NeurIPS 2024 论文列表

We provide a full characterization of the concept classes that are optimistically universally online learnable with {0, 1} labels. The notion of optimistically universal online learning was defined in [Hanneke, 2021] in order to understand learnability under minimal assumptions. In this paper, following the philosophy behind that work, we investigate two questions, namely, for every concept class: (1) What are the minimal assumptions on the data process admitting online learnability? (2) Is there a learning algorithm which succeeds under every data process satisfying the minimal assumptions? Such an algorithm is said to be optimistically universal for the given concept class. We resolve both of these questions for all concept classes, and moreover, as part of our solution we design general learning algorithms for each case. Finally, we extend these algorithms and results to the agnostic case, showing an equivalence between the minimal assumptions on the data process for learnability in the agnostic and realizable cases, for every concept class, as well as the equivalence of optimistically universal learnability.

We propose a novel approach to sequential Bayesian inference based on variational Bayes (VB). The key insight is that, in the online setting, we do not need to add the KL term to regularize to the prior (which comes from the posterior at the previous timestep); instead we can optimize just the expected log-likelihood, performing a single step of natural gradient descent starting at the prior predictive. We prove this method recovers exact Bayesian inference if the model is conjugate. We also show how to compute an efficient deterministic approximation to the VB objective, as well as our simplified objective, when the variational distribution is Gaussian or a sub-family, including the case of a diagonal plus low-rank precision matrix. We show empirically that our method outperforms other online VB methods in the non-conjugate setting, such as online learning for neural networks, especially when controlling for computational costs.

This paper considers the problem of online learning with non-convex loss functions in dynamic environments. Recently, Suggala and Netrapalli [2020] demonstrated that follow the perturbed leader (FTPL) can achieve optimal regret for non-convex losses, but their results are limited to static environments. In this research, we examine dynamic environments and choose \emph{dynamic regret} and \emph{adaptive regret} to measure the performance. First, we propose an algorithm named FTPL-D by restarting FTPL periodically and establish $O(T^\frac{2}{3}(V_T+1)^\frac{1}{3})$ dynamic regret with the prior knowledge of $V_T$, which is the variation of loss functions. In the case that $V_T$ is unknown, we run multiple FTPL-D with different restarting parameters as experts and use a meta-algorithm to track the best one on the fly. To address the challenge of non-convexity, we utilize randomized sampling in the process of tracking experts. Next, we present a novel algorithm called FTPL-A that dynamically maintains a group of FTPL experts and combines them with an advanced meta-algorithm to obtain $O(\sqrt{\tau\log{T}})$ adaptive regret for any interval of length $\tau$. Moreover, we demonstrate that FTPL-A also attains an $\tilde{O}(T^\frac{2}{3}(V_T+1)^\frac{1}{3})$ dynamic regret bound. Finally, we discuss the application to online constrained meta-learning and conduct experiments to verify the effectiveness of our methods.

We study the framework of a dynamic decision-making scenario with resource constraints. In this framework, an agent, whose target is to maximize the total reward under the initial inventory, selects an action in each round upon observing a random request, leading to a reward and resource consumptions that are further associated with an unknown random external factor. While previous research has already established an $\widetilde{O}(\sqrt{T})$ worst-case regret for this problem, this work offers two results that go beyond the worst-case perspective: one for the worst-case gap between benchmarks and another for logarithmic regret rates. We first show that an $\Omega(\sqrt{T})$ distance between the commonly used fluid benchmark and the online optimum is unavoidable when the former has a degenerate optimal solution. On the algorithmic side, we merge the re-solving heuristic with distribution estimation skills and propose an algorithm that achieves an $\widetilde{O}(1)$ regret as long as the fluid LP has a unique and non-degenerate solution. Furthermore, we prove that our algorithm maintains a near-optimal $\widetilde{O}(\sqrt{T})$ regret even in the worst cases and extend these results to the setting where the request and external factor are continuous. Regarding information structure, our regret results are obtained under two feedback models, respectively, where the algorithm accesses the external factor at the end of each round and at the end of a round only when a non-null action is executed.

