Research map / Architectures and efficiency / Quantization
ASVQ: What Reparameterization Is Just Enough for Efficient Codebook Learning?
Wenfeng Zou
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes ASVQ, a lightweight codebook reparameterization for vector quantization that aims to avoid codebook collapse without the overhead of expressive reparameterizations.
Problem
Codebook collapse in vector quantization models hampers efficient discrete representation learning; recent expressive codebook reparameterizations help but add optimization and computational overhead.
Approach
Asks what minimal reparameterization scheme suffices for efficient codebook learning, and proposes a simple scheme (ASVQ) that improves codebook adaptability at low overhead compared with expressive reparameterizations.
Experimental setup
Baselines: expressive codebook reparameterization methods
Builds on: vector quantization VQ-VAE
Closest papers by meaning
- Codebook Transfer with Part-of-Speech for Vector-Quantized Image Modeling (CVPR 2024 · similarity 0.90)
- Learning to Quantize for Training Vector-Quantized Networks (ICML 2025 · similarity 0.89)
- LG-VQ: Language-Guided Codebook Learning (NeurIPS 2024 · similarity 0.88)
- Finite Scalar Quantization: VQ-VAE Made Simple (ICLR 2024 · similarity 0.87)
- Restructuring Vector Quantization with the Rotation Trick (ICLR 2025 · similarity 0.87)
- Residual Quantization with Implicit Neural Codebooks (ICML 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
