Determine the practical efficiency of neural-network full-likelihood training

Determine whether the proposed full-likelihood procedure for simultaneously training neural-network models of the target density and the odds function can work efficiently in practice.

Background

The paper proposes using a full-likelihood objective with neural-network models for the target density and the odds function. When the target model can be sampled, Monte Carlo methods can approximate the normalization integrals and their gradients, making optimization feasible in principle.

The authors note that the resulting procedure is related to REINFORCE, whose Monte Carlo gradients can have high variance, and mention reparameterization as a possible remedy. They do not establish whether the proposed training strategy is computationally effective in real applications and explicitly defer that investigation to future work.

References

While the above procedure offers a possible way of training two neural networks simultaneously based on FL, it remains unclear if this procedure will work efficiently in practice. Investigating the practical utility of this neural network training procedure is beyond the scope of the current paper so we leave this for future work.

— On efficiency gains via augmenting a tiny sample with a massive auxiliary sample  (2608.26610 - Chen, 27 Aug 2026) in Section 4, subsection “Practical challenges”