📄 中文摘要
该研究提出了一种结合多分辨率学习（MRL）与int8和二进制量化的方法，以平衡基础设施成本与检索准确性。通过对比不同的量化技术和马特里奥什卡嵌入，研究展示了如何在保持高效检索性能的同时显著降低计算和存储成本。结果表明，采用这些技术可以实现高达80%的成本节约，从而为大规模向量搜索提供了可行的解决方案。
扩展向量搜索：比较量化与马特里奥什卡嵌入以实现80%的成本降低
📄 English Summary
Scaling Vector Search: Comparing Quantization and Matryoshka Embeddings for 80% Cost Reduction
This study presents a method that combines Multi-Resolution Learning (MRL) with int8 and binary quantization to balance infrastructure costs with retrieval accuracy. By comparing various quantization techniques and Matryoshka embeddings, the research demonstrates how to significantly reduce computational and storage costs while maintaining efficient retrieval performance. The results indicate that employing these techniques can achieve up to an 80% cost reduction, providing a viable solution for large-scale vector search.