Classification of integers based on residue classes via modern deep learning algorithms
From MaRDI portal
Abstract: Computing the residue when dividing a given integer by prime numbers like 2, 3, or others may appear trivial to human beings, but it can be less straightforward for computers in the absence of pre-defined algorithms. In this paper, we tested multiple deep learning architectures and feature engineering approaches on classifying large finite integers (up to ) based on their residues when divided by small prime numbers. It turns out that, regardless of the network architectures (CNN, RNN, Transformer, etc.) or the complexity of the networks, the ability of classification critically depends on the feature space fed into the deep learning models. We also evaluated commercially available Automated Machine Learning (AutoML) pipelines from Amazon, Google and Microsoft, and found that they failed to address this issue unless appropriately engineered features were provided. Furthermore, we introduced a method that utilizes linear regression on Fourier series basis vectors, and successfully demonstrated its effectiveness in the general case. Finally, we evaluated prompt-based learning approaches using GPT-J, Falcon-40B, and LLaMA and demonstrated its apparent failures. To conclude, feature engineering remains an important task to improve the performance, increase the interpretability, and reduce the complexity of models, even in the era of AutoML and Large Language Models (LLMs).
This page was built for publication: Classification of integers based on residue classes via modern deep learning algorithms
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6431936)
