It is proved that every deterministic first-order method requires $\Omega(\ell\Delta\kappa/\epsilon^2)$ oracle queries in the worst case to find $x$ satisfying $\Phi(0)-\inf_x\Phi(x)$ and that the linear dependence on $\kappa$ is unavoidable for deterministic first-order methods.
Abstract
We study the deterministic first-order oracle complexity of finding stationary points of the value function in smooth nonconvex-Polyak-{\L}ojasiewicz (NC-P{\L}) minimax optimization. We assume that the objective is jointly $\ell$-smooth and satisfies the $\mu$-P{\L} condition in the dual variable, and that its value function $\Phi(x):=\max_y f(x;y)$ satisfies $\Phi(0)-\inf_x\Phi(x)\leq\Delta$. When $\kappa:=\ell/\mu\gtrsim 1$ and $0<\epsilon^2\lesssim\ell\Delta$, we prove that every deterministic first-order method requires $\Omega(\ell\Delta\kappa/\epsilon^2)$ oracle queries in the worst case to find $x$ satisfying $\|\nabla\Phi(x)\|\leq\epsilon$. This rate matches the known upper bound in its dependence on $(\ell,\Delta,\kappa,\epsilon)$ [Yang et al., 2022] and shows that the linear dependence on $\kappa$ is unavoidable for deterministic first-order methods.
It is proved that every deterministic first-order algorithm requires a first-order oracle that returns both the function value and the full subdifferential at every query point, and establishes the optimal deterministic oracle complexity.
We study the simultaneous approximation of constant-degree polynomials over convex sets. For any family of $m$ degree-$d$ polynomials and any convex set ${H} \subseteq \mathbb{R}_{\ge0}^n$, we construct an $\epsilon$-Cover of the joint value set $\{(f_1(x), \dots, f_m(x)) : x \in {H}\}$ in the $\ell_\infty$-norm. This cover is of size $n^{O(\log(mn)/\epsilon^2)}$, provided the polynomials have constant range over the smallest $\ell_1$-ball inscribing ${H}$. Our approach extends classical net-based sparsifications for linear functions (e.g., Lipton, Markakis, and Mehta [2003]) to arbitrary families of constant-degree polynomials over general convex sets. We use a two-step scheme: first, we construct a quasi-polynomial pre-cover of the family on the smallest $\ell_1$-ball containing ${H}$ by using a concentration argument and leveraging a connection between Bernstein approximation and multinomial distributions; we then compress the pre-cover to ${H}$ by using a recursive degree reduction and feasibility programs anchored at points of the pre-cover. The existence of these covers immediately yields a unified framework for Quasi-Polynomial Time Approximation Schemes (QPTAS) across a wide range of a problems, including fixed-degree polynomial minimization over polyhedral sets, Constraint Satisfaction Problems (CSPs), Free Games, variational inequalities with polynomial operators (which implies guarantees for local Nash equilibria in polynomial games), and additive approximation for normalized densest $k$-subhypergraph on $O(1)$-uniform hypergraphs.
Martino Bernasconi, Matteo Castiglioni, Andrea Celli et al.· arXiv.org· 1 citation
We study a deterministic family of sharpness-aware minimization methods for smooth nonconvex functions. The perturbation is $$ y_k=x_k+\rho\, \frac{\nabla f(x_k)}{\norm{\nabla f(x_k)}^\alpha}, \qquad 0\leq\alpha\leq 1, $$ so that its effective radius is $\rho\norm{\nabla f(x_k)}^{1-\alpha}$. For $0<\alpha\leq1$, we give an explicit complexity bound above the stationarity level $(L\rho)^{1/\alpha}$. A one-dimensional quadratic example reaches this level exactly, showing that the bound describes a real limitation of the constant-parameter rule. The unnormalized case $\alpha=0$ is treated separately and requires $L\rho<1$. We then introduce a clipped rule which agrees with the constant-$\rho$ rule away from stationary points and becomes proportional to the gradient near them. The clipped method has $\norm{\nabla f(x_k)}\to0$ and the usual $O(T^{-1/2})$ stationarity bound. Numerical tests on a quadratic function, the Rosenbrock function, and a five-dimensional nonconvex function illustrate the stationarity floor of the unclipped rule and the effect of clipping.
For every $2<p<\infty$ and every integer $r\ge2$, we construct a finite set $T\subseteq S_{L^p[0,1]}$ such that $|T|\le2^{Cr^2}$, $\gamma_2(T)\le Cr$, and $\gamma_2(\conv T)\ge c r^{3/2-1/p}$. Consequently, for every fixed $p>2$, both estimates in Talagrand's Research Problem~2.11.3 fail in each of the spaces $L^p[0,1]$ and $\ell_p$. The lower bound follows from a multilevel product principle applied to a scale-separated product of Euclidean simplices. The same construction gives the sharp convexification profile of $\ell_p^r(\ell_2^{2^{128r}})$ for $2<p\le\infty$ and, by finite representability, quantitative failure in every infinite-dimensional Banach space with cotype index greater than $2$.
This work proves the first near-optimal lower bound for arbitrary adaptive randomized algorithms throughout both accuracy regimes of exact value Lipschitz convex optimization, and develops a posterior mean energy method for adaptive exact max observations.
Haihan Zhang, Chen-Heng Zhang, Zhiquan Qi et al.· 1 citation
Let $X\subset\mathbb{R}$ be finite and let $\gamma(t)=(t,f(t))$, where $f$ is strictly convex. We show that \[ J_3(\gamma(X)) =\#\{(x_1,\ldots,x_6)\in X^6:\sum_{i=1}^3\gamma(x_i)=\sum_{i=4}^6\gamma(x_i)\} \ll_{\epsilon}|X|^{3+\epsilon}. \] When specialized to the parabola, our result implies near-optimal estimates for the number of solutions to the diameter-free quadratic Vinogradov system. As a second application, we settle a conjecture from Krishnapur-Kurlberg-Wigman and Bombieri-Bourgain concerning lattice points on dilates of the unit circle. As a third application, we prove that $|A-A|\gg_\epsilon|A|^{5/3-\epsilon}$ and $|A+A|\gg_\epsilon|A|^{8/5-\epsilon}$ for any finite convex sequence $A\subset \mathbb{R}$.
Adam Cushman, C. Demeter, Shukun Wu· 1 citation· ⚡1
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
