The Fuzzy-Pattern Tsetlin Machine (FPTM) makes Tsetlin Machines much smaller and faster to train and run, while matching or improving the accuracy of existing variants. That opens the door to experimenting with problems that were previously impractical, including text generation.
The key idea is to combine FPTM with Hyperdimensional Computing (HDC). HDC encodes variable-length character or token contexts into fixed-size hypervectors, making them a natural fit for Tsetlin Machines.
As a proof of concept, I trained a simple character-level language model that generates Shakespeare-style text one character at a time. Depending on the configuration, training takes anywhere from a few minutes to a few hours on a modern desktop CPU.
The repository includes a complete text generation training example.
The Fuzzy-Pattern Tsetlin Machine (FPTM) makes Tsetlin Machines much smaller and faster to train and run, while matching or improving the accuracy of existing variants. That opens the door to experimenting with problems that were previously impractical, including text generation.
The key idea is to combine FPTM with Hyperdimensional Computing (HDC). HDC encodes variable-length character or token contexts into fixed-size hypervectors, making them a natural fit for Tsetlin Machines.
As a proof of concept, I trained a simple character-level language model that generates Shakespeare-style text one character at a time. Depending on the configuration, training takes anywhere from a few minutes to a few hours on a modern desktop CPU.
The repository includes a complete text generation training example.