Transformers, especially the decoder-only variants, are the backbone of most modern large language models. Yet, we have a very limited understanding of their limitations (i.e., what tasks they cannot solve) besides the simplest 1-layer case. Due to the difficulty of analyzing multi-layer models, all previous work relies on unproven complexity conjectures to show limitations for multi-layer Transformers. In this work, we prove the first unconditional lower bound against multilayer decoder-only transformers. For any constant L, we prove that any L-layer decoder-only transformer needs a polynomial model dimension $\left(n^{\Omega(1)}\right)$ to perform sequential composition of L functions over an input of n tokens. As a consequence, our results give: (1) the first depthwidth trade-off for multi-layer transformers, exhibiting that the L-step composition task is exponentially harder for L-layer models compared to $(L+1)$-layer ones; (2) an unconditional separation between encoder and decoder, exhibiting a hard task for decoders that can be solved by an exponentially shallower and smaller encoder; (3) a provable advantage of chain-of-thought, exhibiting a task that becomes exponentially easier when the model is allowed to produce an intermediate sequence of tokens before outputting the final answer. On the technical side, we propose the multi-party autoregressive communication model that abstracts the key aspects of a decoder-only Transformer. In particular, lower bounds within this communication model imply lower bounds against decoder-only transformers, independent of their implementation details. To prove lower bounds in this communication model, we also introduce a new proof technique that finds a certain indistinguishable decomposition of all possible inputs iteratively. We believe our new communication model and proof techniques will be helpful to understand the computational power of transformers further.

We revisit the problem of selecting $k$-out-of-$n$ elements with the goal of optimizing an objective function, and ask whether it can be solved approximately with sublinear query complexity. For objective functions that are monotone submodular, [Li, Feldman, Kazemi, Karbasi, NeurIPS’22; Kuhnle, AISTATS’21] gave an $\Omega(n/k)$ query lower bound for approximating to within any constant factor. We strengthen their lower bound to a nearly tight $\tilde{\Omega}(n)$. This lower bound holds even for estimating the value of the optimal subset. When the objective function is additive, we prove that finding an approximately optimal subset still requires near-linear query complexity, but we can estimate the value of the optimal subset in $\tilde{O}(n/k)$ queries, and that this is tight up to polylog factors.

The problem of continual learning in the domain of reinforcement learning, often called non-stationary reinforcement learning, has been identified as an important challenge to the application of reinforcement learning. We prove a worst-case complexity result, which we believe captures this challenge: Modifying the probabilities or the reward of a single state-action pair in a reinforcement learning problem requires an amount of time almost as large as the number of states in order to keep the value function up to date, unless the strong exponential time hypothesis (SETH) is false; SETH is a widely accepted strengthening of the P $\neq$ NP conjecture. Recall that the number of states in current applications of reinforcement learning is typically astronomical. In contrast, we show that just adding a new state-action pair is considerably easier to implement.

Multi-distribution learning generalizes the classic PAC learning to handle data coming from multiple distributions. Given a set of $k$ data distributions and a hypothesis class of VC dimension $d$, the goal is to learn a hypothesis that minimizes the maximum population loss over $k$ distributions, up to $\epsilon$ additive error. In this paper, we settle the sample complexity of multi-distribution learning by giving an algorithm of sample complexity $\widetilde{O}((d+k)\epsilon^{-2}) \cdot (k/\epsilon)^{o(1)}$. This matches the lower bound up to sub-polynomial factor and resolves the COLT 2023 open problem of Awasthi, Haghtalab and Zhao.

We study the iteration complexity of decentralized learning of approximate correlated equilibria in incomplete information games. On the negative side, we prove that in extensive-form games, assuming $\mathsf{PPAD} \not\subset \mathsf{TIME}(n^{\polylog(n)})$, any polynomial-time learning algorithms must take at least $2^{\log_2^{1-o(1)}(|\mathcal{I}|)}$ iterations to converge to the set of $\epsilon$-approximate correlated equilibrium, where $|\mathcal{I}|$ is the number of nodes in the game and $\epsilon > 0$ is an absolute constant. This nearly matches, up to the $o(1)$ term, the algorithms of (Peng and Rubinstein STOC’2024, Dagan et al. STOC’2024) for learning $\epsilon$-approximate correlated equilibrium, and resolves an open question of Anagnostides, Kalavasis, Sandholm, and Zampetakis (Anagnostides et al. ITCS 2024). Our lower bound holds even for the easier solution concept of $\epsilon$-approximate coarse correlated equilibrium. On the positive side, we give uncoupled dynamics that reach $\epsilon$-approximate correlated equilibria of a Bayesian game in polylogarithmic iterations, without any dependence of the number of types. This demonstrates a separation between Bayesian games and extensive-form games.

