The role of Inria and academia-industry collaboration in artificial intelligence and the outcomes of OcéanIA: Inria Chile's contributions to KHIPU 2025

From March 10 to 14, 2025, Inria Chile participated in various activities at Khipu, the Latin American meeting on artificial intelligence that brought together more than 800 students, researchers, academics, representatives of startups, and major tech companies. The event featured talks, discussion panels, workshops, and poster presentations from scientists and students across Latin America and major tech companies and startups, such as Mistral AI_from France.

The event, of which Inria Chile was a strategic partner, took place at the San Joaquín Campus of the Pontificia Universidad Católica de Chile and the CEINA Cultural Center, showcasing the dynamism and growth of Latin America's artificial intelligence ecosystem.

In this context, Nayat Sánchez-Pi, Director of Inria Chile, was invited to participate in the panel 'Research in Academia and Industry,' a roundtable discussion moderated by Guillermo Sapiro, Professor at Princeton University and Distinguished Engineer at Apple, Inc. The panel included Laura Alonso Alemany, Professor at the Universidad Nacional de Córdoba, Argentina; Nando De Freitas, Vice President of Artificial Intelligence at Microsoft AI; Jeff Dean, Chief Scientist at Google Research and Google DeepMind; and Nayat Sánchez-Pi, Director of Inria Chile and the Franco-Chilean Binational Center for Artificial Intelligence.

The session focused on the differences between working in academia and in large corporations, but also on the ways both sectors can collaborate to benefit in research and technology development mutually. The unique contributions of each sector were highlighted, from the industry's role in solving real-world problems to the freedom of research and student training in academia, as well as the particular role of institutions like Inria.

The Director of Inria Chile, Nayat Sánchez-Pi, was able to address Inria's unique mission, which, as a public research institute, not only conducts cutting-edge research in artificial intelligence but also seeks to have an impact on society and the public interest by making research results available to society. This is achieved through close collaboration with major companies, creating deep-tech startups, making open-source software and libraries like scikit-learn available to the community, or participating in the training of research talents.

Nayat Sánchez-Pi also explained that often, Inria's mission is “to assume the risk of carrying out some research that we see the value but other players of the ecosystem are unable to do,” illustrating her observations with some historical data on the impact Inria has had over time by supporting the scientists who do this work and providing effective means to transfer the research results to society: “In the 70s we worked in computer networks and created the foundation of the IP protocol, or created the bases of database systems, with results like postgreSQL or snowflake that come from Inria researchers, in the 90s we set the foundation of the web, with results such as XML and CSSs coming from Inria researchers, and now, it not a surprise, that we are at the origins of the AI revolution.”

At Khipu, the Poster Sessions allowed researchers and students from various parts of Latin America to present their work and research findings, sharing and exchanging knowledge with other attendees and researchers present.

On this occasion, Inria Chile had the opportunity to present two outcomes of the OcéanIA project, a Franco-Chilean initiative that leverages artificial intelligence to unravel the impact of oceans on climate change and biodiversity. The OcéanIA research team has made significant progress in understanding the ocean's role in climate change and is now a AI toolbox made available for this purpose.

The first poster presented at Khipu 2025 was 'Decoding Marine Microbiome Life via Explainable Machine Learning,' presented by Luis Valenzuela, a postdoc at Inria Chile. He demonstrated how the use of explainable machine learning models allows for predicting the relationship between metagenomic, metatranscriptomic, taxonomic information, and environmental variables. Luis also highlighted the potential of large language models, ranging from 600 to 40,000 million parameters, which have been pre-trained with metagenomic data, to identify new protein functions and taxonomies from sequences in the Tara Ocean dataset, a member institution of the OcéanIA project.

On the other hand, José Vasquez-Bastías, a student at the Universidad Autónoma de Chile and intern at Inria Chile in 2024, presented the results of his internship with the poster 'Symbolic Regression and LLM Synergy: Enhancing Explainable in Genomic Predictions.' In this poster, José showcased studies of massive language models (125 -72,000M parameters) trained on natural language and/or mathematical reasoning tasks to automate the analysis of equations generated by symbolic regressions, thereby enhancing the interpretation and optimization of genomic models.