A Reproducibility Study of Variance-Aware Quantization

We propose a reproducibility study of Variance-Aware Quantization (VAQ). We propose an implementation in Rust. Moreover, we also experimentally compare the performance of VAQ with the other state-of-the-art methods, in particular, Product Quantization and Optimized Product Quantization. Finally, we propose a method to improve the performance of VAQ by using the HNSW algorithm, a popular—graph-based k-nearest neighbors search method.