Anomaly detection (AD) identifies the defect regions of a given image. Recent works have studied AD, focusing on learning AD without abnormal images, with long-tailed distributed training data, and using a unified model for all classes. In addition, online AD learning has also been explored. In this work, we expand in both directions to a realistic setting by considering the novel task of long-tailed online AD (LTOAD). We first identified that the offline state-of-the-art LTAD methods cannot be directly applied to the online setting. Specifically, LTAD is class-aware, requiring class labels that are not available in the online setting. To address this challenge, we propose a class-agnostic framework for LTAD and then adapt it to our online learning setting. Our method outperforms the SOTA baselines in most offline LTAD settings, including both the industrial manufacturing and the medical domain. In particular, we observe +4.63% image-AUROC on MVTec even compared to methods that have access to class labels and the number of classes. In the most challenging long-tailed online setting, we achieve +0.53% image-AUROC compared to baselines. Our LTOAD benchmark is released here: https://d
Deploying machine learning (ML) algorithms on mobile phones is bottlenecked by performance degradation under dynamic, real-world conditions that differ from the offline training conditions. While continual learning and adaptation are essential to mitigate this distributional shift, conventional online learning methods are often computationally prohibitive for resource-constrained devices. In this paper, we propose LightTune, a lightweight, backpropagation-free online fine-tuning framework with provable convergence guarantees. LightTune opportunistically refines ML models using live test-time data only when performance falls below a predefined threshold, ensuring minimal computational overhead and highly efficient responsiveness. As a practical demonstration, we integrate LightTune into a block error rate (BLER) prediction algorithm for 6G mobile systems. This integration enables the ML BLER prediction model to dynamically adapt to previously unseen channel conditions in real-time. Our extensive results show a substantial reduction in the average BLER prediction error of up to 48.8% with online fine-tuning. Furthermore, we leverage this BLER prediction algorithm for link adaptation
In this paper, we introduce a new problem, Online-MMSI, where the model must perform multimodal social interaction understanding (MMSI) using only historical information. Given a recorded video and a multi-party dialogue, the AI assistant is required to immediately identify the speaker's referent, which is critical for real-world human-AI interaction. Without access to future conversational context, both humans and models experience substantial performance degradation when moving from offline to online settings. To tackle the challenges, we propose Online-MMSI-VLM, a novel framework based on multimodal large language models. The core innovations of our approach lie in two components: (1) multi-party conversation forecasting, which predicts upcoming speaker turns and utterances in a coarse-to-fine manner; and (2) socially-aware visual prompting, which highlights salient social cues in each video frame using bounding boxes and body keypoints. Our model achieves state-of-the-art results on three tasks across two datasets, significantly outperforming the baseline and demonstrating the effectiveness of Online-MMSI-VLM. Project page: https://sampson-lee.github.io/online-mmsi-project-page
This paper proposes an algorithm for real-time learning without explicit feedback. The algorithm combines the ideas of semi-supervised learning on graphs and online learning. In particular, it iteratively builds a graphical representation of its world and updates it with observed examples. Labeled examples constitute the initial bias of the algorithm and are provided offline, and a stream of unlabeled examples is collected online to update this bias. We motivate the algorithm, discuss how to implement it efficiently, prove a regret bound on the quality of its solutions, and apply it to the problem of real-time face recognition. Our recognizer runs in real time, and achieves superior precision and recall on 3 challenging video datasets.
The construction of confidence intervals and hypothesis tests for functionals is a cornerstone of statistical inference. Traditionally, the most efficient procedures - such as the Wald interval or the Likelihood Ratio Test - require both a point estimator and a consistent estimate of its asymptotic variance. However, when estimators are derived from online or sequential algorithms, computational constraints often preclude multiple passes over the data, complicating variance estimation. In this article, we propose a computationally efficient, rate-optimal wrapper method (HulC) that wraps around any online algorithm to produce asymptotically valid confidence regions bypassing the need for explicit asymptotic variance estimation. The method is provably valid for any online algorithm that yields an asymptotically normal estimator. We evaluate the practical performance of the proposed method primarily using Stochastic Gradient Descent (SGD) with Polyak-Ruppert averaging. Furthermore, we provide extensive numerical simulations comparing the performance of our approach (HulC) when used with other online algorithms, including implicit-SGD and ROOT-SGD.
