Architectural Diversity and Open Challenges

View the PDF file for the paper entitled “A comprehensive survey of deep learning to predict a time series: architectural diversity and open challenges, by Jongseon Kim and 4 other authors

PDF HTML (experimental) view

a summary:The prediction of the time chain is a task that provides basic information to make decisions. After traditional statistical and automatic learning methods, many primary deep learning structures such as MLPS, CNNS, RNNS and GNNS were developed. However, the structural restrictions resulting from the inductive biases for each deep educational structure restricting their performance. Transformer models, which excel in dealing with long -term consequences, have become important architectural components to predict a time chain. However, recent research has shown that alternatives such as simple linear layers can outperform transformers. These results have opened new possibilities for the use of various structures, starting from the basic deep learning models to emerging structures and mixed curricula. In this context, the architectural modeling to predict the time of time now entered the Renaissance. This poll not only provides a historical context for predicting the time chain, but also provides a comprehensive and timely analysis of the movement towards architectural diversification. By comparing and reconsidering deep learning models, we reveal new views and present modern trends, including hybrids, spread, mABA and Foundation. By focusing on the inherent characteristics of time chains’ data, we also address the open challenges that have gained interest in predicting the time chain, such as the dependency of the channel, the conversion of distribution, causal, and extraction of features. These contributions help reduce entry barriers to new expatriates by providing a systematic understanding of various fields of research in predicting time chains (TSF), while providing broader views to researchers or new opportunities through in -depth exploration of TSF challenges. (It was shortened by the 1920 -year -old ARXIV. The full version of the paper.)