Tsetlin machine (TM) is a logic-based machine learning approach with the crucial advantages of being transparent and hardware-friendly. While TMs match or surpass deep learning accuracy for an increasing number of applications, large clause pools tend to produce clauses with many literals (long clauses). As such, they become less interpretable. Further, longer clauses increase the switching activity of the clause logic in hardware, consuming more power. This paper introduces a novel variant of TM learning - Clause Size Constrained TMs (CSC-TMs) - where one can set a soft constraint on the clause size. As soon as a clause includes more literals than the constraint allows, it starts expelling literals. Accordingly, oversized clauses only appear transiently. To evaluate CSC-TM, we conduct classification, clustering, and regression experiments on tabular data, natural language text, images, and board games. Our results show that CSC-TM maintains accuracy with up to 80 times fewer literals. Indeed, the accuracy increases with shorter clauses for TREC, IMDb, and BBC Sports. After the accuracy peaks, it drops gracefully as the clause size approaches a single literal. We finally analyze CSC-TM power consumption and derive new convergence properties.

The Tsetlin Machine (TM) [1] employs groups of Tsetlin Automata (TAs) [2], which operate on binary data using propositional logic. Via a game-theoretic collaboration scheme, the TAs self-organize to capture the distinct patterns in the data. In brief, each group of TAs builds a conjunctive clause that captures a specific pattern. The dynamics of the collaboration involves three interacting mechanisms. High pattern recall is enforced by a resource allocation mechanism that diversifies clause construction. Simultaneously, a mechanism that forces the clauses to capture frequent patterns combats overfitting. Finally, without compromising high pattern frequency, the discrimination power of the clauses is optimized by injecting discriminative features. TMs provide two main advantages: transparent inference and learning combined with hardware-near building blocks.