NeurIPS 2024 Papers

To address the uncertainty in function types, recent progress in online convex optimization (OCO) has spurred the development of universal algorithms that simultaneously attain minimax rates for multiple types of convex functions. However, for a $T$-round online problem, state-of-the-art methods typically conduct $O(\log T)$ projections onto the domain in each round, a process potentially time-consuming with complicated feasible sets. In this paper, inspired by the black-box reduction of Cutkosky and Orabona [2018], we employ a surrogate loss defined over simpler domains to develop universal OCO algorithms that only require $1$ projection. Embracing the framework of prediction with expert advice, we maintain a set of experts for each type of functions and aggregate their predictions via a meta-algorithm. The crux of our approach lies in a uniquely designed expert-loss for strongly convex functions, stemming from an innovative decomposition of the regret into the meta-regret and the expert-regret. Our analysis sheds new light on the surrogate loss, facilitating a rigorous examination of the discrepancy between the regret of the original loss and that of the surrogate loss, and carefully controlling meta-regret under the strong convexity condition. With only $1$ projection per round, we establish optimal regret bounds for general convex, exponentially concave, and strongly convex functions simultaneously. Furthermore, we enhance the expert-loss to exploit the smoothness property, and demonstrate that our algorithm can attain small-loss regret for multiple types of convex and smooth functions.

In linear bandits, how can a learner effectively learn when facing corrupted rewards? While significant work has explored this question, a holistic understanding across different adversarial models and corruption measures is lacking, as is a full characterization of the minimax regret bounds. In this work, we compare two types of corruptions commonly considered: strong corruption, where the corruption level depends on the learner’s chosen action, and weak corruption, where the corruption level does not depend on the learner’s chosen action. We provide a unified framework to analyze these corruptions. For stochastic linear bandits, we fully characterize the gap between the minimax regret under strong and weak corruptions. We also initiate the study of corrupted adversarial linear bandits, obtaining upper and lower bounds with matching dependencies on the corruption level. Next, we reveal a connection between corruption-robust learning and learning with gap-dependent misspecification—a setting first studied by Liu et al. (2023a), where the misspecification level of an action or policy is proportional to its suboptimality. We present a general reduction that enables any corruption-robust algorithm to handle gap-dependent misspecification. This allows us to recover the results of Liu et al. (2023a) in a black-box manner and significantly generalize them to settings like linear MDPs, yielding the first results for gap-dependent misspecification in reinforcement learning. However, this general reduction does not attain the optimal rate for gap-dependent misspecification. Motivated by this, we develop a specialized algorithm that achieves optimal bounds for gap-dependent misspecification in linear bandits, thus answering an open question posed by Liu et al. (2023a).

Bandits with preference feedback present a powerful tool for optimizing unknown target functions when only pairwise comparisons are allowed instead of direct value queries. This model allows for incorporating human feedback into online inference and optimization and has been employed in systems for tuning large language models. The problem is fairly understood in toy settings with linear target functions or over finite small domains that limits practical interest. Taking the next step, we consider infinite domains and kernelized rewards. In this setting, selecting a pair of actions is quite challenging and requires balancing exploration and exploitation at two levels: within the pair, and along the iterations of the algorithm. We propose MaxMinLCB, which emulates this trade-off as a zero-sum Stackelberg game and chooses action pairs that are informative and have favorable reward values. MaxMinLCB consistently outperforms algorithms in the literature and satisfies an anytime-valid rate-optimal regret guarantee. This is owed to our novel preference-based confidence sequences for kernelized logistic estimators, which are of independent interest.

We consider online model selection with decentralized data over $M$ clients, and study the necessity of collaboration among clients. Previous work proposed various federated algorithms without demonstrating their necessity, while we answer the question from a novel perspective of computational constraints. We prove lower bounds on the regret, and propose a federated algorithm and analyze the upper bound. Our results show (i) collaboration is unnecessary in the absence of computational constraints on clients; (ii) collaboration is necessary if the computational cost on each client is limited to $o(K)$, where $K$ is the number of candidate hypothesis spaces. We clarify the unnecessary nature of collaboration in previous federated algorithms for distributed online multi-kernel learning, and improve the regret bounds at a smaller computational and communication cost. Our algorithm relies on three new techniques including an improved Bernstein's inequality for martingale, a federated online mirror descent framework, and decoupling model selection and prediction, which might be of independent interest.