The online reconstruction of dynamic scenes from multi-view streaming videos faces significant challenges in training, rendering and storage efficiency. Harnessing superior learning speed and real-time rendering capabilities, 3D Gaussian Splatting (3DGS) has recently demonstrated considerable potential in this field. However, 3DGS can be inefficient in terms of storage and prone to overfitting by excessively growing Gaussians, particularly with limited views. This paper proposes an efficient framework, dubbed HiCoM, with three key components. First, we construct a compact and robust initial 3DGS representation using a perturbation smoothing strategy. Next, we introduce a Hierarchical Coherent Motion mechanism that leverages the inherent non-uniform distribution and local consistency of 3D Gaussians to swiftly and accurately learn motions across frames. Finally, we continually refine the 3DGS with additional Gaussians, which are later merged into the initial 3DGS to maintain consistency with the evolving scene. To preserve a compact representation, an equivalent number of low-opacity Gaussians that minimally impact the representation are removed before processing subsequent frames. Extensive experiments conducted on two widely used datasets show that our framework improves learning efficiency of the state-of-the-art methods by about 20% and reduces the data storage by 85%, achieving competitive free-viewpoint video synthesis quality but with higher robustness and stability. Moreover, by parallel learning multiple frames simultaneously, our HiCoM decreases the average training wall time to <2 seconds per frame with negligible performance degradation, substantially boosting real-world applicability and responsiveness.

In this paper, we investigate the question of whether no-swap-regret dynamics have stronger convergence properties in repeated games than regular no-external-regret dynamics. We prove that in almost all symmetric zero-sum games under symmetric initializations of the agents, no-swap-regret dynamics in self-play are guaranteed to converge in a strong ``frequent-iterate'' sense to the Nash equilibrium: in all but a vanishing fraction of the rounds, the players must play a strategy profile close to a symmetric Nash equilibrium. Remarkably, relaxing any of these three constraints, i.e. by allowing either i) asymmetric initial conditions, or ii) an asymmetric game or iii) no-external regret dynamics suffices to destroy this result and lead to complex non-equilibrating or even chaotic behavior. In a dual type of result, we show that the power of no-swap-regret dynamics comes at a cost of imposing a time-asymmetry on its inputs. While no-external-regret dynamics can be completely determined by the cumulative reward vector received by each player, we show there does not exist any general no-swap-regret dynamics defined on the same state space. In fact, we prove that any no-swap-regret learning algorithm must play a time-asymmetric function over the set of previously observed rewards, ruling out any dynamics based on a symmetric function of the current set of rewards.

Continual learning (CL) empowers pre-trained vision-language (VL) models to efficiently adapt to a sequence of downstream tasks. However, these models often encounter challenges in retaining previously acquired skills due to parameter shifts and limited access to historical data. In response, recent efforts focus on devising specific frameworks and various replay strategies, striving for a typical learning-forgetting trade-off. Surprisingly, both our empirical research and theoretical analysis demonstrate that the stability of the model in consecutive zero-shot predictions serves as a reliable indicator of its anti-forgetting capabilities for previously learned tasks. Motivated by these insights, we develop a novel replay-free CL method named ZAF (Zero-shot Antidote to Forgetting), which preserves acquired knowledge through a zero-shot stability regularization applied to wild data in a plug-and-play manner. To enhance efficiency in adapting to new tasks and seamlessly access historical models, we introduce a parameter-efficient EMA-LoRA neural architecture based on the Exponential Moving Average (EMA). ZAF utilizes new data for low-rank adaptation (LoRA), complemented by a zero-shot antidote on wild data, effectively decoupling learning from forgetting. Our extensive experiments demonstrate ZAF's superior performance and robustness in pre-trained models across various continual VL concept learning tasks, achieving leads of up to 3.70\%, 4.82\%, and 4.38\%, along with at least a 10x acceleration in training speed on three benchmarks, respectively. Additionally, our zero-shot antidote significantly reduces forgetting in existing models by at least 6.37\%. Our code is available at https://github.com/Zi-Jian-Gao/Stabilizing-Zero-Shot-Prediction-ZAF.

