The paper "Efficient Inverted Indexes for Approximate Retrieval over Learned Sparse Representations" by Sebastian Bruch, Cosimo Rulli, Franco Maria Nardini, and Rossano Venturini addresses the challenge of efficient retrieval over learned sparse representations, which are effective models for text retrieval but present unique difficulties due to their distributional differences from traditional lexical models like BM25. The authors propose a novel inverted index organization called SEismic (Spilled Clustering of Inverted Lists with Summaries for Maximum Inner Product Search) that enables fast and effective approximate retrieval over learned sparse embeddings. SEismic organizes inverted lists into geometrically cohesive blocks, each equipped with a summary vector, allowing for quick evaluation of blocks during query processing. Experimental results show that SEismic achieves sub-millisecond per-query latency on various sparse embeddings of the Ms Marco dataset while maintaining high recall, outperforming state-of-the-art inverted index-based solutions and graph-based methods by significant margins. The paper also includes an in-depth analysis of SEismic through ablation studies, demonstrating the impact of its components on performance.The paper "Efficient Inverted Indexes for Approximate Retrieval over Learned Sparse Representations" by Sebastian Bruch, Cosimo Rulli, Franco Maria Nardini, and Rossano Venturini addresses the challenge of efficient retrieval over learned sparse representations, which are effective models for text retrieval but present unique difficulties due to their distributional differences from traditional lexical models like BM25. The authors propose a novel inverted index organization called SEismic (Spilled Clustering of Inverted Lists with Summaries for Maximum Inner Product Search) that enables fast and effective approximate retrieval over learned sparse embeddings. SEismic organizes inverted lists into geometrically cohesive blocks, each equipped with a summary vector, allowing for quick evaluation of blocks during query processing. Experimental results show that SEismic achieves sub-millisecond per-query latency on various sparse embeddings of the Ms Marco dataset while maintaining high recall, outperforming state-of-the-art inverted index-based solutions and graph-based methods by significant margins. The paper also includes an in-depth analysis of SEismic through ablation studies, demonstrating the impact of its components on performance.