We give a simple and computationally efficient algorithm that, for any constant ε>0, obtains ε T-swap regret within only T = (n) rounds; this is an exponential improvement compared to the super-linear number of rounds required by the state-of-the-art algorithm, and resolves the main open problem of ‍[]. Our algorithm has an exponential dependence on ε, but we prove a new, matching lower bound. Our algorithm for swap regret implies faster convergence to ε-Correlated Equilibrium (ε-CE) in several regimes: For normal form two-player games with n actions, it implies the first uncoupled dynamics that converges to the set of ε-CE in polylogarithmic rounds; a (n)-bit communication protocol for ε-CE in two-player games (resolving an open problem mentioned by ‍[, , ]); and an Õ(n)-query algorithm for ε-CE (resolving an open problem of ‍[] and obtaining the first separation between ε-CE and ε-Nash equilibrium in the query complexity model). For extensive-form games, our algorithm implies a PTAS for normal form correlated equilibria, a solution concept often conjectured to be computationally intractable (e.g. ‍[, ]).

We show that any randomized first-order algorithm which minimizes a d-dimensional, 1-Lipschitz convex function over the unit ball must either use $\Omega\left(d^{2-\delta}\right)$ bits of memory or make $\Omega\left(d^{1+\delta / 6-o(1)}\right)$ queries, for any constant $\delta \in(0,1)$ and when the precision $\epsilon$ is quasipolynomially small in d. Our result implies that cutting plane methods, which use $\tilde{O}\left(d^{2}\right)$ bits of memory and $\tilde{O}(d)$ queries, are Pareto-optimal among randomized first-order algorithms, and quadratic memory is required to achieve optimal query complexity for convex optimization.

We settle the complexity of dynamic least-squares regression (LSR), where rows and labels $\left(\mathbf{A}^{(t)}, \mathbf{b}^{(t)}\right)$ can be adaptively inserted and/or deleted, and the goal is to efficiently maintain an $\epsilon$-approximate solution to $\min _{\mathbf{x}^{(t)}}\left\|\mathbf{A}^{(t)} \mathbf{x}^{(t)}-\mathbf{b}^{(t)}\right\|_{2}$ for all $t \in[T]$. We prove sharp separations $\left(d^{2-o(1)}\right.$ vs. $\left.\sim d\right)$ between the amortized update time of: (i) Fully vs. Partially dynamic 0.01-LSR; (ii) High vs. low-accuracy LSR in the partially-dynamic (insertion-only) setting.Our lower bounds follow from a gap-amplification reduction–reminiscent of iterative refinement-from the exact version of the Online Matrix Vector Conjecture (OMv) [HKNS15], to constant approximate OMv over the reals, where the i-th online product $\mathrm{Hv}^{(i)}$ only needs to be computed to 0.1 -relative error. All previous fine-grained reductions from OMv to its approximate versions only show hardness for inverse polynomial approximation $\epsilon=$ $n^{-\omega(1)}$ (additive or multiplicative). This result is of independent interest in fine-grained complexity and for the investigation of the OMv Conjecture, which is still widely open.

In the experts problem, on each of T days, an agent needs to follow the advice of one of n “experts”. After each day, the loss associated with each expert’s advice is revealed. A fundamental result in learning theory says that the agent can achieve vanishing regret, i.e. their cumulative loss is within $o(T)$ of the cumulative loss of the best-in-hindsight expert. Can the agent perform well without sufficient space to remember all the experts? We extend a nascent line of research on this question in two directions: 1) We give a new algorithm against the oblivious adversary, improving over the memory-regret tradeoff obtained by [PZ23], and nearly matching the lower bound of [SWXZ22]. 2) We also consider an adaptive adversary who can observe past experts chosen by the agent. In this setting we give both a new algorithm and a novel lower bound, proving that roughly $\sqrt{n}$ memory is both necessary and sufficient for obtaining $o(T)$ regret.

