Due to the rise of privacy concerns, in many practical applications, the training data is aggregated before being shared with the learner to protect the privacy of users' sensitive responses. In an aggregate learning framework, the dataset is grouped into bags of samples, where each bag is available only with an aggregate response, providing a summary of individuals' responses in that bag. In this paper, we study two natural loss functions for learning from aggregate responses: the bag-level loss and the instance-level loss. In the former, the model is learned by minimizing a loss between the aggregate responses and aggregate model predictions, while in the latter, the model aims to fit individual predictions to the aggregate responses. In this work, we show that the instance-level loss can be perceived as a regularized form of the bag-level loss. This observation allows us to compare the two approaches with respect to the bias and variance of the resulting estimators and to introduce a novel interpolating estimator that combines the two approaches. For linear regression tasks, we provide a precise characterization of the risk of the interpolating estimator in an asymptotic regime where the size of the training set grows in proportion to the feature dimension. Our analysis enables us to theoretically understand the effect of different factors, such as bag size, on the model's prediction risk. Additionally, we propose a mechanism for differentially private learning from aggregate responses and derive the optimal bag size in terms of the prediction risk-privacy trade-off. We also carry out thorough experiments to corroborate our theory and show the efficacy of the interpolating estimator.

Motivated by online retail, we consider the problem of selling one item (e.g., an ad slot) to two non-excludable buyers (say, a merchant and a brand). This problem captures, for example, situations where a merchant and a brand cooperatively bid in an auction to advertise a product, and both benefit from the ad being shown. A mechanism collects bids from the two and decides whether to allocate and which payments the two parties should make. This gives rise to intricate incentive compatibility constraints, e.g., on how to split payments between the two parties. We approach the problem of finding a revenue-maximizing incentive-compatible mechanism from an online learning perspective; this poses significant technical challenges. First, the action space (the class of all possible mechanisms) is huge; second, the function that maps mechanisms to revenue is highly irregular, ruling out standard discretization-based approaches.

Multiple instance regression (MIR) was introduced by Ray and Page (2001) as an analogue of multiple instance learning (MIL) in which we are given bags of feature-vectors (instances) and for each bag there is a bag-label which matches the label of one (unknown) primary instance from that bag. The goal is to compute a hypothesis regressor consistent with the underlying instance-labels. A natural approach is to find the best primary instance assignment and regressor optimizing the mse loss on the bags though no formal generalization guarantees were known. Our work is the first to prove generalization error bounds for MIR when the bags are drawn i.i.d. at random. Essentially, with high probability any MIR regressor with low error on sampled bags also has low error on the underlying instance-label distribution. We next study the complexity of linear regression on MIR bags, shown to be NP-hard in general by Ray and Page (2001), who however left open the possibility of arbitrarily good approximations. Significantly strengthening previous work, we prove a strong inapproximability bound: even if there exists zero bag-loss MIR linear regressor on a collection of $2$-sized bags with labels in $[-1,1]$, it is NP-hard to find an MIR linear regressor with bag-loss $< C$ for some absolute constant $C > 0$. Our work also proposes a model training method for MIR based on a novel weighted assignment loss, geared towards handling overlapping bags which have not received much attention previously. We conduct empirical evaluations on synthetic and real-world datasets showing that our method outperforms the baseline MIR methods.

We propose a new family of label randomizers for training _regression_ models under the constraint of label differential privacy (DP). In particular, we leverage the trade-offs between bias and variance to construct better label randomizers depending on a privately estimated prior distribution over the labels. We demonstrate that these randomizers achieve state-of-the-art privacy-utility trade-offs on several datasets, highlighting the importance of reducing bias when training neural networks with label DP. We also provide theoretical results shedding light on the structural properties of the optimal unbiased randomizers.

