Talk:TurboQuant

Concerns over prior work and methodology

Researchers behind DRIVE (NeurIPS 2021) and EDEN (ICML 2022) argued in a technical note (arXiv:2604.18555) and an OpenReview discussion (tO3ASKZlok) that TurboQuant substantially overlaps with their earlier quantization frameworks. They noted that the original TurboQuant submission did not cite DRIVE or EDEN despite these similarities, and further claimed that TurboQuant can be viewed as a restricted or biased special case of both frameworks and that EDEN achieves stronger theoretical guarantees and better empirical performance on several benchmarks used in the TurboQuant paper.

Another technical note (arXiv:2604.19528) and open review discussion (tO3ASKZlok) by researchers mainly from NTU and ETH Zürich has raised three documented concerns about this paper: (1) methodological description accuracy, (2) theoretical characterisation of RaBitQ, and (3) experimental baseline asymmetry.

I propose adding a 'Criticism and Responses' section sourcing these notes and the public ICLR OpenReview record. Tamim Ehsan (talk) 03:43, 16 May 2026 (UTC)