We provide the first sub-linear space and sub-linear regret algorithm for online learning with expert advice (against an oblivious adversary), addressing an open question raised recently by Srinivas, Woodruff, Xu and Zhou (STOC 2022). We also demonstrate a separation between oblivious and (strong) adaptive adversaries by proving a linear memory lower bound of any sub-linear regret algorithm against an adaptive adversary. Our algorithm is based on a novel pool selection procedure that bypasses the traditional wisdom of leader selection for online learning, and a generic reduction that transforms any weakly sub-linear regret o(T) algorithm to T1-α regret algorithm, which may be of independent interest. Our lower bound utilizes the connection of no-regret learning and equilibrium computation in zero-sum games, leading to a proof of a strong lower bound against an adaptive adversary. * The full version of the paper can be accessed at https://arxiv.org/abs/2207.07974

We study the complexity of finding an approximate (pure) Bayesian Nash equilibrium in a first-price auction with common priors when the tie-breaking rule is part of the input. We show that the problem is PPAD-complete even when the tie-breaking rule is trilateral (i.e., it specifies item allocations when no more than three bidders are in tie, and adopts the uniform tie-breaking rule otherwise). This is the first hardness result for equilibrium computation in first-price auctions with common priors. On the positive side, we give a PTAS for the problem under the uniform tie-breaking rule.

Motivated by the exploding growth of web-based services and the importance of efficiently managing the computational resources of such systems, we introduce and study a theoretical model for load balancing of very large databases such as commercial search engines. Our model is a more realistic version of the well-received \bab model with an additional constraint that limits the number of servers that carry each piece of the data. This additional constraint is necessary when, on one hand, the data is so large that we can not copy the whole data on each server. On the other hand, the query response time is so limited that we can not ignore the fact that the number of queries for each piece of the data changes over time, and hence we can not simply split the data over different machines. In this paper, we develop an almost optimal load balancing algorithm that works given an estimate of the load of each piece of the data. Our algorithm is almost perfectly robust to wrong estimates, to the extent that even when all of the loads are adversarially chosen the performance of our algorithm is $1-1/e$, which is provably optimal. Along the way, we develop various techniques for analyzing the balls-into-bins process under certain correlations and build a novel connection with the multiplicative weights update scheme.

Continual learning, or lifelong learning, is a formidable current challenge to machine learning. It requires the learner to solve a sequence of k different learning tasks, one after the other, while retaining its aptitude for earlier tasks; the continual learner should scale better than the obvious solution of developing and maintaining a separate learner for each of the k tasks. We embark on a complexity-theoretic study of continual learning in the PAC framework. We make novel uses of communication complexity to establish that any continual learner, even an improper one, needs memory that grows linearly with k, strongly suggesting that the problem is intractable. When logarithmically many passes over the learning tasks are allowed, we provide an algorithm based on multiplicative weights update whose memory requirement scales well; we also establish that improper learning is necessary for such performance. We conjecture that these results may lead to new promising approaches to continual learning.

Continual learning is an emerging paradigm in machine learning, wherein a model is exposed in an online fashion to data from multiple different distributions (i.e. environments), and is expected to adapt to the distribution change. Precisely, the goal is to perform well in the new environment, while simultaneously retaining the performance on the previous environments (i.e. avoid ``catastrophic forgetting'').While this setup has enjoyed a lot of attention in the applied community, there hasn't be theoretical work that even formalizes the desired guarantees. In this paper, we propose a framework for continual learning through the framework of feature extraction---namely, one in which features, as well as a classifier, are being trained with each environment. When the features are linear, we design an efficient gradient-based algorithm $\mathsf{DPGrad}$, that is guaranteed to perform well on the current environment, as well as avoid catastrophic forgetting. In the general case, when the features are non-linear, we show such an algorithm cannot exist, whether efficient or not.

