🧩Collage Lab

An R&D Lab focusing on Continual, Decentralized Compositionality for Sustainable Artificial Intelligence

Current unsustainable AI monolithic approach: build, use and throw away (left); COLLAGE main approach (right): incrementally build, share and re-use knowledge and skills as a set of compositional building blocks that can be efficiently and effectively composed, adapted, and ultimately used for the specific task at hand in a decentralized network of AI applications.

The COLLAGE Lab advances the vision of Sustainable Artificial Intelligence understood as the pursuit of efficiency and democratization in AI development and use. While modern AI has achieved remarkable progress, it remains heavily dependent on vast computational and data resources, limiting accessibility and long-term scalability. At COLLAGE, we address this challenge by rethinking AI from the ground up through the concept of reusability—the idea that AI systems should be designed to evolve, adapt, and build upon existing knowledge rather than being recreated from scratch.

Reusability serves as the foundation of our approach and can only be realized through continuality, compositionality, and decentralization. Continuality ensures that AI models can grow and adapt over time without catastrophic forgetting. Compositionality allows new systems to emerge from the combination and refinement of existing ones. Decentralization enables the open exchange and collective improvement of AI components across decentralized peer-to-peer networks.

Together, these principles define a new paradigm for efficient and democratic AI—one that promotes shared progress and reduces the structural inefficiencies of current, siloed AI ecosystems. The COLLAGE Lab thus represents a shift from isolated, application-specific AI toward a reusable, adaptable, and collectively built infrastructure for intelligent systems. By developing the theoretical and methodological underpinnings of this vision, the lab aims to chart a path toward a truly sustainable AI—one that is not only computationally efficient but also accessible, transparent, and inclusive by design. The group is supported by several grants including the prestigious FIS2 - Starting Grant (a.k.a. the Italian ERC) with a budget over 1.3M euros.

Members

Mentees

PreviousTowards Sustainable AINextBio & CV

Last updated