IM-LPG: Inverse Modeling Approach to Laser Pulse Shape Generation in Inertial Confinement Fusion

Fusion energy remains one of the greatest scientific challenges of our time, with transformative potential for sustainable, carbon-free power. In Inertial Confinement Fusion (ICF), achieving successful implosions critically depends on the design of Laser Pulse (LP) shapes that can efficiently drive fusion targets within stringent physical constraints. Traditional LP design relies heavily on expensive simulations and manual iterative tuning, which limits scalability. We propose an Inverse Modeling Approach to Laser Pulse Shape Generation (IM-LPG) that maps target pellet parameters and desired fusion implosion outcomes directly to tailored LP shapes. IM-LPG supports both diffusion-based and autoregressive architectures, offering flexibility for diverse modeling scenarios. To balance accuracy and feasibility, we introduce a multi-objective training setup that produces LPs satisfying physical constraints while achieving <2\% error on implosion outcomes. Furthermore, we incorporate constraint conditioning through inpainting and gradient-based editing, enabling fine-grained control of pulse characteristics during generation.Our framework provides a data-driven, flexible, and controllable solution to LP design in ICF, representing a step toward accelerating the path to practical fusion energy.