Searcharxiv⌕ Search

arXiv subjects

Nicholas J. A. Harvey

Publications and source records attributed to Nicholas J. A. Harvey.

At least 19 recordsLinked to original sources
arXiv2026

A second order regret bound for NormalHedge

We consider the problem of prediction with expert advice for ``easy'' sequences. We show that a variant of NormalHedge enjoys a second-order $ε$-quantile regret bound of $O\big(\sqrt{V_T \log(V_T/ε)}\big) $ when $V_T > \log N$, where $V_T$ is the cumulative second moment of instantaneous per-expert regret averaged with respect to a natural distribution determined by the algorithm. The algorithm is motivated by a continuous time limit using Stochastic Differential Equations. The discrete time analysis uses self-concordance techniques.

cs.LG↗
arXiv2022

Continuous Prediction with Experts' Advice

Prediction with experts' advice is one of the most fundamental problems in online learning and captures many of its technical challenges. A recent line of work has looked at online learning through the lens of differential equations and continuous-time analysis. This viewpoint has yielded optimal results for several problems in online learning. In this paper, we employ continuous-time stochastic calculus in order to study the discrete-time experts' problem. We use these tools to design a continuous-time, parameter-free algorithm with improved guarantees for the quantile regret. We then develop an analogous discrete-time algorithm with a very similar analysis and identical quantile regret bounds. Finally, we design an anytime continuous-time algorithm with regret matching the optimal fixed-time rate when the gains are independent Brownian Motions; in many settings, this is the most difficult case. This gives some evidence that, even with adversarial gains, the optimal anytime and fixed-time regrets may coincide.

cs.LG↗
arXiv2022

Efficient and Optimal Fixed-Time Regret with Two Experts

Prediction with expert advice is a foundational problem in online learning. In instances with $T$ rounds and $n$ experts, the classical Multiplicative Weights Update method suffers at most $\sqrt{(T/2)\ln n}$ regret when $T$ is known beforehand. Moreover, this is asymptotically optimal when both $T$ and $n$ grow to infinity. However, when the number of experts $n$ is small/fixed, algorithms with better regret guarantees exist. Cover showed in 1967 a dynamic programming algorithm for the two-experts problem restricted to $\{0,1\}$ costs that suffers at most $\sqrt{T/2π} + O(1)$ regret with $O(T^2)$ pre-processing time. In this work, we propose an optimal algorithm for prediction with two experts' advice that works even for costs in $[0,1]$ and with $O(1)$ processing time per turn. Our algorithm builds up on recent work on the experts problem based on techniques and tools from stochastic calculus.

cs.LG↗
arXiv2021

Online mirror descent and dual averaging: keeping pace in the dynamic case

Online mirror descent (OMD) and dual averaging (DA) -- two fundamental algorithms for online convex optimization -- are known to have very similar (and sometimes identical) performance guarantees when used with a fixed learning rate. Under dynamic learning rates, however, OMD is provably inferior to DA and suffers a linear regret, even in common settings such as prediction with expert advice. We modify the OMD algorithm through a simple technique that we call stabilization. We give essentially the same abstract regret bound for OMD with stabilization and for DA by modifying the classical OMD convergence analysis in a careful and modular way that allows for straightforward and flexible proofs. Simple corollaries of these bounds show that OMD with stabilization and DA enjoy the same performance guarantees in many applications -- even under dynamic learning rates. We also shed light on the similarities between OMD and DA and show simple conditions under which stabilized-OMD and DA generate the same iterates.

cs.LG↗
arXiv2021

Optimal anytime regret with two experts

We consider the classical problem of prediction with expert advice. In the fixed-time setting, where the time horizon is known in advance, algorithms that achieve the optimal regret are known when there are two, three, or four experts or when the number of experts is large. Much less is known about the problem in the anytime setting, where the time horizon is not known in advance. No minimax optimal algorithm was previously known in the anytime setting, regardless of the number of experts. Even for the case of two experts, Luo and Schapire have left open the problem of determining the optimal algorithm. We design the first minimax optimal algorithm for minimizing regret in the anytime setting. We consider the case of two experts, and prove that the optimal regret is $γ\sqrt{t} / 2$ at all time steps $t$, where $γ$ is a natural constant that arose 35 years ago in studying fundamental properties of Brownian motion. The algorithm is designed by considering a continuous analogue of the regret problem, which is solved using ideas from stochastic calculus.

cs.LG↗
arXiv2020

Regret Bounds without Lipschitz Continuity: Online Learning with Relative-Lipschitz Losses