Standard contextual bandit problem assumes that all the relevant contexts are observed before the algorithm chooses an arm. This modeling paradigm, while useful, often falls short when dealing with problems in which additional valuable contexts can be observed after arm selection. For example, content recommendation platforms like Youtube, Instagram, Tiktok receive much additional features about a user's reward after the user clicks a content (e.g., how long the user stayed, what is the user's watch speed, etc.). To improve online learning efficiency in these applications, we study a novel contextual bandit problem with post-serving contexts and design a new algorithm, poLinUCB, that achieves tight regret under standard assumptions. Core to our technical proof is a robustified and generalized version of the well-known Elliptical Potential Lemma (EPL), which can accommodate noise in data. Such robustification is necessary for tackling our problem, though we believe it could also be of general interest. Extensive empirical tests on both synthetic and real-world datasets demonstrate the significant benefit of utilitzing post-serving contexts as well as the superior performance of our algorithm over the state-of-the-art approaches.

In this work, we investigate the problem of how to bid in repeated contextual first price auctions for a single learner (the bidder). Concretely, at each time t, the learner receives a context and decides the bid based on historical information and xt. We assume that the maximum bid of all the others follows a linear model, i.e., mt = ⟨α0, xt⟩ + zt, where is unknown to the learner and zt is randomly sampled from a noise distribution with log-concave density. In this work, we consider both binary feedback (the learner can only observe whether she wins or not) and full information feedback (the learner can observe mt) models of the learner. For binary feedback, when the noise distribution is (partially) known, we propose a bidding algorithm that achieves at most regret, where Δ is a constant reserve in first price auctions.1 For the full information feedback with unknown noise distribution, we provide an algorithm that achieves regret at most . Our approach combines an estimator for log-concave density functions and the Maximum Likelihood Estimation (MLE) method to learn the noise distribution and linear weight α0 simultaneously. We complement our results with a lower bound result such that any bidding policy in a broad class must achieve regret at least , even when the learner receives the full information feedback and is known.

Incrementality, which measures the causal effect of showing an ad to a potential customer (e.g. a user in an internet platform) versus not, is a central object for advertisers in online advertising platforms. This paper investigates the problem of how an advertiser can learn to optimize the bidding sequence in an online manner \emph{without} knowing the incrementality parameters in advance. We formulate the offline version of this problem as a specially structured episodic Markov Decision Process (MDP) and then, for its online learning counterpart, propose a novel reinforcement learning (RL) algorithm with regret at most $\widetilde{O}(H^2\sqrt{T})$, which depends on the number of rounds $H$ and number of episodes $T$, but does not depend on the number of actions (i.e., possible bids). A fundamental difference between our learning problem from standard RL problems is that the realized reward feedback from conversion incrementality is \emph{mixed} and \emph{delayed}. To handle this difficulty we propose and analyze a novel pairwise moment-matching algorithm to learn the conversion incrementality, which we believe is of independent interest.

Motivated by online decision-making in time-varying combinatorial environments, we study the problem of transforming offline algorithms to their online counterparts. We focus on offline combinatorial problems that are amenable to a constant factor approximation using a greedy algorithm that is robust to local errors. For such problems, we provide a general framework that efficiently transforms offline robust greedy algorithms to online ones using Blackwell approachability.

In this paper, we introduce a novel technique for constrained submodular maximization, inspired by barrier functions in continuous optimization. This connection not only improves the running time for constrained submodular maximization but also provides the state of the art guarantee. More precisely, for maximizing a monotone submodular function subject to the combination of a $k$-matchoid and $\ell$-knapsack constraints (for $\ell\leq k$), we propose a potential function that can be approximately minimized. Once we minimize the potential function up to an $\epsilon$ error, it is guaranteed that we have found a feasible set with a $2(k+1+\epsilon)$-approximation factor which can indeed be further improved to $(k+1+\epsilon)$ by an enumeration technique. We extensively evaluate the performance of our proposed algorithm over several real-world applications, including a movie recommendation system, summarization tasks for YouTube videos, Twitter feeds and Yelp business locations, and a set cover problem.