Training offline RL models using visual inputs poses two significant challenges, *i.e.*, the overfitting problem in representation learning and the overestimation bias for expected future rewards. Recent work has attempted to alleviate the overestimation bias by encouraging conservative behaviors. This paper, in contrast, tries to build more flexible constraints for value estimation without impeding the exploration of potential advantages. The key idea is to leverage off-the-shelf RL simulators, which can be easily interacted with in an online manner, as the “*test bed*” for offline policies. To enable effective online-to-offline knowledge transfer, we introduce CoWorld, a model-based RL approach that mitigates cross-domain discrepancies in state and reward spaces. Experimental results demonstrate the effectiveness of CoWorld, outperforming existing RL approaches by large margins.

We study the fundamental problem of sequential probability assignment, also known as online learning with logarithmic loss, with respect to an arbitrary, possibly nonparametric hypothesis class. Our goal is to obtain a complexity measure for the hypothesis class that characterizes the minimax regret and to determine a general, minimax optimal algorithm. Notably, the sequential $\ell_{\infty}$ entropy, extensively studied in the literature (Rakhlin and Sridharan, 2015, Bilodeau et al., 2020, Wu et al., 2023), was shown to not characterize minimax regret in general. Inspired by the seminal work of Shtarkov (1987) and Rakhlin, Sridharan, and Tewari (2010), we introduce a novel complexity measure, the \emph{contextual Shtarkov sum}, corresponding to the Shtarkov sum after projection onto a multiary context tree, and show that the worst case log contextual Shtarkov sum equals the minimax regret. Using the contextual Shtarkov sum, we derive the minimax optimal strategy, dubbed \emph{contextual Normalized Maximum Likelihood} (cNML). Our results hold for sequential experts, beyond binary labels, which are settings rarely considered in prior work. To illustrate the utility of this characterization, we provide a short proof of a new regret upper bound in terms of sequential $\ell_{\infty}$ entropy, unifying and sharpening state-of-the-art bounds by Bilodeau et al. (2020) and Wu et al. (2023).

Causal knowledge about the relationships among decision variables and a reward variable in a bandit setting can accelerate the learning of an optimal decision. Current works often assume the causal graph is known, which may not always be available a priori. Motivated by this challenge, we focus on the causal bandit problem in scenarios where the underlying causal graph is unknown and may include latent confounders. While intervention on the parents of the reward node is optimal in the absence of latent confounders, this is not necessarily the case in general. Instead, one must consider a set of possibly optimal arms/interventions, each being a special subset of the ancestors of the reward node, making causal discovery beyond the parents of the reward node essential. For regret minimization, we identify that discovering the full causal structure is unnecessary; however, no existing work provides the necessary and sufficient components of the causal graph. We formally characterize the set of necessary and sufficient latent confounders one needs to detect or learn to ensure that all possibly optimal arms are identified correctly. We also propose a randomized algorithm for learning the causal graph with a limited number of samples, providing a sample complexity guarantee for any desired confidence level. In the causal bandit setup, we propose a two-stage approach. In the first stage, we learn the induced subgraph on ancestors of the reward, along with a necessary and sufficient subset of latent confounders, to construct the set of possibly optimal arms. We show that for our proposed algorithm, the number of intervention samples required to learn the set of possibly optimal arms scales polynomially with respect to the number of nodes. The second phase involves the application of a standard bandit algorithm, such as the UCB algorithm. We also establish a regret bound for our two-phase approach, which is sublinear in the number of rounds.

