Tamás Bitai

This is an analysis of Bayesian spam filtering from the viewpoint of Bayesian epistemology. I review the Bayesian spam filtering algorithm, point out where it deviates from the principles of Bayesian epistemology, and present test results showing that the “principled” Bayesian spam filtering is also more effective. Finally, I compare Bayesian filtering to neural networks, with mixed results.

Bitai, T. (2021). Spamszűrés bayesiánus episztemológiával. In Dr. Koncz, I., Szova, I. (eds.), A PEME XXII. PhD-konferenciájának előadásai (pp. 62–67). PEME. https://peme.hu/2021/12/31/xxii-phd-konferencia/

Postprint

GitHub

The washing out of the priors is an argument for Bayesian confirmation theory, which claims that Bayesian agents converge to a consensus regardless of the differences in their initial beliefs. Here we review the textbook formalizations of the argument, and conclude that they are not committed to any interpretation of objective probability, but they only prove almost sure convergence (in the sense of objective probability). We propose a new formalization, additionally assuming the frequentist interpretation of objective probability, that proves actual, unconditional convergence. We consider this a more convincing argument for Bayesianism to those accepting the frequentist interpretation of objective probability.

Bitai, T. (2019). A bayesiánus kezdeti valószínűségek kimosódása frekventista objektív valószínűségre alapozva. In Ruszkai, Sz. É. et al. (eds.), Ütközéspontok V. – A Doktoranduszok Országos Szövetsége Filozófiatudományi Osztálya konferenciájának kötete (pp. 33–42). JATEPress. https://www.academia.edu/41647821

Postprint

Ruzsa is an educational software for introductory logic courses. It’s an analytic tableau proof editor for Tarski’s World. Thus, Ruzsa is an alternative to Fitch, but instead of Fitch’s natural deduction proof method, Ruzsa uses the more beginner-friendly method of analytic tableaux.

The idea of Ruzsa was András Máté’s.

Open

Documentation