Motivated by the exploding growth of web-based services and the importance of efficiently managing the computational resources of such systems, we introduce and study a theoretical model for load balancing of very large databases such as commercial search engines. Our model is a more realistic version of the well-received balls-into-bins model with an additional constraint that limits the number of servers that carry each piece of the data. This additional constraint is necessary when, on one hand, the data is so large that we can not copy the whole data on each server. On the other hand, the query response time is so limited that we can not ignore the fact that the number of queries for each piece of the data changes over time, and hence we can not simply split the data over different machines In this paper, we develop an almost optimal load balancing algorithm that works given an estimate of the load of each piece of the data. Our algorithm is almost perfectly robust to wrong estimates, to the extent that even when all of the loads are adversarially chosen the performance of our algorithm is 1–1/e, which is provably optimal. Along the way, we develop various techniques for analyzing the balls-into-bins process under certain correlations and build a novel connection with the multiplicative weights update scheme.

We study the complexity of the classic Hylland-Zeckhauser scheme [21] for one-sided matching markets. We show that the problem of finding an ∊-approximate equilibrium in the HZ scheme is PPAD-hard, and this holds even when ∊ is polynomially small and when each agent has no more than four distinct utility values. Our hardness result, when combined with the PPAD membership result of [29], resolves the approximation complexity of the HZ scheme. We also show that the problem of approximating within a certain constant factor the optimal social welfare (the weight of the matching) achievable by HZ equilibria is NP-hard.

We study dynamic algorithms for the problem of maximizing a monotone submodular function over a stream of n insertions and deletions. We show that any algorithm that maintains a (0.5+є)-approximate solution under a cardinality constraint, for any constant є>0, must have an amortized query complexity that is polynomial in n. Moreover, a linear amortized query complexity is needed in order to maintain a 0.584-approximate solution. This is in sharp contrast with recent dynamic algorithms of [LMN+20, Mon20] that achieve (0.5−є)-approximation with a polylog(n) amortized query complexity.

We consider the adaptive influence maximization problem: given a network and a budget k, iteratively select k seeds in the network to maximize the expected number of adopters. In the full-adoption feedback model, after selecting each seed, the seed-picker observes all the resulting adoptions. In the myopic feedback model, the seed-picker only observes whether each neighbor of the chosen seed adopts. Motivated by the extreme success of greedy-based algorithms/heuristics for influence maximization, we propose the concept of greedy adaptivity gap, which compares the performance of the adaptive greedy algorithm to its non-adaptive counterpart. Our first result shows that, for submodular influence maximization, the adaptive greedy algorithm can perform up to a (1 − 1/e)-fraction worse than the non-adaptive greedy algorithm, and that this ratio is tight. More specifically, on one side we provide examples where the performance of the adaptive greedy algorithm is only a (1−1/e) fraction of the performance of the non-adaptive greedy algorithm in four settings: for both feedback models and both the independent cascade model and the linear threshold model. On the other side, we prove that in any submodular cascade, the adaptive greedy algorithm always outputs a (1 − 1/e)-approximation to the expected number of adoptions in the optimal non-adaptive seed choice. Our second result shows that, for the general submodular diffusion model with full-adoption feedback, the adaptive greedy algorithm can outperform the non-adaptive greedy algorithm by an unbounded factor. Finally, we propose a risk-free variant of the adaptive greedy algorithm that always performs no worse than the non-adaptive greedy algorithm.

Shunyu Yao, Binghui Peng, Christos Papadimitriou, Karthik Narasimhan. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Inspired by applications on search engines and web servers, we consider a load balancing problem with a general convex objective function. In this problem, we are given a bipartite graph on a set of sources S and a set of workers W and the goal is to distribute the load from each source among its neighboring workers such that the total load of workers are as balanced as possible. We present a new distributed algorithm that works with any symmetric non-decreasing convex function for evaluating the balancedness of the workers' load. Our algorithm computes a nearly optimal allocation of loads in O(log n log² d/ε³) rounds where n is the number of nodes, d is the maximum degree, and ε is the desired precision. If the objective is to minimize the maximum load, we modify the algorithm to obtain a nearly optimal solution in O(log n log d/ε²) rounds. This improves a line of algorithms that require a polynomial number of rounds in n and d and appear to encounter a fundamental barrier that prevents them from obtaining poly-logarithmic runtime [Berenbrink et al., 2005; Berenbrink et al., 2009; Subramanian and Scherson, 1994; Rabani et al., 1998]. In our paper, we introduce a novel primal-dual approach with multiplicative weight updates that allows us to circumvent this barrier. Our algorithm is inspired by [Agrawal et al., 2018] and other distributed algorithms for optimizing linear objectives but introduces several new twists to deal with general convex objectives.