We consider the classic problem of online convex optimisation. Whereas the notion of static regret is relevant for stationary problems, the notion of switching regret is more appropriate for non-stationary problems. A switching regret is defined relative to any segmentation of the trial sequence, and is equal to the sum of the static regrets of each segment. In this paper we show that, perhaps surprisingly, we can achieve the asymptotically optimal switching regret on every possible segmentation simultaneously. Our algorithm for doing so is very efficient: having a space and per-trial time complexity that is logarithmic in the time-horizon. Our algorithm also obtains novel bounds on its dynamic regret: being adaptive to variations in the rate of change of the comparator sequence.

We consider contextual bandits with graph feedback, a class of interactive learning problems with richer structures than vanilla contextual bandits, where taking an action reveals the rewards for all neighboring actions in the feedback graph under all contexts. Unlike the multi-armed bandits setting where a growing literature has painted a near-complete understanding of graph feedback, much remains unexplored in the contextual bandits counterpart. In this paper, we make inroads into this inquiry by establishing a regret lower bound $\Omega(\sqrt{\beta_M(G) T})$, where $M$ is the number of contexts, $G$ is the feedback graph, and $\beta_M(G)$ is our proposed graph-theoretic quantity that characterizes the fundamental learning limit for this class of problems. Interestingly, $\beta_M(G)$ interpolates between $\alpha(G)$ (the independence number of the graph) and $\mathsf{m}(G)$ (the maximum acyclic subgraph (MAS) number of the graph) as the number of contexts $M$ varies. We also provide algorithms that achieve near-optimal regret for important classes of context sequences and/or feedback graphs, such as transitively closed graphs that find applications in auctions and inventory control. In particular, with many contexts, our results show that the MAS number essentially characterizes the statistical complexity for contextual bandits, as opposed to the independence number in multi-armed bandits.

The classical theory of statistical estimation aims to estimate a parameter of interest under data generated from a fixed design (''offline estimation''), while the contemporary theory of online learning provides algorithms for estimation under adaptively chosen covariates (''online estimation''). Motivated by connections between estimation and interactive decision making, we ask: is it possible to convert offline estimation algorithms into online estimation algorithms in a black-box fashion? We investigate this question from an information-theoretic perspective by introducing a new framework, Oracle-Efficient Online Estimation (OEOE), where the learner can only interact with the data stream indirectly through a sequence of offline estimators produced by a black-box algorithm operating on the stream. Our main results settle the statistical and computational complexity of online estimation in this framework. $\bullet$ Statistical complexity. We show that information-theoretically, there exist algorithms that achieve near-optimal online estimation error via black-box offline estimation oracles, and give a nearly-tight characterization for minimax rates in the OEOE framework. $\bullet$ Computational complexity. We show that the guarantees above cannot be achieved in a computationally efficient fashion in general, but give a refined characterization for the special case of conditional density estimation: computationally efficient online estimation via black-box offline estimation is possible whenever it is possible via unrestricted algorithms. Finally, we apply our results to give offline oracle-efficient algorithms for interactive decision making.

We study the performance guarantees of exploration-free greedy algorithms for the linear contextual bandit problem. We introduce a novel condition, named the \textit{Local Anti-Concentration} (LAC) condition, which enables a greedy bandit algorithm to achieve provable efficiency. We show that the LAC condition is satisfied by a broad class of distributions, including Gaussian, exponential, uniform, Cauchy, and Student's~$t$ distributions, along with other exponential family distributions and their truncated variants. This significantly expands the class of distributions under which greedy algorithms can perform efficiently. Under our proposed LAC condition, we prove that the cumulative expected regret of the greedy algorithm for the linear contextual bandit is bounded by $\mathcal{O}(\operatorname{poly} \log T)$. Our results establish the widest range of distributions known to date that allow a sublinear regret bound for greedy algorithms, further achieving a sharp poly-logarithmic regret.

