The sbi library offers a wide range of utilities for neural inference methods, yet progress on the ABC side has been stagnant. Utilizing statistical distances across multiple observations in ABC has considerably improved ABC methods [1].

To add competitive ABC baselines, I would suggest to add sample-based approximations of the maximum-mean discrepancy [2] and the 2-Wasserstein distance [3].

If you deem the extension relevant to the sbi package, I would be happy implementing these changes.

[1] Bernton, E., Jacob, P. E., Gerber, M., & Robert, C. P. (2019). Approximate Bayesian computation with the Wasserstein distance. Journal of the Royal Statistical Society Series B: Statistical Methodology.

[2] Gretton, A., Borgwardt, K. M., Rasch, M. J., Schölkopf, B., & Smola, A. (2012). A kernel two-sample test. The Journal of Machine Learning Research.

[3] Peyré, G., & Cuturi, M. (2019). Computational optimal transport: With applications to data science. Foundations and Trends® in Machine Learning.