We provide a technique for OLO that obtains regret $G\|w_\star\|\sqrt{T\log(\|w_\star\|G\sqrt{T})} + \|w_\star\|^2 + G^2$ on $G$-Lipschitz losses for any comparison point $w_\star$ without knowing either $G$ or $\|w_\star\|$. Importantly, this matches the optimal bound $G\|w_\star\|\sqrt{T}$ available with such knowledge (up to logarithmic factors), unless either $\|w_\star\|$ or $G$ is so large that even $G\|w_\star\|\sqrt{T}$ is roughly linear in $T$. Thus, at a high level it matches the optimal bound in all cases in which one can achieve sublinear regret.

Consider a hiring process with candidates coming from different universities. It is easy to order candidates with the same background, yet it can be challenging to compare them otherwise. The latter case requires additional costly assessments, leading to a potentially high total cost for the hiring organization. Given an assigned budget, what would be an optimal strategy to select the most qualified candidate? We model the above problem as a multicolor secretary problem, allowing comparisons between candidates from distinct groups at a fixed cost. Our study explores how the allocated budget enhances the success probability of online selection algorithms.

We study the interplay between local differential privacy (LDP) and robustness to Huber corruption and possibly heavy-tailed rewards in the context of multi-armed bandits (MABs). We consider two different practical settings: LDP-then-Corruption (LTC) where each user's locally private response might be further corrupted during the data collection process, and Corruption-then-LDP (CTL) where each user's raw data may be corrupted such that the LDP mechanism will only be applied to the corrupted data. To start with, we present the first tight characterization of the mean estimation error in high probability under both LTC and CTL settings. Leveraging this new result, we then present an almost tight characterization (up to log factor) of the minimax regret in online MABs and sub-optimality in offline MABs under both LTC and CTL settings, respectively. Our theoretical results in both settings are also corroborated by a set of systematic simulations. One key message in this paper is that LTC is a more difficult setting that leads to a worse performance guarantee compared to the CTL setting (in the minimax sense). Our sharp understanding of LTC and CTL also naturally allows us to give the first tight performance bounds for the most practical setting where corruption could happen both before and after the LDP mechanism. As an important by-product, we also give the first correct and tight regret bound for locally private and heavy-tailed online MABs, i.e., without Huber corruption, by identifying a fundamental flaw in the state-of-the-art.

Motivated by online display advertising, this work considers repeated second-price auctions, where agents sample their value from an unknown distribution with cumulative distribution function $F$. In each auction $t$, a decision-maker bound by limited observations selects $n_t$ agents from a coalition of $N$ to compete for a prize with $p$ other agents, aiming to maximize the cumulative reward of the coalition across all auctions. The problem is framed as an $N$-armed structured bandit, each number of player sent being an arm $n$, with expected reward $r(n)$ fully characterized by $F$ and $p+n$. We present two algorithms, Local-Greedy (LG) and Greedy-Grid (GG), both achieving *constant* problem-dependent regret. This relies on three key ingredients: **1.** an estimator of $r(n)$ from feedback collected from any arm $k$, **2.** concentration bounds of these estimates for $k$ within an estimation neighborhood of $n$ and **3.** the unimodality property of $r$ under standard assumptions on $F$. Additionally, GG exhibits problem-independent guarantees on top of best problem-dependent guarantees. However, by avoiding to rely on confidence intervals, LG practically outperforms GG, as well as standard unimodal bandit algorithms such as OSUB or multi-armed bandit algorithms.