We study the algorithmic problem of finding large $γ$-balanced independent sets in dense random bipartite graphs; an independent set is $γ$-balanced if a $γ$ proportion of its vertices lie on one side of the bipartition. In the sparse regime, Perkins and Wang established tight bounds within the low-degree polynomial (LDP) framework, showing a factor-$1/(1-γ)$ statistical-computational gap via the Overlap Gap Property (OGP) framework tailored for stable algorithms. However, these techniques do not appear to extend to the dense setting. For the related large independent set problem in dense random graph, the best known algorithm is an online greedy procedure that is inherently unstable, and LDP algorithms are conjectured to fail even in the "easy" regime where greedy succeeds. We show that the largest $γ$-balanced independent set in dense random bipartite graphs has size $α:=\frac{\log_b n}{γ(1-γ)}$ whp, where $n$ is the size of each bipartition, $p$ is the edge probability, and $b=1/(1-p)$. We design an online algorithm that achieves $(1-ε)(1-γ)α$ whp for any $ε>0$. We complement this with a sharp lower bound, showing that no online algorithm can achieve $(1+ε)(1-γ)α$ with nonneg
Recently, a wide range of memory-efficient LLM training algorithms have gained substantial popularity. These methods leverage the low-rank structure of gradients to project optimizer states into a subspace using projection matrix found by singular value decomposition (SVD). However, convergence of these algorithms is highly dependent on the update rules of their projection matrix. In this work, we provide the \emph{first} convergence guarantee for arbitrary update rules of projection matrix. This guarantee is generally applicable to optimizers that can be analyzed with Hamiltonian Descent, including most common ones, such as LION, Adam. Inspired by our theoretical understanding, we propose Online Subspace Descent, a new family of subspace descent optimizer without SVD. Instead of updating the projection matrix with eigenvectors, Online Subspace Descent updates the projection matrix with online PCA. Online Subspace Descent is flexible and introduces only minimum overhead to training. We show that for the task of pretraining LLaMA models ranging from 60M to 7B parameters on the C4 dataset, Online Subspace Descent achieves lower perplexity and better downstream tasks performance than
This paper addresses the problem of reconstructing a scene online at the level of objects given an RGB-D video sequence. While current object-aware neural implicit representations hold promise, they are limited in online reconstruction efficiency and shape completion. Our main contributions to alleviate the above limitations are twofold. First, we propose a feature grid interpolation mechanism to continuously update grid-based object-centric neural implicit representations as new object parts are revealed. Second, we construct an object library with previously mapped objects in advance and leverage the corresponding shape priors to initialize geometric object models in new videos, subsequently completing them with novel views as well as synthesized past views to avoid losing original object details. Extensive experiments on synthetic environments from the Replica dataset, real-world ScanNet sequences and videos captured in our laboratory demonstrate that our approach outperforms state-of-the-art neural implicit models for this task in terms of reconstruction accuracy and completeness.
We investigate the geometric hitting set problem in the online setup for the range space $Σ=({\cal P},{\cal S})$, where the set $¶\subset\mathbb{R}^2$ is a collection of $n$ points and the set $\cal S$ is a family of geometric objects in $\mathbb{R}^2$. In the online setting, the geometric objects arrive one by one. Upon the arrival of an object, an online algorithm must maintain a valid hitting set by making an irreversible decision, i.e., once a point is added to the hitting set by the algorithm, it can not be deleted in the future. The objective of the geometric hitting set problem is to find a hitting set of the minimum cardinality. Even and Smorodinsky (Discret. Appl. Math., 2014) considered an online model (Model-I) in which the range space $Σ$ is known in advance, but the order of arrival of the input objects in $\cal S$ is unknown. They proposed online algorithms having optimal competitive ratios of $Θ(\log n)$ for intervals, half-planes and unit disks in $\mathbb{R}^2$. Whether such an algorithm exists for unit squares remained open for a long time. This paper considers an online model (Model-II) in which the entire range space $Σ$ is not known in advance. We only know the
Online machine learning (ML) is often used in self-adaptive systems to strengthen the adaptation mechanism and improve the system utility. Despite such benefits, applying online ML for self-adaptation can be challenging, and not many papers report its limitations. Recently, we experimented with applying online ML for self-adaptation of a smart farming scenario and we had faced several unexpected difficulties -- traps -- that, to our knowledge, are not discussed enough in the community. In this paper, we report our experience with these traps. Specifically, we discuss several traps that relate to the specification and online training of the ML-based estimators, their impact on self-adaptation, and the approach used to evaluate the estimators. Our overview of these traps provides a list of lessons learned, which can serve as guidance for other researchers and practitioners when applying online ML for self-adaptation.
