Reproduce “Optimization on Pareto sets: On a theory of multi-objective optimization” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Abhishek Roy, Geelon So, Yi-An Ma

Researchers from the University of California, San Diego introduce a theoretical and algorithmic framework for optimizing a preference function over Pareto sets by reformulating the problem as smooth optimization on the standard simplex. Their Pareto Majorization-Minimization (PMM) algorithm converges to an approximate preference stationary point with an $O(\varepsilon_0^{-2})$ rate.

Key insights

  • The Pareto manifold, defined by the set of optimal solutions to scalarized problems, is diffeomorphic to the standard simplex, enabling a transformation of the intractable original problem into a smooth, convex-constrained optimization.
  • Defining a non-trivial, necessary, and locally decidable first-order stationarity condition for this problem fundamentally requires the use of second-order information (Hessians).
  • A family of explicit quadratic majorizing surrogate functions can be constructed, allowing for an iterative optimization algorithm despite the implicit nature of the objective function.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.