Modern ad auctions allow advertisers to target more specific segments of the user population. Unfortunately, this is not always in the best interest of the ad platform – partially hiding some information could be more beneficial for the platform's revenue. In this paper, we examine the following basic question in the context of second-price ad auctions: how should an ad platform optimally reveal information about the ad opportunity to the advertisers in order to maximize revenue? We consider a model in which bidders’ valuations depend on a random state of the ad opportunity. Different from previous work, we focus on a more practical, and challenging, situation where the space of possible realizations of ad opportunities is extremely large. We thus focus on developing algorithms whose running time is polynomial in the number of bidders, but is independent of the number of ad opportunity realizations. We assume that the auctioneer can commit to a signaling scheme to reveal noisy information about the realized state of the ad opportunity, and examine the auctioneer's algorithmic question of designing the optimal signaling scheme. We first consider that the auctioneer is restricted to send a public signal to all bidders. As a warm-up, we start with a basic (though less realistic) setting in which the auctioneer knows the bidders’ valuations, and show that an e-optimal scheme can be implemented in time polynomial in the number of bidders and 1/∊. We then move to a well-motivated Bayesian valuation setting in which the auctioneer and bidders both have private information, and present two results. First, we exhibit a characterization result regarding approximately optimal schemes and prove that any constant-approximate public signaling scheme must use exponentially many signals. Second, we present a “simple” public signaling scheme that serves as a constant approximation under mild assumptions. Finally, we initiate an exploration on the power of being able to send different signals privately to different bidders. In the basic setting where the auctioneer knows bidders’ valuations, we exhibit a polynomial-time private scheme that extracts almost full surplus even in the worst Bayes Nash equilibrium. This illustrates the surprising power of private signaling schemes in extracting revenue.

Multi-armed bandit problems are the predominant theoretical model of exploration-exploitation tradeoffs in learning, and they have countless applications ranging from medical trials, to communication networks, to Web search and advertising. In many of these application domains, the learner may be constrained by one or more supply (or budget) limits, in addition to the customary limitation on the time horizon. The literature lacks a general model encompassing these sorts of problems. We introduce such a model, called bandits with knapsacks , that combines bandit learning with aspects of stochastic integer programming. In particular, a bandit algorithm needs to solve a stochastic version of the well-known knapsack problem , which is concerned with packing items into a limited-size knapsack. A distinctive feature of our problem, in comparison to the existing regret-minimization literature, is that the optimal policy for a given latent distribution may significantly outperform the policy that plays the optimal fixed arm. Consequently, achieving sublinear regret in the bandits-with-knapsacks problem is significantly more challenging than in conventional bandit problems. We present two algorithms whose reward is close to the information-theoretic optimum: one is based on a novel “balanced exploration” paradigm, while the other is a primal-dual algorithm that uses multiplicative updates. Further, we prove that the regret achieved by both algorithms is optimal up to polylogarithmic factors. We illustrate the generality of the problem by presenting applications in a number of different domains, including electronic commerce, routing, and scheduling. As one example of a concrete application, we consider the problem of dynamic posted pricing with limited supply and obtain the first algorithm whose regret, with respect to the optimal dynamic policy, is sublinear in the supply.

Can we summarize multi-category data based on user preferences in a scalable manner? Many utility functions used for data summarization satisfy submodularity, a natural diminishing returns property. We cast personalized data summarization as an instance of a general submodular maximization problem subject to multiple constraints. We develop the first practical and FAst coNsTrained submOdular Maximization algorithm, FANTOM, with strong theoretical guarantees. FANTOM maximizes a submodular function (not necessarily monotone) subject to intersection of a p-system and l knapsacks constrains. It achieves a (1 + ε)(p + 1)(2p + 2l + 1)/p approximation guarantee with only O(nrp log(n)/ε) query complexity (n and r indicate the size of the ground set and the size of the largest feasible solution, respectively). We then show how we can use FANTOM for personalized data summarization. In particular, a p-system can model different aspects of data, such as categories or time stamps, from which the users choose. In addition, knapsacks encode users’ constraints including budget or time. In our set of experiments, we consider several concrete applications: movie recommendation over 11K movies, personalized image summarization with 10K images, and revenue maximization on the YouTube social networks with 5000 communities. We observe that FANTOM constantly provides the highest utility against all the baselines.