This paper examines two extensions of multi-armed bandit problems: multi-armed bandits with expert advice and contextual linear bandits. For the former problem, multi-armed bandits with expert advice, the previously known best upper and lower bounds have been $O(\sqrt{KT \log \frac{N}{K} })$ and $\Omega( \sqrt{KT \frac{ \log N }{\log K }} )$, respectively. Here, $K$, $N$, and $T$ represent the numbers of arms, experts, and rounds, respectively. We provide a lower bound of $\Omega( \sqrt{KT \log \frac{N}{K}} )$ for the setup in which the player chooses an expert before observing the advices in each round. For the latter problem, contextual linear bandits, we provide an algorithm that achieves $O ( \sqrt{d T \log ( K \min\{ 1, \frac{S}{d} \} )} )$ together with a matching lower bound, where $d$ and $S$ represent the dimensionality of feature vectors and the size of the context space, respectively.

We study the batched best arm identification (BBAI) problem, where the learner's goal is to identify the best arm while switching the policy as less as possible. In particular, we aim to find the best arm with probability $1-\delta$ for some small constant $\delta>0$ while minimizing both the sample complexity (total number of arm pulls) and the batch complexity (total number of batches). We propose the three-batch best arm identification (Tri-BBAI) algorithm, which is the first batched algorithm that achieves the optimal sample complexity in the asymptotic setting (i.e., $\delta\rightarrow 0$) and runs in $3$ batches in expectation. Based on Tri-BBAI, we further propose the almost optimal batched best arm identification (Opt-BBAI) algorithm, which is the first algorithm that achieves the near-optimal sample and batch complexity in the non-asymptotic setting (i.e., $1/\delta$ is finite), while enjoying the same batch and sample complexity as Tri-BBAI when $\delta$ tends to zero. Moreover, in the non-asymptotic setting, the complexity of previous batch algorithms is usually conditioned on the event that the best arm is returned (with a probability of at least $1-\delta$), which is potentially unbounded in cases where a sub-optimal arm is returned. In contrast, the complexity of Opt-BBAI does not rely on such an event. This is achieved through a novel procedure that we design for checking whether the best arm is eliminated, which is of independent interest.

In online learning, a decision maker repeatedly selects one of a set of actions, with the goal of minimizing the overall loss incurred. Following the recent line of research on algorithms endowed with additional predictive features, we revisit this problem by allowing the decision maker to acquire additional information on the actions to be selected. In particular, we study the power of \emph{best-action queries}, which reveal beforehand the identity of the best action at a given time step. In practice, predictive features may be expensive, so we allow the decision maker to issue at most $k$ such queries. We establish tight bounds on the performance any algorithm can achieve when given access to $k$ best-action queries for different types of feedback models. In particular, we prove that in the full feedback model, $k$ queries are enough to achieve an optimal regret of $\Theta(\min\{\sqrt T, \frac{T}{k}\})$. This finding highlights the significant multiplicative advantage in the regret rate achievable with even a modest (sublinear) number $k \in \Omega(\sqrt{T})$ of queries. Additionally, we study the challenging setting in which the only available feedback is obtained during the time steps corresponding to the $k$ best-action queries. There, we provide a tight regret rate of $\Theta(\min\{\frac{T}{\sqrt k},\frac{T^2}{k^2}\})$, which improves over the standard $\Theta(\frac{T}{\sqrt k})$ regret rate for label efficient prediction for $k \in \Omega(T^{2/3})$.

Consider the domain of multiclass classification within the adversarial online setting. What is the price of relying on bandit feedback as opposed to full information? To what extent can an adaptive adversary amplify the loss compared to an oblivious one? To what extent can a randomized learner reduce the loss compared to a deterministic one? We study these questions in the mistake bound model and provide nearly tight answers. We demonstrate that the optimal mistake bound under bandit feedback is at most $O(k)$ times higher than the optimal mistake bound in the full information case, where $k$ represents the number of labels. This bound is tight and provides an answer to an open question previously posed and studied by Daniely and Helbertal ['13] and by Long ['17, '20], who focused on deterministic learners. Moreover, we present nearly optimal bounds of $\tilde{\Theta}(k)$ on the gap between randomized and deterministic learners, as well as between adaptive and oblivious adversaries in the bandit feedback setting. This stands in contrast to the full information scenario, where adaptive and oblivious adversaries are equivalent, and the gap in mistake bounds between randomized and deterministic learners is a constant multiplicative factor of $2$. In addition, our results imply that in some cases the optimal randomized mistake bound is approximately the square-root of its deterministic parallel. Previous results show that this is essentially the smallest it can get. Some of our results are proved via a reduction to prediction with expert advice under bandit feedback, a problem interesting on its own right. For this problem, we provide a randomized algorithm which is nearly optimal in some scenarios.

