Dr. Franco Maria Nardini (CNR-ISTI, Italy) delivered an engaging and thought-provoking keynote at the International Conference on Similarity Search and Applications 2025 (SISAP 2025), held on October 1-3 in Reykjavik, Iceland.
Dr. Nardini’s talk, titled “Efficient Retrieval over Learned Representations of Text: Challenges and Opportunities,” offered an insightful exploration of how transformer-based large language models (LLMs) are revolutionizing the way we index and retrieve information from large text collections.
In recent years, LLMs have enabled the creation of high-dimensional, contextual representations of text—both dense and sparse—that capture semantic meaning far beyond traditional keyword-based methods. These representations allow systems to identify semantically similar items through nearest-neighbor search, driving major progress in information retrieval, recommender systems, and retrieval-augmented generation.
During his keynote, Dr. Nardini reviewed the state-of-the-art advancements that make efficient indexing and retrieval possible with these learned embeddings. He also highlighted the current challenges—such as scalability, efficiency, and interpretability—and outlined exciting research opportunities in this rapidly evolving field.
His presentation underscored how these innovations not only improve retrieval performance but also reshape how we think about search, recommendation, and knowledge grounding in AI systems.