The Adaptive Seeding problem is an algorithmic challenge motivated by influence maximization in social networks: One seeks to select among certain accessible nodes in a network, and then select, adaptively, among neighbors of those nodes as they become accessible in order to maximize a global objective function. More generally, adaptive seeding is a stochastic optimization framework where the choices in the first stage affect the realizations in the second stage, over which we aim to optimize. Our main result is a (1 – 1/e)2-approximation for the adaptive seeding problem for any monotone submodular function. While adaptive policies are often approximated via non-adaptive policies, our algorithm is based on a novel method we call locally-adaptive policies. These policies combine a non-adaptive global structure, with local adaptive optimizations. This method enables the (1–1/e)2-approximation for general monotone submodular functions and circumvents some of the impossibilities associated with non-adaptive policies. We also introduce a fundamental problem in submodular optimization that may be of independent interest: given a ground set of elements where every element appears with some small probability, find a set of expected size at most k that has the highest expected value over the realization of the elements. We show a surprising result: there are classes of monotone submodular functions (including coverage) that can be approximated almost optimally as the probability vanishes. For general monotone submodular functions we show via a reduction from Planted-Clique that approximations for this problem are not likely to be obtainable. This optimization problem is an important tool for adaptive seeding via non-adaptive policies, and its hardness motivates the introduction of locally-adaptive policies we use in the main result.

Is it possible to maximize a monotone submodular function faster than the widely used lazy greedy algorithm (also known as accelerated greedy), both in theory and practice? In this paper, we develop the first linear-time algorithm for maximizing a general monotone submodular function subject to a cardinality constraint. We show that our randomized algorithm, STOCHASTIC-GREEDY, can achieve a (1 − 1/e − ε) approximation guarantee, in expectation, to the optimum solution in time linear in the size of the data and independent of the cardinality constraint. We empirically demonstrate the effectiveness of our algorithm on submodular functions arising in data summarization, including training large-scale kernel methods, exemplar-based clustering, and sensor placement. We observe that STOCHASTIC-GREEDY practically achieves the same utility value as lazy greedy but runs much faster. More surprisingly, we observe that in many practical scenarios STOCHASTIC-GREEDY does not evaluate the whole fraction of data points even once and still achieves indistinguishable results compared to lazy greedy.

How can one find a subset, ideally as small as possible, that well represents a massive dataset? I.e., its corresponding utility, measured according to a suitable utility function, should be comparable to that of the whole dataset. In this paper, we formalize this challenge as a submodular cover problem. Here, the utility is assumed to exhibit submodularity, a natural diminishing returns condition preva- lent in many data summarization applications. The classical greedy algorithm is known to provide solutions with logarithmic approximation guarantees compared to the optimum solution. However, this sequential, centralized approach is imprac- tical for truly large-scale problems. In this work, we develop the first distributed algorithm – DISCOVER – for submodular set cover that is easily implementable using MapReduce-style computations. We theoretically analyze our approach, and present approximation guarantees for the solutions returned by DISCOVER. We also study a natural trade-off between the communication cost and the num- ber of rounds required to obtain such a solution. In our extensive experiments, we demonstrate the effectiveness of our approach on several applications, includ- ing active set selection, exemplar based clustering, and vertex cover on tens of millions of data points using Spark.

We study decision making when each action is described by a set of objectives, all of which are to be maximized. During the training phase, we have access to the actions of an outside agent (“mentor”). In the test phase, our goal is to maximally improve upon the mentor’s (unobserved) actions across all objectives. We present an algorithm with a vanishing regret compared with the optimal possible improvement, and show that our regret bound is the best possible. The bound is independent of the number of actions, and scales only as the logarithm of the number of objectives.

We study contextual bandits with ancillary constraints on resources, which are common in real-world applications such as choosing ads or dynamic pricing of items. We design the first algorithm for solving these problems that improves over a trivial reduction to the non-contextual case. We consider very general settings for both contextual bandits (arbitrary policy sets, Dudik et al. (2011)) and bandits with resource constraints (bandits with knapsacks, Badanidiyuru et al. (2013a)), and prove a regret guarantee with near-optimal statistical properties.

