Motivated by our collaboration with Food Rescue U.S. (FRUS), a food recovery organization that relies on volunteers to complete recurring tasks, we study how crowdsourcing platforms can use commitment to promote growth and engagement. Despite reducing match uncertainty, high levels of commitment can decrease the probability of forming new matches in the spot market, which in turn can suppress growth. To better understand this trade-off, we develop a model for two-sided random markets which repeatedly match volunteers with tasks. Our model incorporates match uncertainty as well as the negative impact of failing to match on future engagement. We study the optimal level of commitment to maximize the total discounted number of matches.

Finding the optimal pass sequence of compilation can lead to a significant reduction in program size. Prior works on compilation pass ordering have two major drawbacks. They either require an excessive budget (in terms of the number of compilation passes) at compile time or fail to generalize to unseen programs. In this work, instead of predicting passes sequentially, we directly learn a policy on the pass sequence space, which outperforms the default -Oz flag by an average of 4.5% over a large collection (4683) of unseen code repositories from diverse domains across 14 datasets. To achieve this, we first identify a small set (termed coreset) of pass sequences that generally optimize the size of most programs. Then, a policy is learned to pick the optimal sequences by predicting the normalized values of the pass sequences in the coreset. Our results demonstrate that existing human-designed compiler passes can be improved with a simple yet effective technique that leverages pass sequence space which contains dense rewards, while approaches operating on the individual pass space may suffer from issues of sparse reward, and do not generalize well to held-out programs from different domains. Website: https://rlcompopt.github.io.

We develop novel learning rates for conditional mean embeddings by applying the theory of interpolation for reproducing kernel Hilbert spaces (RKHS). We derive explicit, adaptive convergence rates for the sample estimator under the misspecifed setting, where the target operator is not Hilbert-Schmidt or bounded with respect to the input/output RKHSs. We demonstrate that in certain parameter regimes, we can achieve uniform convergence rates in the output RKHS. We hope our analyses will allow the much broader application of conditional mean embeddings to more complex ML/RL settings involving infinite dimensional RKHSs and continuous state spaces.

We introduce and study a variation of the submodular maximization problem motivated by applications in online retail. A platform displays a list of products to a user in response to a search query. The user inspects the first k items in the list for a k chosen at random from a given distribution, and decides whether to purchase an item from that set based on a choice model. The goal of the platform is to maximize the engagement of the shopper defined as the probability of purchase. This problem gives rise to a less-studied variation of submodular maximization in which we are asked to choose an ordering of a set of elements to maximize a linear combination of different submodular functions.

We study a continuous and infinite time horizon counterpart to the classic prophet inequality, which we term the stationary prophet inequality problem. Here, copies of a good arrive and perish according to Poisson point processes. Buyers arrive similarly and make take-it-or-leave-it offers for unsold items. The objective is to maximize the (infinite) time average revenue of the seller. Our main results are pricing-based policies which (i) achieve a 1/2-approximation of the optimal offline policy, which is best possible, and (ii) achieve a better than (1-1/e)-approximation of the optimal online policy. Result (i) improves upon bounds implied by recent work of Collina et al. (WINE'20), and is the first optimal prophet inequality for a stationary problem. Result (ii) improves upon a 1-1/e bound implied by recent work of Aouad and Sarita (EC'20), and shows that this prevalent bound in online algorithms is not optimal for this problem.

We study sample complexity of optimizing “hill-climbing friendly” functions defined on a graph under noisy observations. We define a notion of convexity, and we show that a variant of best-arm identification can find a near-optimal solution after a small number of queries that is independent of the size of the graph. For functions that have local minima and are nearly convex, we show a sample complexity for the classical simulated annealing under noisy observations. We show effectiveness of the greedy algorithm with restarts and the simulated annealing on problems of graph-based nearest neighbor classification as well as a web advertising application.

We study the assignment problem of objects to agents with heterogeneous preferences under distributional constraints. Each agent is associated with a publicly known type and has a private ordinal ranking over objects. We are interested in assigning as many agents as possible. Our first contribution is a generalization of the well-known and widely used serial dictatorship. Our mechanism maintains several desirable properties of serial dictatorship, including strategyproofness, Pareto efficiency, and computational tractability while satisfying the distributional constraints with a small error. We also propose a generalization of the probabilistic serial algorithm, which finds an ordinally efficient and envy-free assignment, and also satisfies the distributional constraints with a small error. We show, however, that no ordinally efficient and envy-free mechanism is also weakly strategyproof. Both of our algorithms assign at least the same number of students as the optimum fractional assignment.