the minimum euclidean norm
play

The minimum Euclidean norm point in a polytope: Wolfe's method is - PowerPoint PPT Presentation

The minimum Euclidean norm point in a polytope: Wolfe's method is exponential Luis Rademacher, UC Davis Simons Institute Join work with Jess de Loera, Jamie Haddock Minimum norm point in a polytope Given 1 , ,


  1. The minimum Euclidean norm point in a polytope: Wolfe's method is exponential Luis Rademacher, UC Davis Simons Institute Join work with JesΓΊs de Loera, Jamie Haddock

  2. Minimum norm point in a polytope β€’ Given π‘ž 1 , … , π‘ž π‘œ ∈ 𝑆 𝑒 , determine argmin π‘¦βˆˆπ‘„ 𝑦 for 𝑄 = conv(π‘ž 1 , … , π‘ž π‘œ ) . β€’ A convex quadratic program.

  3. Projection onto a simplex: not so easy β€’ Only polynomial time algorithms I know are general purpose convex programming algorithms like the ellipsoid method. β€’ No strongly polynomial time algorithm known.

  4. Wolfe’s method (Wolfe ’74, Lawson Hanson β€˜74) β€’ A combinatorial algorithm to find the minimum norm point in a polytope. β€’ Not the same as Frank-Wolfe

  5. Motivation β€’ In machine learning: β€’ Fujishige-Wolfe algorithm β€’ one of the most practical algorithms for submodular function minimization . β€’ Wolfe’s method is a subroutine in it β€’ Optimal loading of recursive neural networks [CDKSV β€˜95] β€’ Non-negative least squares

  6. This talk: β€’ Complexity of Wolfe’s method β€’ Relationship between Linear Programming (LP) and minimum norm point problem.

  7. Related work β€’ [Lawson, Hanson β€˜74] An algorithm very similar to Wolfe’s. β€’ [Fujishige, Hayashi, Isotani ’06] Polynomial time reduction from LP to minimum norm point in a polytope. β€’ [Chakrabarty, Jain, Kothari β€˜14] [Lacoste-Julien, Jaggi β€˜15] Rates of convergence of Wolfe’s method.

  8. Our results β€’ LP reduces in strongly polynomial time to β€œminimum norm point in a simplex ”. β€’ A (strongly) polynomial time algorithm for minimum norm point in a simplex would give a (strongly) polynomial time algorithm for general LPs. β€’ Step 1: LP reduc es to β€œmembership in a V - polytope”. β€’ Step 2: β€œMembership in a V - polytope” reduces to β€œdistance to a simplex”.

  9. Our results β€’ Wolfe’s method takes exponential time in the worst case. β€’ We construct explicit sets of points in every dimension. β€’ Similar in spirit to Klee-Minty cubes for LP.

  10. Idea of Wolfe’s method β€’ Given points 𝑄 βŠ† 𝑆 𝑒 , maintain the following invariant : A subset 𝑇 βŠ† 𝑄 whose vertices determine a simplex, and the minimum norm point, 𝑦 , in the simplex. β€’ Start with any point in 𝑄 as 𝑇 and 𝑦 . β€’ Alternate between the following two steps: β€’ Find point π‘ž ∈ 𝑄 such that 𝑇 ← π‘‘π‘π‘œπ‘€ 𝑇 βˆͺ π‘ž contains a better point (entering rule). Add π‘ž to 𝑇 . β€’ Let 𝑦 β€œ follow gravity ” (towards 0) within π‘‘π‘π‘œπ‘€ 𝑇 while dimension of current face decreases. Let 𝑇 be the (vertices of the) current face of 𝑦 .

  11. Inefficiency of Wolfe’s method with β€œ minnorm ” entering rule β€’ A point, 𝑏 , enters current set 𝑇 , leaves, and then re-enters. β€’ 𝑏 < π‘ž < π‘Ÿ < 𝑠 < |𝑑| S . a ap apq pq pqr qr qrs rs rsa

  12. Exponential lower bound β€’ Replace point 𝑏 and 1 st coordinate by subspace and a set of points in it, constructed recursively. β€’ Seq 𝑇 : 𝑄 𝑒 βˆ’ 2 β†’ β‹― β†’ 𝑠 𝑒 𝑑 𝑒 β†’ 𝑄 𝑒 βˆ’ 2 𝑠 𝑒 𝑑 𝑒

  13. Open questions β€’ Find exponential lower bound for other β€œentering rules.” ( linopt) β€’ For submodular function minimization, polytopes are so- called β€œbase polytopes.” β€’ Our exponential example is not of that kind. β€’ Could Wolfe’s method be faster on base polytopes? β€’ Smoothed analysis of Wolfe’s method.

Recommend


More recommend