Learned Sparse IR models, such as SPLADE, offer an excellent efficiency-effectiveness tradeoff. However, they rely on the underlying backbone vocabulary, which might hinder performance (polysemicity and synonymy) and pose a challenge for multi-lingual and multi-modal usages. To solve this limitation, we propose to replace the backbone vocabulary with a latent space of semantic concepts learned using Sparse Auto-Encoders (SAE). Throughout this paper, we study the compatibility of these 2 concepts, explore training approaches, and analyze the differences between our SAE-SPLADE model and traditional SPLADE models. Our experiments demonstrate that SAE-SPLADE achieves retrieval performance comparable to SPLADE on both in-domain and out-of-domain tasks while offering improved efficiency.
@article{arxiv.2604.21511,
title = {From Tokens to Concepts: Leveraging SAE for SPLADE},
author = {Yuxuan Zong and Mathias Vast and Basile Van Cooten and Laure Soulier and Benjamin Piwowarski},
journal= {arXiv preprint arXiv:2604.21511},
year = {2026}
}
Comments
11 pages, 3 figures, 9 tables. To appear at SIGIR 2025