In online convex optimization (OCO), Lipschitz continuity of the functions is commonly assumed in order to obtain sublinear regret. Moreover, many algorithms have only logarithmic regret when these functions are also strongly convex. Recently, researchers from convex optimization proposed the notions of "relative Lipschitz continuity" and "relative strong convexity". Both of the notions are generalizations of their classical counterparts. It has been shown that subgradient methods in the relative setting have performance analogous to their performance in the classical setting. In this work, we consider OCO for relative Lipschitz and relative strongly convex functions. We extend the known regret bounds for classical OCO algorithms to the relative setting. Specifically, we show regret bounds for the follow the regularized leader algorithms and a variant of online mirror descent. Due to the generality of these methods, these results yield regret bounds for a wide variety of OCO algorithms. Furthermore, we further extend the results to algorithms with extra regularization such as regularized dual averaging.

cs.LG↗
arXiv2019

Simple and optimal high-probability bounds for strongly-convex stochastic gradient descent

We consider stochastic gradient descent algorithms for minimizing a non-smooth, strongly-convex function. Several forms of this algorithm, including suffix averaging, are known to achieve the optimal $O(1/T)$ convergence rate in expectation. We consider a simple, non-uniform averaging strategy of Lacoste-Julien et al. (2011) and prove that it achieves the optimal $O(1/T)$ convergence rate with high probability. Our proof uses a recently developed generalization of Freedman's inequality. Finally, we compare several of these algorithms experimentally and show that this non-uniform averaging strategy outperforms many standard techniques, and with smaller variance.

cs.LG↗
arXiv2018

Tight Analyses for Non-Smooth Stochastic Gradient Descent

Consider the problem of minimizing functions that are Lipschitz and strongly convex, but not necessarily differentiable. We prove that after $T$ steps of stochastic gradient descent, the error of the final iterate is $O(\log(T)/T)$ with high probability. We also construct a function from this class for which the error of the final iterate of deterministic gradient descent is $Ω(\log(T)/T)$. This shows that the upper bound is tight and that, in this setting, the last iterate of stochastic gradient descent has the same general error rate (with high probability) as deterministic gradient descent. This resolves both open questions posed by Shamir (2012). An intermediate step of our analysis proves that the suffix averaging method achieves error $O(1/T)$ with high probability, which is optimal (for any first-order optimization method). This improves results of Rakhlin (2012) and Hazan and Kale (2014), both of which achieved error $O(1/T)$, but only in expectation, and achieved a high probability error bound of $O(\log \log(T)/T)$, which is suboptimal. We prove analogous results for functions that are Lipschitz and convex, but not necessarily strongly convex or differentiable. After $T$ steps of stochastic gradient descent, the error of the final iterate is $O(\log(T)/\sqrt{T})$ with high probability, and there exists a function for which the error of the final iterate of deterministic gradient descent is $Ω(\log(T)/\sqrt{T})$.

cs.LG↗
arXiv2018

Greedy and Local Ratio Algorithms in the MapReduce Model

MapReduce has become the de facto standard model for designing distributed algorithms to process big data on a cluster. There has been considerable research on designing efficient MapReduce algorithms for clustering, graph optimization, and submodular optimization problems. We develop new techniques for designing greedy and local ratio algorithms in this setting. Our randomized local ratio technique gives $2$-approximations for weighted vertex cover and weighted matching, and an $f$-approximation for weighted set cover, all in a constant number of MapReduce rounds. Our randomized greedy technique gives algorithms for maximal independent set, maximal clique, and a $(1+ε)\ln Δ$-approximation for weighted set cover. We also give greedy algorithms for vertex colouring with $(1+o(1))Δ$ colours and edge colouring with $(1+o(1))Δ$ colours.

cs.DS↗
arXiv2017

Short proofs for generalizations of the Lovász Local Lemma: Shearer's condition and cluster expansion

The Lovász Local Lemma is a seminal result in probabilistic combinatorics. It gives a sufficient condition on a probability space and a collection of events for the existence of an outcome that simultaneously avoids all of those events. Over the years, more general conditions have been discovered under which the conclusion of the lemma continues to hold. In this note we provide short proofs of two of those more general results: Shearer's lemma and the cluster expansion lemma, in their "lopsided" form. We conclude by using the cluster expansion lemma to prove that the symmetric form of the local lemma holds with probabilities bounded by $1/ed$, rather than the bound $1/e(d+1)$ required by the traditional proofs.

