Comments on: Query Learning Algorithm for Symbolic Weighted Finite Automata https://icgi2020.lis-lab.fr August 23-27, 2021 Tue, 31 Aug 2021 11:14:06 +0000 hourly 1 https://wordpress.org/?v=7.1 By: Kaito Suzuki https://icgi2020.lis-lab.fr/query-learning-algorithm-for-symbolic-weighted-finite-automata/#comment-26 Mon, 23 Aug 2021 13:53:05 +0000 https://icgi2020.lis-lab.fr/?page_id=517#comment-26 Thank you for your question. In principle, SWFAs can be used for any applications of WFAs, such as extraction from RNN, data compression, NLP, etc. If the alphabet size is huge or infinite, SWFAs have the advantage in the sense of representation size. I think the example of extraction from RNN is particularly interesting since many RNNs have huge discrete (because of multi-dimensional) or continuous input.

However, our algorithm requires some additional assumptions for polynomial-time learning, including the existence of MAT-learner for the function class G and linear-closedness of G. This limits the range of applications. Our paper includes some discussions to show examples of G, but the applicable example of applications is still unclear.

]]> By: Jane Chandlee https://icgi2020.lis-lab.fr/query-learning-algorithm-for-symbolic-weighted-finite-automata/#comment-14 Sun, 22 Aug 2021 14:46:01 +0000 https://icgi2020.lis-lab.fr/?page_id=517#comment-14 Thank you for the talk – it was very clear! I was wondering what kind of domains or applications are best suited to SWFAs. Are there examples in which another representation is possible, but the SWFA representation is more efficient?

]]>