arXiv research

We prove for any positive integer $n$ there exist boundary-sum irreducible ${\mathbb Z}_n$-corks with Stein structure. Here `boundary-sum irreducible' means the manifold is indecomposable with respect to boundary-sum. We also verify that some of the finite order corks admit hyperbolic boundary by HIKMOT.

Summing over 3-manifolds using TQFT partition functions.

A thesis submitted for the degree of Doctor of Philosophy of The Australian National University. In this work we introduce several new optimisation methods for problems in machine learning. Our algorithms broadly fall into two categories: optimisation of finite sums and of graph structured objectives. The finite sum pr…

Novel analysis of EFP for finite-sum problems in neural networks.

We study the conditions under which one is able to efficiently apply variance-reduction and acceleration schemes on finite sum optimization problems. First, we show that, perhaps surprisingly, the finite sum structure by itself, is not sufficient for obtaining a complexity bound of $\tilde{\cO}((n+L/μ)\ln(1/ε))$ for $L…

New method reduces complexity of minimizing convex finite sums without needing individual function indices.

Stochastic optimization algorithms with variance reduction have proven successful for minimizing large finite sums of functions. Unfortunately, these techniques are unable to deal with stochastic perturbations of input data, induced for example by data augmentation. In such cases, the objective is no longer a finite su…

Lower bounds for higher-order methods in non-convex optimization.

Characterizes compact complex surfaces with finite homotopy rank-sum.

Characterizes Stein surfaces with finite homotopy rank-sum.

We study Farrell Nil-groups associated to a finite order automorphism of a ring $R$. We show that any such Farrell Nil-group is either trivial, or infinitely generated (as an abelian group). Building on this first result, we then show that any finite group that occurs in such a Farrell Nil-group occurs with infinite mu…

Genus 3 Heegaard groups of lens space connected sums are finitely generated.

Finite-sum optimization problems are ubiquitous in machine learning, and are commonly solved using first-order methods which rely on gradient computations. Recently, there has been growing interest in \emph{second-order} methods, which rely on both gradients and Hessians. In principle, second-order methods can require …

For most positive integer pairs $(a,b)$, the topological space $#a{\mathbb C \mathbb P}^2#b{\bar{\mathbb C \mathbb P^2}}$ is shown to admit infinitely many inequivalent smooth structures which dissolve upon performing a single connected sum with $S^2\times S^2$. This is then used to construct infinitely many non-equiva…

New methods optimize sums of bivariate functions on finite domains.

Two new Frank-Wolfe algorithms improve convergence for constrained optimization.

Smooth finite-sum optimization has been widely studied in both convex and nonconvex settings. However, existing lower bounds for finite-sum optimization are mostly limited to the setting where each component function is (strongly) convex, while the lower bounds for nonconvex finite-sum optimization remain largely unsol…

New lower bounds for gradient methods in strongly convex finite-sum optimization.

New framework guarantees convergence of multi-step MAML.

Paper establishes lower bounds for finite-sum optimization problems using novel construction methods.

We describe a novel optimization method for finite sums (such as empirical risk minimization problems) building on the recently introduced SAGA method. Our method achieves an accelerated convergence rate on strongly convex smooth problems. Our method has only one parameter (a step size), and is radically simpler than o…

We show that the finiteness length of an $S$-arithmetic subgroup $Γ$ in a noncommutative isotropic absolutely almost simple group $G$ over a global function field is one less than the sum of the local ranks of $G$ taken over the places in $S$. This determines the finiteness properties for arithmetic subgroups in isotro…

The popular cubic smoothing spline estimate of a regression function arises as the minimizer of the penalized sum of squares $\sum_j(Y_j - μ(t_j))^2 + λ\int_a^b [μ"(t)]^2 dt$, where the data are $t_j,Y_j$, $j=1,..., n$. The minimization is taken over an infinite-dimensional function space, the space of all functions wi…

Crane and Frenkel proposed a state sum invariant for triangulated 4-manifolds.They defined and used new algebraic structures called Hopf categories for their construction. Crane and Yetter studied Hopf categories and gave some examples using group cocycles that are associated to the Drinfeld double of a finite group. I…

We study Frank-Wolfe methods for nonconvex stochastic and finite-sum optimization problems. Frank-Wolfe methods (in the convex case) have gained tremendous recent interest in machine learning and optimization communities due to their projection-free property and their ability to exploit structured constraints. However,…

Study on convergence of Langevin dynamics for zero-sum games in probability distributions.

