Safe Creative
Identifier 2306244668470
Entry date Jun 24, 2023, 8:55 AM UTC
License Creative Commons Attribution-NonCommercial-ShareAlike 4.0
-------------------------
Copyright registered declarations
Author 100.00 %. Holder José Antonio Esteban Rodríguez. Date Jun 24, 2023.
Information available at https://www.safecreative.org/work/2306244668470-neural-models-for-time-series-forecasting
/
Registration: 2306244668470
