Dynamic partial reconfiguration implementation of the SVM/KNN multi-classifier on FPGA for bioinformatics...

Area of Science:

• Computational Biology
• Computer Engineering
• Bioinformatics

Background:

• Bioinformatics data is highly dimensional, demanding significant computational power.
• Conventional computing methods struggle with these demands, necessitating high-performance solutions like GPUs and FPGAs.
• Field Programmable Gate Arrays (FPGAs) offer efficiency and performance, enhanced by dynamic partial reconfiguration (DPR) for flexibility.

Purpose of the Study:

• To propose a dynamic multi-classifier architecture leveraging FPGAs and DPR for processing bioinformatics data.
• To enable the simultaneous application of diverse classification algorithms for more reliable and consensus-driven bioinformatics analysis.
• To address the time-consuming nature of applying multiple classifiers on conventional PCs.

Main Methods:

• Implementation of two common classifiers, Support Vector Machines (SVMs) and K-nearest neighbor (KNN), using FPGA dynamic partial reconfiguration (DPR).
• Integration of these DPR-implemented classifiers into a single multi-classifier FPGA architecture.
• The architecture allows specific FPGA regions to function as either an SVM or KNN classifier dynamically.

Main Results:

• The proposed multi-classifier DPR implementation achieved at least an 8x reduction in reconfiguration time compared to single, non-DPR classifier implementations.
• The architecture occupied less space and hardware resources than implementing both classifiers separately.
• Demonstrated the potential for extension into an ensemble classifier system.

Conclusions:

• The dynamic multi-classifier FPGA architecture effectively processes bioinformatics data with improved efficiency.
• DPR on FPGAs provides a flexible and resource-efficient platform for complex computational biology tasks.
• This architecture offers a promising approach for accelerating bioinformatics analysis and enabling ensemble methods.