> ML_LITERATURE // LLOYD-1982-LEAST-SQUARES-QUANTIZATION-PCM_v1.0
Least squares quantization in PCM
Stuart P. Lloyd · IEEE Transactions on Information Theory (1982)
foundational1982foundationalthirdPartyReproduced
Principal Contribution
Standard mathematical formulation of the k-means clustering algorithm (Lloyd algorithm) for optimal scalar and vector quantization.
Operational Relevance
Underpins vector quantization in FAISS, modern codebook neural generation (VQ-VAE), and customer segmentation.
Assumptions
- Continuous probability densities can be approximated by a discrete set of representative centroid points minimizing mean squared quantization error
Limitations
- Sensitive to initial seeds; prone to trapping in suboptimal local minima