Get Started
Home
Topics
Search
Library
Research questionHow can residual quantization encode semantic IDs while preserving graded similarity and directional residual variation?Hierarchical semantic IDs assign codewords across residual-quantization levels for retrieval and recommendation. Shared corpus structure can consume early codebook capacity, while hard assignments and full-codeword subtraction may lose graded similarity and directional residual information.
AI
Information Retrieval
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.PRQ-KMeans: Projection Residual Quantization for Semantic ID TokenizationThe setting is post-hoc semantic-ID construction, where residual construction is not constrained by input reconstruction. Evidence comes from a large-scale industrial search dataset and four public recommendation benchmarks, comparing the proposed tokenizer with other evaluated tokenizers.research paper · Sep 3, 2026
Related questions
How can discrete visual quantizers be compared fairly at matched coding rates?How can discrete audio tokenizers preserve semantics and acoustic fidelity for both understanding and generation?How can low-bit KV caches save autoregressive decoding memory without losing long-context retrieval quality?How can extreme low-bit LLM quantization adapt its basis across layers, calibration data, and quantizers?