Spreading Vectors For Similarity Search

Abstract

Discretizing multi-dimensional data distributions is a fundamental step of modern indexing methods. State-of-the-art techniques learn parameters of quantizers on training data for optimal performance, thus adapting quantizers to the data. In this work, we propose to reverse this paradigm and adapt the data to the quantizer: we train a neural net which last layer forms a fixed parameter-free quantizer, such as pre-defined points of a hyper-sphere. As a proxy objective, we design and train a neural network that favors uniformity in the spherical latent space, while preserving the neighborhood structure after the mapping. We propose a new regularizer derived from the Kozachenko--Leonenko differential entropy estimator to enforce uniformity and combine it with a locality-aware triplet loss. Experiments show that our end-to-end approach outperforms most learned quantization methods, and is competitive with the state of the art on widely adopted benchmarks. Furthermore, we show that training

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keysablayrolles2018spreading

Related papers

• Accurate Deep Representation Quantization With Gradient Snapping Layer For Similarity Search (2016)0.00
• Interleaved Composite Quantization For High-dimensional Similarity Search (2019)0.00
• Leanvec: Searching Vectors Faster By Making Them Fit (2023)0.00
• Quantization Meets Projection: A Happy Marriage For Approximate K-nearest Neighbor Search (2024)0.00
• Deep Metric Learning Using Similarities From Nonlinear Rank Approximations (2019)2.26
• Nearest Neighbor Search With Compact Codes: A Decoder Perspective (2021)3.58
• Lossless Compression Of Vector Ids For Approximate Nearest Neighbor Search (2025)11.11
• Central Similarity Quantization For Efficient Image And Video Retrieval (2019)23.49