The slow convergence rate and pathological curvature issues of first-order gradient methods for training deep neural networks, initiated an ongoing effort for developing faster $\mathit{second}$-$\mathit{order}$ optimization algorithms beyond SGD, without compromising the generalization error. Despite their remarkable convergence rate ($\mathit{independent}$ of the training batch size $n$), second-order algorithms incur a daunting slowdown in the $\mathit{cost}$ $\mathit{per}$ $\mathit{iteration}$ (inverting the Hessian matrix of the loss function), which renders them impractical. Very recently, this computational overhead was mitigated by the works of [ZMG19,CGH+19}, yielding an $O(mn^2)$-time second-order algorithm for training two-layer overparametrized neural networks of polynomial width $m$. We show how to speed up the algorithm of [CGH+19], achieving an $\tilde{O}(mn)$-time backpropagation algorithm for training (mildly overparametrized) ReLU networks, which is near-linear in the dimension ($mn$) of the full gradient (Jacobian) matrix. The centerpiece of our algorithm is to reformulate the Gauss-Newton iteration as an $\ell_2$-regression problem, and then use a Fast-JL type dimension reduction to $\mathit{precondition}$ the underlying Gram matrix in time independent of $M$, allowing to find a sufficiently good approximate solution via $\mathit{first}$-$\mathit{order}$ conjugate gradient. Our result provides a proof-of-concept that advanced machinery from randomized linear algebra -- which led to recent breakthroughs in $\mathit{convex}$ $\mathit{optimization}$ (ERM, LPs, Regression) -- can be carried over to the realm of deep learning as well.

We initiate a systematic study on {\em dynamic influence maximization} (DIM). In the DIM problem, one maintains a seed set $S$ of at most $k$ nodes in a dynamically involving social network, with the goal of maximizing the expected influence spread while minimizing the amortized updating cost. We consider two evolution models. In the {\em incremental model}, the social network gets enlarged over time and one only introduces new users and establishes new social links, we design an algorithm that achieves $(1-1/e-\epsilon)$-approximation to the optimal solution and has $k \cdot\mathsf{poly}(\log n, \epsilon^{-1})$ amortized running time, which matches the state-of-art offline algorithm with only poly-logarithmic overhead. In the fully dynamic model, users join in and leave, influence propagation gets strengthened or weakened in real time, we prove that under the Strong Exponential Time Hypothesis (SETH), no algorithm can achieve $2^{-(\log n)^{1-o(1)}}$-approximation unless the amortized running time is $n^{1-o(1)}$. On the technical side, we exploit novel adaptive sampling approaches that reduce DIM to the dynamic MAX-k coverage problem, and design an efficient $(1-1/e-\epsilon)$-approximation algorithm for it. Our lower bound leverages the recent developed distributed PCP framework.

Public goods games in undirected networks are generally known to have pure Nash equilibria, which are easy to find. In contrast, we prove that, in directed networks, a broad range of public goods games have intractable equilibrium problems: The existence of pure Nash equilibria is NP-hard to decide, and mixed Nash equilibria are PPAD-hard to find. We define general utility public goods games, and prove a complexity dichotomy result for finding pure equilibria, and a PPAD-completeness proof for mixed Nash equilibria. Even in the divisible goods variant of the problem, where existence is easy to prove, finding the equilibrium is PPAD-complete. Finally, when the treewidth of the directed network is appropriately bounded, we prove that polynomial-time algorithms are possible.