Algorithms for playing in Stackelberg games have been deployed in real-world domains including airport security, anti-poaching efforts, and cyber-crime prevention. However, these algorithms often fail to take into consideration the additional information available to each player (e.g. traffic patterns, weather conditions, network congestion), a salient feature of reality which may significantly affect both players' optimal strategies. We formalize such settings as Stackelberg games with side information, in which both players observe an external context before playing. The leader commits to a (context-dependent) strategy, and the follower best-responds to both the leader's strategy and the context. We focus on the online setting in which a sequence of followers arrive over time, and the context may change from round-to-round. In sharp contrast to the non-contextual version, we show that it is impossible for the leader to achieve good performance (measured by regret) in the full adversarial setting. Motivated by our impossibility result, we show that no-regret learning is possible in two natural relaxations: the setting in which the sequence of followers is chosen stochastically and the sequence of contexts is adversarial, and the setting in which the sequence of contexts is stochastic and the sequence of followers is chosen by an adversary.

Online Class Incremental Learning (OCIL) aims to train models incrementally, where data arrive in mini-batches, and previous data are not accessible. A major challenge in OCIL is Catastrophic Forgetting, i.e., the loss of previously learned knowledge. Among existing baselines, replay-based methods show competitive results but requires extra memory for storing exemplars, while exemplar-free (i.e., data need not be stored for replay in production) methods are resource friendly but often lack accuracy. In this paper, we propose an exemplar-free approach—Forward-only Online Analytic Learning (F-OAL). Unlike traditional methods, F-OAL does not rely on back-propagation and is forward-only, significantly reducing memory usage and computational time. Cooperating with a pre-trained frozen encoder with Feature Fusion, F-OAL only needs to update a linear classifier by recursive least square. This approach simultaneously achieves high accuracy and low resource consumption. Extensive experiments on bench mark datasets demonstrate F-OAL’s robust performance in OCIL scenarios. Code is available at: https://github.com/liuyuchen-cz/F-OAL

A natural strategy for continual learning is to weigh a Bayesian ensemble of fixed functions. This suggests that if a (single) neural network could be interpreted as an ensemble, one could design effective algorithms that learn without forgetting. To realize this possibility, we observe that a neural network classifier with N parameters can be interpreted as a weighted ensemble of N classifiers, and that in the lazy regime limit these classifiers are fixed throughout learning. We call these classifiers the *neural tangent experts* and show they output valid probability distributions over the labels. We then derive the likelihood and posterior probability of each expert given past data. Surprisingly, the posterior updates for these experts are equivalent to a scaled and projected form of stochastic gradient descent (SGD) over the network weights. Away from the lazy regime, networks can be seen as ensembles of adaptive experts which improve over time. These results offer a new interpretation of neural networks as Bayesian ensembles of experts, providing a principled framework for understanding and mitigating catastrophic forgetting in continual learning settings.

We provide a new understanding of the stochastic gradient bandit algorithm by showing that it converges to a globally optimal policy almost surely using \emph{any} constant learning rate. This result demonstrates that the stochastic gradient algorithm continues to balance exploration and exploitation appropriately even in scenarios where standard smoothness and noise control assumptions break down. The proofs are based on novel findings about action sampling rates and the relationship between cumulative progress and noise, and extend the current understanding of how simple stochastic gradient methods behave in bandit settings.

We consider the reinforcement learning problem for the constrained Markov decision process (CMDP), which plays a central role in satisfying safety or resource constraints in sequential learning and decision-making. In this problem, we are given finite resources and a MDP with unknown transition probabilities. At each stage, we take an action, collecting a reward and consuming some resources, all assumed to be unknown and need to be learned over time. In this work, we take the first step towards deriving optimal problem-dependent guarantees for the CMDP problems. We derive a logarithmic regret bound, which translates into a $O(\frac{1}{\Delta\cdot\epsilon}\cdot\log^2(1/\epsilon))$ sample complexity bound, with $\Delta$ being a problem-dependent parameter, yet independent of $\epsilon$. Our sample complexity bound improves upon the state-of-art $O(1/\epsilon^2)$ sample complexity for CMDP problems established in the previous literature, in terms of the dependency on $\epsilon$. To achieve this advance, we develop a new framework for analyzing CMDP problems. To be specific, our algorithm operates in the primal space and we resolve the primal LP for the CMDP problem at each period in an online manner, with \textit{adaptive} remaining resource capacities. The key elements of our algorithm are: i) a characterization of the instance hardness via LP basis, ii) an eliminating procedure that identifies one optimal basis of the primal LP, and; iii) a resolving procedure that is adaptive to the remaining resources and sticks to the characterized optimal basis.

