Tuning the Performance of a Computational Persistent Homology Package.

Area of Science:

• Computational topology
• Data analysis algorithms
• Scientific software engineering

Background:

• Persistent homology (PH) is a powerful technique for analyzing topological features in data.
• Existing PH software, like Eirene, can face performance limitations.
• Efficient computation is crucial for applying PH to large datasets.

Purpose of the Study:

• To improve the computational performance of the Eirene persistent homology software library.
• To identify and address performance bottlenecks within the Eirene package.
• To leverage multicore/manycore hardware for accelerating PH computations.

Main Methods:

• Utilized Julia's built-in profiling tools to pinpoint performance bottlenecks.
• Developed and implemented novel methods for performance optimization.
• Applied parallelization techniques to computationally intensive functions within Eirene.

Main Results:

• Successfully identified key areas for performance improvement in Eirene.
• Implemented parallelized functions demonstrating significant speedups.
• Empirical results confirm substantial performance gains in the optimized Eirene package.

Conclusions:

• The performance enhancements significantly improve Eirene's efficiency.
• Parallelization strategies are effective for accelerating persistent homology computations.
• Optimized Eirene offers a more performant tool for topological data analysis.