We consider the adaptive influence maximization problem: given a network and a budget k, iteratively select k seeds in the network to maximize the expected number of adopters. In the full-adoption feedback model, after selecting each seed, the seed-picker observes all the resulting adoptions. In the myopic feedback model, the seed-picker only observes whether each neighbor of the chosen seed adopts. Motivated by the extreme success of greedy-based algorithms/heuristics for influence maximization, we propose the concept of greedy adaptivity gap, which compares the performance of the adaptive greedy algorithm to its non-adaptive counterpart. Our first result shows that, for submodular influence maximization, the adaptive greedy algorithm can perform up to a (1-1/e)-fraction worse than the non-adaptive greedy algorithm, and that this ratio is tight. More specifically, on one side we provide examples where the performance of the adaptive greedy algorithm is only a (1-1/e) fraction of the performance of the non-adaptive greedy algorithm in four settings: for both feedback models and both the independent cascade model and the linear threshold model. On the other side, we prove that in any submodular cascade, the adaptive greedy algorithm always outputs a (1-1/e)-approximation to the expected number of adoptions in the optimal non-adaptive seed choice. Our second result shows that, for the general submodular cascade model with full-adoption feedback, the adaptive greedy algorithm can outperform the non-adaptive greedy algorithm by an unbounded factor. Finally, we propose a risk-free variant of the adaptive greedy algorithm that always performs no worse than the non-adaptive greedy algorithm.

In many social systems in which individuals and organizations interact with each other, there can be no easy laws to govern the rules of the environment, and agents' payoffs are often influenced by other agents' actions. We examine such a social system in the setting of sponsored search auctions and tackle the search engine's dynamic pricing problem by combining the tools from both mechanism design and the AI domain. In this setting, the environment not only changes over time, but also behaves strategically. Over repeated interactions with bidders, the search engine can dynamically change the reserve prices and determine the optimal strategy that maximizes the profit. We first train a buyer behavior model, with a real bidding data set from a major search engine, that predicts bids given information disclosed by the search engine and the bidders' performance data from previous rounds. We then formulate the dynamic pricing problem as an MDP and apply a reinforcement-based algorithm that optimizes reserve prices over time. Experiments demonstrate that our model outperforms static optimization strategies including the ones that are currently in use as well as several other dynamic ones.

We consider the setting where players run the Hedge algorithm or its optimistic variant \cite{syrgkanis2015fast} to play an n-action game repeatedly for T rounds. 1) For two-player games, we show that the regret of optimistic Hedge decays at \tilde{O}( 1/T ^{5/6} ), improving the previous bound O(1/T^{3/4}) by \cite{syrgkanis2015fast}. 2) In contrast, we show that the convergence rate of vanilla Hedge is no better than \tilde{\Omega}(1/ \sqrt{T})}, addressing an open question posted in \cite{syrgkanis2015fast}. For general m-player games, we show that the swap regret of each player decays at rate \tilde{O}(m^{1/2} (n/T)^{3/4}) when they combine optimistic Hedge with the classical external-to-internal reduction of Blum and Mansour \cite{blum2007external}. The algorithm can also be modified to achieve the same rate against itself and a rate of \tilde{O}(\sqrt{n/T}) against adversaries. Via standard connections, our upper bounds also imply faster convergence to coarse correlated equilibria in two-player games and to correlated equilibria in multiplayer games.

Over the past decades, various theories and algorithms have been developed under the framework of Stackelberg games and part of these innovations have been fielded under the scenarios of national security defenses and wildlife protections. However, one of the remaining difficulties in the literature is that most of theoretical works assume full information of the payoff matrices, while in applications, the leader often has no prior knowledge about the follower’s payoff matrix, but may gain information about the follower’s utility function through repeated interactions. In this paper, we study the problem of learning the optimal leader strategy in Stackelberg (security) games and develop novel algorithms as well as new hardness results.

Vizing's celebrated theorem asserts that any graph of maximum degree Δ admits an edge coloring using at most Δ+1 colors. In contrast, Bar-Noy, Motwani and Naor showed over a quarter century ago that the trivial greedy algorithm, which uses 2Δ-1 colors, is optimal among online algorithms. Their lower bound has a caveat, however: it only applies to low-degree graphs, with Δ=O(log n), and they conjectured the existence of online algorithms using Δ(1+o(1)) colors for Δ=ω(log n). Progress towards resolving this conjecture was only made under stochastic arrivals (Aggarwal et al., FOCS'03 and Bahmani et al., SODA'10). We resolve the above conjecture for adversarial vertex arrivals in bipartite graphs, for which we present a (1+o(1))Δ-edge-coloring algorithm for Δ=ω(log n) known a priori. Surprisingly, if Δ is not known ahead of time, we show that no (e/(e-1) - Ω(1)) Δ-edge-coloring algorithm exists. We then provide an optimal, (e/(e-1)+o(1)) Δ-edge-coloring algorithm for unknown Δ=ω(log n). To obtain our results, we study a nonstandard fractional relaxation for edge coloring, for which we present optimal fractional online algorithms and a near-lossless online rounding scheme, yielding our optimal randomized algorithms.

