Additive Quantization for Extreme Vector Compression

By Artem Babenko, Victor S. Lempitsky. 2014

319 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Efficiency & Compression

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Efficiency & Compression

Citations in the corpus

Quantization based Fast Inner Product Search (2015, background): “Unlike the composite quantization [10] or additive quantization [2], our lookup table construction has the same speed as that of the original product quantization [11] and does not grow with the length of the code.”
Multiscale Quantization for Fast Similarity Search (2017, methodology): “This includes additive quantization [5, 6, 26], composite quantization [37, 38] and stacked quantization [27].”
Shared Predictive Cross-Modal Deep Quantization (2018, unknown): “Subsequently, in the shared subspace, we learn common representations by using label alignment and adopt additive quantization [47] to obtain the compact codes.”
The Faiss Library (2024, unknown): “As such, we measure the speed and accuracy of several index types and hyperparameter settings under a ﬁxed memory budget.”
A Survey on Learning to Hash (2016, methodology): “an approximation of a data item has already been studied in the signal processing research area, known as multi-stage vector quantization, residual quantization, ormore generally structured vector quantization [38], and recently re-developed for similarity search under the Euclidean distance (additi”
Polysemous Codes (2016, background): “Subsequent works [5,57,58] have pushed possible memory/eﬃciency trade-oﬀ by adopting a more general point of view, such as “Ad-ditive quantization” [5], which provides an excellent approximation and search performance, yet obtained with a much higher computational encoding cost [7].”
Sparse composite quantization (2015, methodology): “Additive quantization [1, 5] further improves the compositionabilty by approximating a database vector using the summation of dictionary words selected from different dictionaries, whose idea is similar to structured vector quantization [9] (a.”
Efficient Indexing of Billion-Scale Datasets of Deep Descriptors (2016, background): “Several works [8, 3, 21, 6] have proposed modifications of PQ that do not decompose the dataspace into orthogonal subspaces.”
Hashing for Similarity Search: A Survey (2014, unknown): “The idea of using the summation of several dictionary items as an approximation of a data item has already been studied in the signal processing area, known as multi-stage vector quantization, residual quantization, or more generally structured vector quantization [34], and recently re-developed for”
Accelerating Large-Scale Inference with Anisotropic Vector Quantization (2019, methodology): “…for binary quantization (He et al., 2013; Liong et al., 2015; Dai et al., 2017), product quantization (Jegou et al., 2011; Guo et al., 2016b; Zhang et al., 2014; Wu et al., 2017), additive quantization (Babenko & Lempitsky, 2014; Martinez et al., 2018), and ternary quantization (Zhu et al., 2016).”
Efficient Large-Scale Approximate Nearest Neighbor Search on the GPU (2016, background): “However, reported results from previous state-of-the-art methods [11, 9, 3, 2] excluded expensive query-dependent pre-computations from the timing of their online-phase.”
Jointly Optimizing Query Encoder and Product Quantization to Improve Retrieval Performance (2021, background): “Some variants [2, 35] use different architectures to better minimize the reconstruction error, while training and searching is more complex.”
Learning Binary Codes for Maximum Inner Product Search (2015, background): “Besides hashing, vector quantization (VQ) [2, 6, 10, 27, 40, 41] is another promising direction for large-scale search problems in high-dimensional space.”
Autoregressive Image Generation using Residual Quantization (2022, methodology): “As a generalized version of PQ, additive quantization (AQ) [1] uses the dependent vectors in the codebook, but finding the codes is an NP-hard task [6].”
MRVQ: One Resident Index for Dimension- and Rate-Elastic Vector Search (2026, unknown): “PQ, OPQ, and additive quantization are standard fixed-rate compressors [Jégou et al., 2011, Ge et al., 2013, Babenko and Lempitsky, 2014].”