How can one summarize a massive data set "on the fly", i.e., without even having seen it in its entirety? In this paper, we address the problem of extracting representative elements from a large stream of data. I.e., we would like to select a subset of say k data points from the stream that are most representative according to some objective function. Many natural notions of "representativeness" satisfy submodularity, an intuitive notion of diminishing returns. Thus, such problems can be reduced to maximizing a submodular set function subject to a cardinality constraint. Classical approaches to submodular maximization require full access to the data set. We develop the first efficient streaming algorithm with constant factor 1/2-ε approximation guarantee to the optimum solution, requiring only a single pass through the data, and memory independent of data size. In our experiments, we extensively evaluate the effectiveness of our approach on several applications, including training large-scale kernel methods and exemplar-based clustering, on millions of data points. We observe that our streaming method, while achieving practically the same utility value, runs about 100 times faster than previous work.

There has been much progress recently on improved approximations for problems involving submodular objective functions, and many interesting techniques have been developed. However, the resulting algorithms are often slow and impractical. In this paper we develop algorithms that match the best known approximation guarantees, but with significantly improved running times, for maximizing a monotone submodular function f : 2[n] → ℝ+ subject to various constraints. As in previous work, we measure the number of oracle calls to the objective function which is the dominating term in the running time. Our first result is a simple algorithm that gives a (1 − 1/∊ − ∊)-approximation for a cardinality constraint using queries, and a 1/(p + 2ℓ + 1 + ∊)-approximation for the intersection of a p-system and ℓ knapsack (linear) constraints using queries. This is the first approximation for a p-system combined with linear constraints. (We also show that the factor of p cannot be improved for maximizing over a p-system.) The main idea behind these algorithms serves as a building block in our more sophisticated algorithms. Our main result is a new variant of the continuous greedy algorithm, which interpolates between the classical greedy algorithm and a truly continuous algorithm. We show how this algorithm can be implemented for matroid and knapsack constraints using Õ(n2) oracle calls to the objective function. (Previous variants and alternative techniques were known to use at least Õ(n4) oracle calls.) This leads to an -time (1 − 1/∊ − ∊)-approximation for a matroid constraint. For a knapsack constraint, we develop a more involved (1 − 1/∊ − ∊)-approximation algorithm that runs in time .

Multi-armed bandit problems are the predominant theoretical model of exploration-exploitation tradeoffs in learning, and they have countless applications ranging from medical trials, to communication networks, to Web search and advertising. In many of these application domains the learner may be constrained by one or more supply (or budget) limits, in addition to the customary limitation on the time horizon. The literature lacks a general model encompassing these sorts of problems. We introduce such a model, called "bandits with knapsacks", that combines aspects of stochastic integer programming with online learning. A distinctive feature of our problem, in comparison to the existing regret-minimization literature, is that the optimal policy for a given latent distribution may significantly outperform the policy that plays the optimal fixed arm. Consequently, achieving sub linear regret in the bandits-with-knapsacks problem is significantly more challenging than in conventional bandit problems. We present two algorithms whose reward is close to the information-theoretic optimum: one is based on a novel "balanced exploration" paradigm, while the other is a primal-dual algorithm that uses multiplicative updates. Further, we prove that the regret achieved by both algorithms is optimal up to polylogarithmic factors. We illustrate the generality of the problem by presenting applications in a number of different domains including electronic commerce, routing, and scheduling. As one example of a concrete application, we consider the problem of dynamic posted pricing with limited supply and obtain the first algorithm whose regret, with respect to the optimal dynamic policy, is sub linear in the supply.