We study the minimum-cost metric perfect matching problem under online i.i.d arrivals. We are given a fixed metric with a server at each of the points, and then requests arrive online, each drawn independently from a known probability distribution over the points. Each request has to be matched to a free server, with cost equal to the distance. The goal is to minimize the expected total cost of the matching. Such stochastic arrival models have been widely studied for the maximization variants of the online matching problem; however, the only known result for the minimization problem is a tight $O(\log n)$-competitiveness for the random-order arrival model. This is in contrast with the adversarial model, where an optimal competitive ratio of $O(\log n)$ has long been conjectured and remains a tantalizing open question. In this paper, we show improved results in the i.i.d arrival model. We show how the i.i.d model can be used to give substantially better algorithms: our main result is an $O((\log \log \log n)^2)$-competitive algorithm in this model. Along the way we give a $9$-competitive algorithm for the line and tree metrics. Both results imply a strict separation between the i.i.d model and the adversarial and random order models, both for general metrics and these much-studied metrics.

In this paper, we study the adaptivity gap of the influence maximization problem under the independent cascade model when full-adoption feedback is available. Our main results are to derive upper bounds on several families of well-studied influence graphs, including in-arborescences, out-arborescences and bipartite graphs. Especially, we prove that the adaptivity gap for the in-arborescences is between [e/(e-1), 2e/(e-1)], and for the out-arborescences the gap is between [e/(e-1), 2]. These are the first constant upper bounds in the full-adoption feedback model. Our analysis provides several novel ideas to tackle the correlated feedback appearing in adaptive stochastic optimization, which may be of independent interest.

We study the adaptive influence maximization problem with myopic feedback under the independent cascade model: one sequentially selects k nodes as seeds one by one from a social network, and each selected seed returns the immediate neighbors it activates as the feedback available for by later selections, and the goal is to maximize the expected number of total activated nodes, referred as the influence spread. We show that the adaptivity gap, the ratio between the optimal adaptive influence spread and the optimal non-adaptive influence spread, is at most 4 and at least e/(e-1), and the approximation ratios with respect to the optimal adaptive influence spread of both the non-adaptive greedy and adaptive greedy algorithms are at least \frac{1}{4}(1 - \frac{1}{e}) and at most \frac{e^2 + 1}{(e + 1)^2} < 1 - \frac{1}{e}. Moreover, the approximation ratio of the non-adaptive greedy algorithm is no worse than that of the adaptive greedy algorithm, when considering all graphs. Our result confirms a long-standing open conjecture of Golovin and Krause (2011) on the constant approximation ratio of adaptive greedy with myopic feedback, and it also suggests that adaptive greedy may not bring much benefit under myopic feedback.

Huang et al. (STOC 2018) introduced the fully online matching problem, a generalization of the classic online bipartite matching problem in that it allows all vertices to arrive online and considers general graphs. They showed that the ranking algorithm by Karp et al. (STOC 1990) is strictly better than 0.5-competitive and the problem is strictly harder than the online bipartite matching problem in that no algorithms can be (1 – 1/e)-competitive. This paper pins down two tight competitive ratios of classic algorithms for the fully online matching problem. For the fractional version of the problem, we show that a natural instantiation of the water-filling algorithm is 2 – ≈ 0.585-competitive, together with a matching hardness result. Interestingly, our hardness result applies to arbitrary algorithms in the edge-arrival models of the online matching problem, improving the state-of-art upper bound. For integral algorithms, we show a tight competitive ratio of ≈ 0.567 for the ranking algorithm on bipartite graphs, matching a hardness result by Huang et al. (STOC 2018).