GSGP-hardware: instantaneous symbolic regression with an FPGA implementation of geometric semantic genetic programming
Yazmin Maldonado,Ruben Salas,Joel A. Quevedo,Rogelio Valdez,Leonardo Trujillo
DOI: https://doi.org/10.1007/s10710-024-09491-5
2024-06-26
Genetic Programming and Evolvable Machines
Abstract:Geometric Semantic Genetic Programming (GSGP) proposed an important enhancement to GP-based learning, incorporating search operators that operate directly on the semantics of the parents with bounded effects on the semantics of the offspring. This approach posed any symbolic regression fitness landscape as a unimodal function, allowing for more directed search. Moreover, it became evident that the search could be implemented in a much more efficient manner, that does not require the execution, evaluation or manipulation of variable length syntactic models. Hence, efficient implementations of this algorithm have been developed using both CPU and GPU processing. However, current implementations are still ill-suited for real-time learning, or learning on devices with limited resources, scenarios that are becoming more prevalent with the continued development of the Internet-of-Things and the increased need for efficient and distributed learning on the Edge. This paper presents GSGP-Hardware, a fully pipelined and parallel design of GSGP developed fully using VHDL, for implementation on FPGA devices. Using Vivado AMD-Xilinx for synthesis and simulation, GSGP-Hardware achieves an approximate improvement in efficiency, in terms of run time and Gpops/s, of three and four orders of magnitude, respectively, compared with the state-of-the-art GPU implementation. This is a performance increase that has not been achieved by other FPGA-based implementations of genetic programming. This is possible due to the manner in which GSGP evolves a model, and competitive accuracy is achieved by incorporating simple but powerful enhancements to the original GSGP algorithm. GSGP-Hardware allows for instantaneous symbolic regression, opening up new application domains for this powerful variant of genetic programming.
computer science, artificial intelligence, theory & methods