ADELIA

Automatic Differentiation for Efficient Laplace Inference Approximations

ADELIA: Automatic Differentiation for Efficient Laplace Inference Approximations

The first AD-enabled INLA implementation, replacing finite-difference gradients with a structure-exploiting multi-GPU backward pass.

Achieves 4.2–7.9× per-gradient speedups and 5–8× energy savings over DALIA (the SOTA), and enables reliable convergence on large models where finite-difference stalls.

  • Role: Lead author (PhD work)
  • Methods: Automatic differentiation, multi-GPU, INLA, structure-exploiting backward pass
  • Status: Under review at SC ‘26
  • Paper: arXiv:2605.06392

References