Online continual learning (OCL) requires the models to learn from constant, endless streams of data. While significant efforts have been made in this field, most were focused on mitigating the \textit{catastrophic forgetting} issue to achieve better classification ability, at the cost of a much heavier training workload. They overlooked that in real-world scenarios, e.g., in high-speed data stream environments, data do not pause to accommodate slow models. In this paper, we emphasize that \textit{model throughput}-- defined as the maximum number of training samples that a model can process within a unit of time -- is equally important. It directly limits how much data a model can utilize and presents a challenging dilemma for current methods. With this understanding, we revisit key challenges in OCL from both empirical and theoretical perspectives, highlighting two critical issues beyond the well-documented catastrophic forgetting: (\romannumeral1) Model's ignorance: the single-pass nature of OCL challenges models to learn effective features within constrained training time and storage capacity, leading to a trade-off between effective learning and model throughput; (\romannumeral2) Model's myopia: the local learning nature of OCL on the current task leads the model to adopt overly simplified, task-specific features and \textit{excessively sparse classifier}, resulting in the gap between the optimal solution for the current task and the global objective. To tackle these issues, we propose the Non-sparse Classifier Evolution framework (NsCE) to facilitate effective global discriminative feature learning with minimal time cost. NsCE integrates non-sparse maximum separation regularization and targeted experience replay techniques with the help of pre-trained models, enabling rapid acquisition of new globally discriminative features. Extensive experiments demonstrate the substantial improvements of our framework in performance, throughput and real-world practicality.

In Economics, the concept of externality refers to any indirect effect resulting from an interaction between players and affecting a third party without compensation. Most of the models within which externality has been studied assume that agents have perfect knowledge of their environment and preferences. This is a major hindrance to the practical implementation of many proposed solutions. To adress this issue, we consider a two-players bandit game setting where the actions of one of the player affect the other one. Building upon this setup, we extend the Coase theorem [Coase, 2013], which suggests that the optimal approach for maximizing the social welfare in the presence of externality is to establish property rights, i.e., enabling transfers and bargaining between the players. Nonetheless, this fundamental result relies on the assumption that bargainers possess perfect knowledge of the underlying game. We first demonstrate that in the absence of property rights in the considered online scenario, the social welfare breaks down. We then provide a policy for the players, which allows them to learn a bargaining strategy which maximizes the total welfare, recovering the Coase theorem under uncertainty.

Adapting to dynamic data distributions is a practical yet challenging task. One effective strategy is to use a model ensemble, which leverages the diverse expertise of different models to transfer knowledge to evolving data distributions. However, this approach faces difficulties when the dynamic test distribution is available only in small batches and without access to the original source data. To address the challenge of adapting to dynamic distributions in such practical settings, we propose continual multi-source adaptation to dynamic distributions (CONTRAST), a novel method that optimally combines multiple source models to adapt to the dynamic test data. CONTRAST has two distinguishing features. First, it efficiently computes the optimal combination weights to combine the source models to adapt to the test data distribution continuously as a function of time. Second, it identifies which of the source model parameters to update so that only the model which is most correlated to the target data is adapted, leaving the less correlated ones untouched; this mitigates the issue of ``forgetting" the source model parameters by focusing only on the source model that exhibits the strongest correlation with the test batch distribution. Through theoretical analysis we show that the proposed method is able to optimally combine the source models and prioritize updates to the model least prone to forgetting. Experimental analysis on diverse datasets demonstrates that the combination of multiple source models does at least as well as the best source (with hindsight knowledge), and performance does not degrade as the test data distribution changes over time (robust to forgetting).