Quantum Inspired Encoding Strategies for Machine Learning Models: Proposing and Evaluating Instance Level, Global Discrete, and Class Conditional Representations
This work addresses encoding efficiency for researchers using quantum-inspired methods in classical ML, but it is incremental as it builds on existing quantum encoding concepts without introducing a new paradigm.
The study tackled the problem of high encoding time in quantum-inspired data transformations for classical machine learning by proposing and evaluating three encoding strategies (ILS, GDS, CCVS), finding that these strategies reduced encoding time while maintaining classification performance, though specific numerical results were not provided.
In this study, we propose, evaluate and compare three quantum inspired data encoding strategies, Instance Level Strategy (ILS), Global Discrete Strategy (GDS) and Class Conditional Value Strategy (CCVS), for transforming classical data into quantum data for use in pure classical machine learning models. The primary objective is to reduce high encoding time while ensuring correct encoding values and analyzing their impact on classification performance. The Instance Level Strategy treats each row of dataset independently; mimics local quantum states. Global Discrete Value Based encoding strategy maps all unique feature values across the full dataset to quantum states uniformly. In contrast, the Class conditional Value based encoding strategy encodes unique values separately for each class, preserving class dependent information. We apply these encoding strategies to a classification task and assess their impact on en-coding efficiency, correctness, model accuracy, and computational cost. By analyzing the trade offs between encoding time, precision, and predictive performance, this study provides insights into optimizing quantum inspired data transformations for classical machine learning workflows.