We will construct an Artificial Intelligence (AI) agent for PDF determination.

The motivation is to arrive at a PDF determination which is optimized without any bias in the optimization choices. This goes beyond the current NNPDF methodology in three respects. First, a single AI agent will replace a Monte Carlo set of neural networks trained to data. Second, the optimization of this agent will not rely on tuning a specific neural architecture and a genetic algorithm used for its minimization. Third, the validation of the method will not be based on an external closure test but rather it will be an outcome of the learning process.

The main challenge is the application to PDF determination of deep-reinforcement learning and Q-network techniques.

As an intermediate goal, we will replace the genetic minimization which is currently adopted with gradient based methods such as stochastic gradient method, and we will replace the neural network architecture currently adopted by NNPDF with architectures based on the deep neural network paradigm.

The final deliverables will include both a suite of tools for PDF determination, such as a framework optimized to solve the problem of convolution regression of PDFs, and first PDF sets determined using them, developed in close collaboration with NNPDF.