A new arXiv paper asks whether algorithms can be treated as models that are learned and evolved, rather than as programs that must be redesigned from scratch. To explore this, the authors introduce Algorithm Modelization, a paradigm that turns iterative algorithms into trainable models.
The core idea is to move away from hand-crafting optimization algorithms and instead allow them to adapt through training. While the abstract does not provide implementation details, it frames the approach as a fundamental shift in how algorithms are conceived and developed.
The paper is a single source, so there are no conflicting findings to compare. It remains an early-stage proposal, and the abstract does not yet describe experimental results or practical applications.