How randomness tames vast network problems
ETH Zurich computer scientists led by Rasmus Kyng have developed a new solver for Laplacian equations that uses randomized Approximate Cholesky factorization, achieving nearly linear running time and up to five times faster performance than existing software. The prototype, published in the SIAM Journal on Scientific Computing, reliably solves large network problems—including those where previous solvers fail—by sampling a small set of connections to reduce computational cost while maintaining accuracy. This work demonstrates how theoretical breakthroughs can be translated into practical, hig…
ETH Zurich computer scientists led by Rasmus Kyng have developed a new solver for Laplacian equations that uses randomized Approximate Cholesky factorization, achieving nearly linear running time and up to five times faster performance than existing software. The prototype, published in the SIAM Journal on Scientific Computing, reliably solves large network problems—including those where previous solvers fail—by sampling a small set of connections to reduce computational cost while maintaining accuracy. This work demonstrates how theoretical breakthroughs can be translated into practical, hig…
The solver’s efficient handling of Laplacian equations, which also underpin machine learning models, supports the development of AI systems in Switzerland by providing faster, scalable tools for network‑based computations.
Original source record. Open the original record to verify the underlying announcement.