2000 character limit reached
A Comparison of PDF Projection with Normalizing Flows and SurVAE (2311.14412v2)
Published 24 Nov 2023 in cs.LG and stat.ME
Abstract: Normalizing flows (NF) recently gained attention as a way to construct generative networks with exact likelihood calculation out of composable layers. However, NF is restricted to dimension-preserving transformations. Surjection VAE (SurVAE) has been proposed to extend NF to dimension-altering transformations. Such networks are desirable because they are expressive and can be precisely trained. We show that the approaches are a re-invention of PDF projection, which appeared over twenty years earlier and is much further developed.
- P. M. Baggenstoss, “A theoretically optimum approach to classification using class-specific features.” Proceedings of ICPR, Barcelona, 2000.
- ——, “The PDF projection theorem and the class-specific method,” IEEE Trans Signal Processing, pp. 672–685, March 2003.
- ——, “On the duality between belief networks and feed-forward neural networks,” IEEE Transactions on Neural Networks and Learning Systems, pp. 1–11, 2018.
- I. Kobyzev, S. J. D. Prince, and M. A. Brubaker, “Normalizing flows: An introduction and review of current methods,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 11, pp. 3964–3979, 2021.
- D. Nielsen, P. Jaini, E. Hoogeboom, O. Winther, and M. Welling, “Survae flows: Surjections to bridge the gap between vaes and flows,” in NIPS 2020 (Virtual), 2020.
- P. M. Baggenstoss and S. Kay, “Nonlinear dimension reduction by pdf estimation,” IEEE Transactions on Signal Processing, 2022.
- P. M. Baggenstoss, “Beyond moments: Extending the maximum entropy principle to feature distribution constraints,” Entropy, vol. 20, no. 9, 2018. [Online]. Available: http://www.mdpi.com/1099-4300/20/9/650
- ——, “Maximum entropy PDF design using feature density constraints: Applications in signal processing,” IEEE Trans. Signal Processing, vol. 63, no. 11, Jun. 2015.
- S. M. Kay, A. H. Nuttall, and P. M. Baggenstoss, “Multidimensional probability density function approximations for detection, classification, and model order selection,” IEEE Transactions on Signal Processing, vol. 49, no. 10, pp. 2240–2252, Oct 2001.
- P. M. Baggenstoss, “Discriminative alignment of projected belief networks,” IEEE Signal Processing Letters, Sep 2021.
- ——, “Evaluating the RBM without integration using pdf projection,” in Proceedings of EUSIPCO 2017, Island of Kos, Greece, Aug 2017.
- ——, “Uniform manifold sampling (UMS): Sampling the maximum entropy pdf,” IEEE Transactions on Signal Processing, vol. 65, no. 9, pp. 2455–2470, May 2017.
- ——, “A neural network based on first principles,” in ICASSP 2020, Barcelona (virtual), Barcelona, Spain, Sep 2020.
- ——, “The class-specific classifier: Avoiding the curse of dimensionality (tutorial),” IEEE Aerospace and Electronic Systems Magazine, special Tutorial addendum, vol. 19, no. 1, pp. 37–52, January 2004.
- P. Baggenstoss, “PBN Toolkit,” {http://class-specific.com/pbntk}, accessed: 2022-02-28.