[4] US Army CERDEC. "Performance standards for night vision goggles." MIL-STD-3009G, 2022.

[3] Vann, J. G., & Zhang, W. "Recurrent priors for low-light video." CVPR 2024: 887-896.

Standard maximum likelihood fails due to near-zero counts. We instead maximize the Evidence Lower Bound (ELBO): [ \mathcalL(\theta,\phi;\mathbfx) = \mathbbE q \phi(\mathbfz [\log p_\theta(\mathbfx|\mathbfz)] - D_KL(q_\phi(\mathbfz|\mathbfx) | p(\mathbfz)) ] where ( q_\phi ) is a neural encoder (inference network) and ( p_\theta ) is a decoder that maps latent variables to Poisson rates.

[2] Nováková, L. "SPAD arrays with on-chip histogramming." JSSC 2025; 60(3): 550-562.

[5] Itzler, M., et al. "Single-photon counting for night vision." IEEE JSTQE 2024; 30(2): 1-14.

[6] Kingma, D. P., & Welling, M. "Auto-encoding variational Bayes." ICLR 2014.

It is important to clarify that does not refer to a specific, publicly documented scientific paper or a widely known academic model (like "Elli-Nova Neural Variational Gradient").