math.CO↗
arXiv2017

Computing the Independence Polynomial: from the Tree Threshold down to the Roots

We study an algorithm for approximating the multivariate independence polynomial $Z(\mathbf{z})$, with negative and complex arguments, an object that has strong connections to combinatorics and to statistical physics. In particular, the independence polynomial with negative arguments, $Z(-\mathbf{p})$, determines the Shearer region, the maximal region of probabilities to which the Lovasz Local Lemma (LLL) can be extended (Shearer 1985). In statistical physics, complex zeros of the independence polynomial relate to existence of phase transitions. Our main result is a deterministic algorithm to compute approximately the independence polynomial in any root-free complex polydisc centered at the origin. Our algorithm is essentially the same as Weitz's algorithm for positive parameters up to the tree uniqueness threshold, and the core of our analysis is a novel multivariate form of the correlation decay technique, which can handle non-uniform complex parameters. In particular, in the univariate real setting our work implies that Weitz's algorithm works in an interval between two critical points $(λ'_c(d), λ_c(d))$, and outside of this interval an approximation of $Z(\mathbf{z})$ is known to be NP-hard. As an application, we give a sub-exponential time algorithm for testing approximate membership in the Shearer region. We also give a new rounding based deterministic algorithm for Shearer's lemma (an extension of the LLL), which, however, runs in sub-exponential time. On the hardness side, we prove that evaluating $Z(\mathbf{z})$ at an arbitrary point in Shearer's region, and testing membership in Shearer's region, are #P-hard problems. We also establish the best possible dependence of the exponent of the run time of Weitz's correlation decay technique in the negative regime on the distance to the boundary of the Shearer region.

cs.DS↗
arXiv2013

The complexity of UNO

This paper investigates the popular card game UNO from the viewpoint of algorithmic combinatorial game theory. We define simple and concise mathematical models for the game, including both cooperative and uncooperative versions, and analyze their computational complexity. In particular, we prove that even a single-player version of UNO is NP-complete, although some restricted cases are in P. Surprisingly, we show that the uncooperative two-player version is also in P.

cs.DM↗
arXiv2013

Pipage Rounding, Pessimistic Estimators and Matrix Concentration

Pipage rounding is a dependent random sampling technique that has several interesting properties and diverse applications. One property that has been particularly useful is negative correlation of the resulting vector. Unfortunately negative correlation has its limitations, and there are some further desirable properties that do not seem to follow from existing techniques. In particular, recent concentration results for sums of independent random matrices are not known to extend to a negatively dependent setting. We introduce a simple but useful technique called concavity of pessimistic estimators. This technique allows us to show concentration of submodular functions and concentration of matrix sums under pipage rounding. The former result answers a question of Chekuri et al. (2009). To prove the latter result, we derive a new variant of Lieb's celebrated concavity theorem in matrix analysis. We provide numerous applications of these results. One is to spectrally-thin trees, a spectral analog of the thin trees that played a crucial role in the recent breakthrough on the asymmetric traveling salesman problem. We show a polynomial time algorithm that, given a graph where every edge has effective conductance at least $κ$, returns an $O(κ^{-1} \cdot \log n / \log \log n)$-spectrally-thin tree. There are further applications to rounding of semidefinite programs, to the column subset selection problem, and to a geometric question of extracting a nearly-orthonormal basis from an isotropic distribution.

cs.DS↗
arXiv2012

Submodular Functions: Learnability, Structure, and Optimization

Submodular functions are discrete functions that model laws of diminishing returns and enjoy numerous algorithmic applications. They have been used in many areas, including combinatorial optimization, machine learning, and economics. In this work we study submodular functions from a learning theoretic angle. We provide algorithms for learning submodular functions, as well as lower bounds on their learnability. In doing so, we uncover several novel structural results revealing ways in which submodular functions can be both surprisingly structured and surprisingly unstructured. We provide several concrete implications of our work in other domains including algorithmic game theory and combinatorial optimization. At a technical level, this research combines ideas from many areas, including learning theory (distributional learning and PAC-style analyses), combinatorics and optimization (matroids and submodular functions), and pseudorandomness (lossless expander graphs).

cs.DS↗
arXiv2011

Sparse Sums of Positive Semidefinite Matrices

Recently there has been much interest in "sparsifying" sums of rank one matrices: modifying the coefficients such that only a few are nonzero, while approximately preserving the matrix that results from the sum. Results of this sort have found applications in many different areas, including sparsifying graphs. In this paper we consider the more general problem of sparsifying sums of positive semidefinite matrices that have arbitrary rank. We give several algorithms for solving this problem. The first algorithm is based on the method of Batson, Spielman and Srivastava (2009). The second algorithm is based on the matrix multiplicative weights update method of Arora and Kale (2007). We also highlight an interesting connection between these two algorithms. Our algorithms have numerous applications. We show how they can be used to construct graph sparsifiers with auxiliary constraints, sparsifiers of hypergraphs, and sparse solutions to semidefinite programs.

cs.DM↗
arXiv2010

Graph Sparsification by Edge-Connectivity and Random Spanning Trees

We present new approaches to constructing graph sparsifiers --- weighted subgraphs for which every cut has the same value as the original graph, up to a factor of $(1 \pm ε)$. Our first approach independently samples each edge $uv$ with probability inversely proportional to the edge-connectivity between $u$ and $v$. The fact that this approach produces a sparsifier resolves a question posed by Benczúr and Karger (2002). Concurrent work of Hariharan and Panigrahi also resolves this question. Our second approach constructs a sparsifier by forming the union of several uniformly random spanning trees. Both of our approaches produce sparsifiers with $O(n \log^2(n)/ε^2)$ edges. Our proofs are based on extensions of Karger's contraction algorithm, which may be of independent interest.

cs.DS↗