AutoML Co-Designer: A Collaborative Framework for Accelerated AI Development

Descripción del proyecto

Devolver al ser humano al diseño automatizado de la inteligencia artificial

Para crear sistemas de inteligencia artificial (IA) fiables hay que tomar decisiones complejas, desde seleccionar el modelo adecuado hasta ajustar los numerosos parámetros que influyen en su aprendizaje. Las herramientas de aprendizaje automático automatizado (AutoML, por sus siglas en inglés) pueden simplificar este proceso probando distintos modelos y ajustando sus hiperparámetros. Por ejemplo, con estos sistemas se podrían comparar diferentes arquitecturas de redes neuronales en un conjunto de datos determinado, como la puntuación crediticia o la satisfacción del cliente. Teniendo esto en cuenta, el equipo del proyecto AutoML Co-Designer, financiado por el CEI, pretende que estos procesos de diseño sean colaborativos. En concreto, desarrollará una plataforma prototipo en la que los desarrolladores puedan guiar los sistemas de AutoML, aportar conocimientos del dominio e interpretar los resultados. El objetivo es acelerar el desarrollo de la IA y garantizar aplicaciones más fuertes.

Objetivo

The growth of the AI market depends on our ability to rapidly and efficiently develop robust AI applications, for which AI developers can be supported by Automated Machine Learning (AutoML), e.g. in model selection, neural network design, or hyperparameter settings. However, current AutoML systems are designed as rigid 'black boxes' and thus create a critical bottleneck. They sideline expert developers, preventing the integration of crucial domain knowledge. This inflexibility slows down development cycles, hinders innovation, and limits the creation of sophisticated, market-ready AI solutions for complex, high-value problems.

This project, AutoML Co-Designer, will deliver the proof of concept for a collaborative development paradigm that directly addresses this efficiency gap. Our objective is to validate a framework that transforms AutoML from a simple automation tool into a powerful interactive system between AutoML and AI developers.
We will engineer and validate a prototype platform with novel interactive interfaces. This platform will serve as the core of our proof of concept, enabling developers to inject domain knowledge, steer optimization towards market-critical goals like robustness and efficiency, and gain actionable insights through advanced explanations.

The successful validation of the AutoML Co-Designer framework will provide a clear business case for a new class of AI development tools. This PoC will establish a new framework for efficiency, demonstrating a faster path from idea to deployment. It will unlock new markets for specialized AI applications previously too complex or resource-intensive to build. This project will lay the groundwork for a commercially viable product that empowers development teams to innovate faster, creating a competitive advantage in the rapidly expanding AI economy.