Search papers, labs, and topics across Lattice.
This paper explores replacing the token-based vocabulary of SPLADE models with a latent space of semantic concepts learned via Sparse Auto-Encoders (SAE) to improve performance and generalization. They investigate training methodologies for SAE-SPLADE and compare its characteristics to traditional SPLADE. Results show that SAE-SPLADE achieves comparable retrieval performance to SPLADE on both in-domain and out-of-domain tasks, while also improving efficiency.
SPLADE models can ditch their token-based vocabularies for a latent semantic space learned by Sparse Auto-Encoders, maintaining retrieval performance while boosting efficiency.
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.
