[KongchangAI]
ConceptResidual Quantization / RQ

Residual Quantization

A hierarchical quantization technique that performs successive quantization of residual errors after an initial vector quantization step, accumulating corrections layer by layer. This approach achieves finer discrete representations at a relatively low codebook cost and is commonly combined with VQ-VAE to improve reconstruction accuracy.

Timeline (last 90 days)

Oct 26

Semantic IDs are usually obtained by quantizing dense semantic representations of items, with common approaches including K-means clustering, residual quantization (RQ), and product quantization (PQ).

Unverified50%

All Facts (1)

Unverified

Semantic IDs are usually obtained by quantizing dense semantic representations of items, with common approaches including K-means clustering, residual quantization (RQ), and product quantization (PQ).

50%

Source Articles

Cross-Item and Cross-Encoding-Space Knowledge Distillation: A New Approach for Parallel Generative RecommendationOct 26Lossy Text Autoencoders: Where Compression Meets Semantic RepresentationOct 9