We consider estimation of a linear functional of the treatment effect from adaptively collected data. This problem finds a variety of applications including off-policy evaluation in contextual bandits, and estimation of the average treatment effect in causal inference. While a certain class of augmented inverse propensity weighting (AIPW) estimators enjoys desirable asymptotic properties including the semi-parametric efficiency, much less is known about their non-asymptotic theory with adaptively collected data. To fill in the gap, we first present generic upper bounds on the mean-squared error of the class of AIPW estimators that crucially depends on a sequentially weighted error between the treatment effect and its estimates. Motivated by this, we propose a general reduction scheme that allows one to produce a sequence of estimates for the treatment effect via online learning to minimize the sequentially weighted estimation error. To illustrate this, we provide three concrete instantiations in (1) the tabular case; (2) the case of linear function approximation; and (3) the case of general function approximation for the outcome model. We then provide a local minimax lower bound to show the instance-dependent optimality of the AIPW estimator using no-regret online learning algorithms.

We introduce the first algorithmic framework for Blackwell approachability on the sequence-form polytope, the class of convex polytopes capturing the strategies of players in extensive-form games (EFGs). This leads to a new class of regret-minimization algorithms that are stepsize-invariant, in the same sense as the Regret Matching and Regret Matching$^+$ algorithms for the simplex. Our modular framework can be combined with any existing regret minimizer over cones to compute a Nash equilibrium in two-player zero-sum EFGs with perfect recall, through the self-play framework. Leveraging predictive online mirror descent, we introduce *Predictive Treeplex Blackwell$^+$* (PTB$^+$), and show a $O(1/\sqrt{T})$ convergence rate to Nash equilibrium in self-play. We then show how to stabilize PTB$^+$ with a stepsize, resulting in an algorithm with a state-of-the-art $O(1/T)$ convergence rate. We provide an extensive set of experiments to compare our framework with several algorithmic benchmarks, including CFR$^+$ and its predictive variant, and we highlight interesting connections between practical performance and the stepsize-dependence or stepsize-invariance properties of classical algorithms.

We present Saliency-driven Experience Replay - SER - a biologically-plausible approach based on replicating human visual saliency to enhance classification models in continual learning settings. Inspired by neurophysiological evidence that the primary visual cortex does not contribute to object manifold untangling for categorization and that primordial saliency biases are still embedded in the modern brain, we propose to employ auxiliary saliency prediction features as a modulation signal to drive and stabilize the learning of a sequence of non-i.i.d. classification tasks. Experimental results confirm that SER effectively enhances the performance (in some cases up to about twenty percent points) of state-of-the-art continual learning methods, both in class-incremental and task-incremental settings. Moreover, we show that saliency-based modulation successfully encourages the learning of features that are more robust to the presence of spurious features and to adversarial attacks than baseline methods. Code is available at: https://github.com/perceivelab/SER

Posterior sampling in contextual bandits with a Gaussian prior can be implemented exactly or approximately using the Laplace approximation. The Gaussian prior is computationally efficient but it cannot describe complex distributions. In this work, we propose approximate posterior sampling algorithms for contextual bandits with a diffusion model prior. The key idea is to sample from a chain of approximate conditional posteriors, one for each stage of the reverse diffusion process, which are obtained by the Laplace approximation. Our approximations are motivated by posterior sampling with a Gaussian prior, and inherit its simplicity and efficiency. They are asymptotically consistent and perform well empirically on a variety of contextual bandit problems.