We study the complexity of the maximum coverage problem, restricted to set systems of bounded VC-dimension. Our main result is a fixed-parameter tractable approximation scheme: an algorithm that outputs a (1-ε)-approximation to the maximum-cardinality union of k sets, in running time $O(f(ε,k,d)⋅ poly(n)) where n is the problem size, d is the VC-dimension of the set system, and f(ε,k,d) is exponential in (kd/ε)c for some constant c. We complement this positive result by showing that the function f(ε,k,d) in the running-time bound cannot be replaced by a function depending only on (ε,d) or on (k,d), under standard complexity assumptions. We also present an improved upper bound on the approximation ratio of the greedy algorithm in special cases of the problem, including when the sets have bounded cardinality and when they are two-dimensional halfspaces. Complementing these positive results, we show that when the sets are four-dimensional halfspaces neither the greedy algorithm nor local search is capable of improving the worst-case approximation ratio of 1-1/e that the greedy algorithm achieves on arbitrary instances of maximum coverage.

We study combinatorial procurement auctions, where a buyer with a valuation function v and budget B wishes to buy a set of items. Each item i has a cost ci and the buyer is interested in a set S that maximizes v(S) subject to ∑i∈Sci ≤ β. Special cases of combinatorial procurement auctions are well-studied problems from submodular optimization. In particular, when the costs are all equal (cardinality constraint), a classic result by Nemhauser et al shows that the greedy algorithm provides an e/e-1 approximation.

We study online procurement markets where agents arrive in a sequential order and a mechanism must make an irrevocable decision whether or not to procure the service as the agent arrives. Our mechanisms are subject to a budget constraint and are designed for stochastic settings in which the bidders are either identically distributed or, more generally, permuted in random order. Thus, the problems we study contribute to the literature on budget-feasible mechanisms as well as the literature on secretary problems and online learning in auctions.

Motivated by the problem of querying and communicating bidders' valuations in combinatorial auctions, we study how well different classes of set functions can be sketched. More formally let f be a function mapping subsets of some ground set [n] to the non-negative real numbers. We say that f′ is an α-sketch of f if for every set S, the value f′(S) lies between f(S)/α and f(S), and f′ can be specified by poly(n) bits. We show that for every subadditive function f there exists an α-sketch where α = n1/2 · O(polylog(n)). Furthermore, we provide an algorithm that finds these sketches with a polynomial number of demand queries. This is essentially the best we can hope for since: 1. We show that there exist subadditive functions (in fact, XOS functions) that do not admit an o(n1/2) sketch. (Balcan and Harvey [3] previously showed that there exist functions belonging to the class of substitutes valuations that do not admit an O(n1/3) sketch.) 2. We prove that every deterministic algorithm that accesses the function via value queries only cannot guarantee a sketching ratio better than n1−ε. We also show that coverage functions, an interesting subclass of submodular functions, admit arbitrarily good sketches. Finally, we show an interesting connection between sketching and learning. We show that for every class of valuations, if the class admits an α-sketch, then it can be α-approximately learned in the PMAC model of Balcan and Harvey. The bounds we prove are only information-theoretic and do not imply the existence of computationally efficient learning algorithms in general.

Perfectly reliable message transmission (PRMT) is one of the fundamental problems in distributed computing. It allows a sender to reliably transmit a message to a receiver in an unreliable network, even in the presence of a computationally unbounded adversary. In this article, we study the inherent trade-off between the three important parameters of the PRMT protocols, namely, the network connectivity ( n ), the round complexity ( r ), and the communication complexity by considering the following generic question (which can be considered as the holy grail problem) in the context of the PRMT protocols. Given an n -connected network, a message of size ℓ (to be reliably communicated) and a limit c for the total communication allowed between the sender and the receiver, what is the minimum number of communication rounds required by a PRMT protocol to send the message, such that the communication complexity of the protocol is O( c )? We answer this interesting question by deriving a nontrivial lower bound on the round complexity. Moreover, we show that the lower bound is tight in the amortized sense, by designing a PRMT protocol whose round complexity matches the lower bound. The lower bound is the first of its kind, that simultaneously captures the inherent tradeoff between the three important parameters of a PRMT protocol.

In this paper, we study the inherent tradeoff between the network connectivity, phase complexity and communication complexity of perfectly reliable message transmission (PRMT) problem in undirected synchronous network, tolerating a mixed adversary A(tb,tf), who has unbounded computing power and can corrupt tb and tf nodes in the network in Byzantine and fail-stop fashion respectively.