
This is DOI page.
Paper ID:26204501
ISSN:2760-8646:Online
Authos: Rikka Botan
Title: SSE: A Parameter-Efficient Training Recipe for Static Retrieval Embeddings
Publisher: Algorithm_Lab.
Conference: ICAITD 2026: The Third International Conference of AI New Technologies and Open Discussions
Location: Hokutopia, Oji, Kita-City, Tokyo Japan
Date: 12-15 October 2026
Editor: Kazuo Ohzeki (Algorithm_Lab. Professor Emeritus of Shibaura Institute of Technology, Professor of Emeritus of International Professional University of Technology in Tokyo)
Citation: Rikka Botan. SSE: A Parameter-Efficient Training Recipe for Static Retrieval Embeddings. In Proceedings of the Conference ICAITD 2026 (pp. 1–12). Algorithm_Lab. https://doi.org/10.63211/j.p.26.204501
doi: https://doi.org/10.63211/j.p.26.204501
pages: 1-12
Abstract:
Static retrieval embeddings provide fast inference but often trail contextual encoders in accuracy. We present SSE (Stable Static Embedding), a complete training recipe that combines contrastive ranking, Matryoshka prefix supervision, a 15-source data mixture, large-batch one-epoch optimization, and a lightweight post-pooling transform. The released checkpoint has 15.6M parameters, approximately half the 31.3M parameters of the prior static reference. Its 256-dimensional prefix reaches NanoBEIR (English) nDCG@10 of 0.5036, matching the reference’s native 1024-dimensional score of 0.5030 with one quarter of the vector storage; the full 512-dimensional representation reaches 0.5159. The model encodes 119,832 queries per second on CPU, has the highest measured quality within the throughput-comparable static family, and lies on the measured speed–quality Pareto frontier. These results support SSE as a parameter- and storage-efficient recipe for static dense retrieval.
Keywords: static retrieval embeddings; parameter efficiency; dense retrieval; Matryoshka representation learning; NanoBEIR
Category: Full paper
Review process: Committee review
Publication date: October 20 2026
First submission date: Aug. 10 2026
Accepted date: Sept. 15 2026
Copyright :Author(Full), AlgorithmLab.(First-in-the-world publishing rights granted by the author as the Prceedings of ICAITD 2026)
Licence:Viewer can download and view this review paper, but cannot secondary distribute (redistribute) it. (It is not Creative Commons License, nor MIT Licence) In other words, "Do not distribute" is the License.