On a smooth closed oriented $4$-manifold $M$ with a smooth action of a finite group $G$ on a Spin$^c$ structure, $G$-monopole invariant is defined by "counting" $G$-invariant solutions of Seiberg-Witten equations for any $G$-invariant Riemannian metric on $M$. We compute $G$-monopole invariants on some $G$-manifolds. F…

SVRG reduces gradient evaluations for policy evaluation in reinforcement learning.

Spatial graphs are decomposed into planar forests and braids.

We provide an explicit section for a mapping class group sequence.

Unified framework for stability and generalization of Push-Sum in decentralized learning over directed graphs.

On a smooth closed oriented $4$-manifold $M$ with a smooth action by a compact Lie group $G$, we define a $G$-monopole class as an element of $H^2(M;\Bbb Z)$ which is the first Chern class of a $G$-equivariant Spin$^c$ structure which has a solution of the Seiberg-Witten equations for any $G$-invariant Riemannian metri…

Defines a universal state sum construction for various TQFTs.

We propose a fast proximal Newton-type algorithm for minimizing regularized finite sums that returns an $ε$-suboptimal point in $\tilde{\mathcal{O}}(d(n + \sqrt{κd})\log(\frac{1}ε))$ FLOPS, where $n$ is number of samples, $d$ is feature dimension, and $κ$ is the condition number. As long as $n > d$, the proposed method…

Study on Nesterov's method in stochastic settings, revealing divergence under certain conditions.

DESTRESS optimizes decentralized nonconvex optimization with optimal IFO complexity and efficient communication.

A new method, Residual-Permuted Sums, improves confidence region construction for linear regression models.

Paper proposes a faster SPIDER-EM variant for large-scale nonconvex optimization.

New insights into tail behavior of heavy-tailed random vectors and processes.

The height function of various surfaces decomposes into finite sums of scaled and translated versions of itself.

Proves a formula for a special invariant of 4-manifolds.

This paper presents a lower bound for optimizing a finite sum of $n$ functions, where each function is $L$-smooth and the sum is $μ$-strongly convex. We show that no algorithm can reach an error $ε$ in minimizing all functions from this class in fewer than $Ω(n + \sqrt{n(κ-1)}\log(1/ε))$ iterations, where $κ=L/μ$ is a …

We give a short proof that if a non-trivial band sum of two knots results in a tight fibered knot, then the band sum is a connected sum. In particular, this means that any prime knot obtained by a non-trivial band sum is not tight fibered. Since a positive L-space knot is tight fibered, a non-trivial band sum never yie…

We prove that the expectation value of the index function i(x) over a probability space of injective function f on any finite simple graph G=(V,E) is equal to the curvature K(x) at the vertex x. This result complements and links Gauss-Bonnet sum K(x) = chi(G) and Poincare-Hopf sum i(x) = chi(G) which both hold for arbi…

SVRN accelerates Newton methods by reducing variance and improving performance.

The study classifies strongly irreducible Heegaard splittings in hyperbolic 3-manifolds.

New method solves root-finding problems with faster convergence.

Let H_g denote the closed 3-manifold obtained as the connected sum of g copies of S^2 times S^1, with free fundamental group of rank g. We prove that, for a finite group G acting on H_g which induces a faithful action on the fundamental group, there is an upper bound for the order of G which is quadratic in g, but that…