Non-stationary online learning has drawn much attention in recent years. In particular, dynamic regret and adaptive regret are proposed as two principled performance measures for online convex optimization in non-stationary environments. To optimize them, a two-layer online ensemble is usually deployed due to the inherent uncertainty of non-stationarity, in which multiple base-learners are maintained and a meta-algorithm is employed to track the best one on the fly. However, the two-layer structure raises concerns about computational complexity -- such methods typically maintain $O(\log T)$ base-learners simultaneously for a $T$-round online game and thus perform multiple projections onto the feasible domain per round, which becomes the computational bottleneck when the domain is complicated. In this paper, we present efficient methods for optimizing dynamic regret and adaptive regret that reduce the number of projections per round from $O(\log T)$ to $1$. The proposed algorithms require only one gradient query and one function evaluation at each round. Our technique hinges on the reduction mechanism developed in parameter-free online learning and requires non-trivial modifications
This paper studies the adversarial-robustness of importance-sampling (aka sensitivity sampling); a useful algorithmic technique that samples elements with probabilities proportional to some measure of their importance. A streaming or online algorithm is called adversarially-robust if it succeeds with high probability on input streams that may change adaptively depending on previous algorithm outputs. Unfortunately, the dependence between stream elements breaks the analysis of most randomized algorithms, and in particular that of importance-sampling algorithms. Previously, Braverman et al. [NeurIPS 2021] suggested that streaming algorithms based on importance-sampling may be adversarially-robust; however, they proved it only for well-behaved inputs. We focus on the adversarial-robustness of online importance-sampling, a natural variant where sampling decisions are irrevocable and made as data arrives. Our main technical result shows that, given as input an adaptive stream of elements $x_1,\ldots,x_T\in \mathbb{R}_+$, online importance-sampling maintains a $(1\pmε)$-approximation of their sum while matching (up to lower order terms) the storage guarantees of the oblivious (non-adapti
We introduce a novel method for the rigorous quantitative evaluation of online algorithms that relaxes the "radical worst-case" perspective of classic competitive analysis. In contrast to prior work, our method, referred to as randomly infused advice (RIA), does not make any probabilistic assumptions about the input sequence and does not rely on the development of designated online algorithms. Rather, it can be applied to existing online randomized algorithms, introducing a means to evaluate their performance in scenarios that lie outside the radical worst-case regime. More concretely, an online algorithm ALG with RIA benefits from pieces of advice generated by an omniscient but not entirely reliable oracle. The crux of the new method is that the advice is provided to ALG by writing it into the buffer B from which ALG normally reads its random bits, hence allowing us to augment it through a very simple and non-intrusive interface. The (un)reliability of the oracle is captured via a parameter 0 {\le} α {\le} 1 that determines the probability (per round) that the advice is successfully infused by the oracle; if the advice is not infused, which occurs with probability 1 - α, then the
Optimizing ranking systems based on user interactions is a well-studied problem. State-of-the-art methods for optimizing ranking systems based on user interactions are divided into online approaches - that learn by directly interacting with users - and counterfactual approaches - that learn from historical interactions. Existing online methods are hindered without online interventions and thus should not be applied counterfactually. Conversely, counterfactual methods cannot directly benefit from online interventions. We propose a novel intervention-aware estimator for both counterfactual and online Learning to Rank (LTR). With the introduction of the intervention-aware estimator, we aim to bridge the online/counterfactual LTR division as it is shown to be highly effective in both online and counterfactual scenarios. The estimator corrects for the effect of position bias, trust bias, and item-selection bias by using corrections based on the behavior of the logging policy and on online interventions: changes to the logging policy made during the gathering of click data. Our experimental results, conducted in a semi-synthetic experimental setup, show that, unlike existing counterfactu
We study online classification of features into labels with general hypothesis classes. In our setting, true labels are determined by some function within the hypothesis class but are corrupted by unknown stochastic noise, and the features are generated adversarially. Predictions are made using observed noisy labels and noiseless features, while the performance is measured via minimax risk when comparing against true labels. The noise mechanism is modeled via a general noise kernel that specifies, for any individual data point, a set of distributions from which the actual noisy label distribution is chosen. We show that minimax risk is tightly characterized (up to a logarithmic factor of the hypothesis class size) by the Hellinger gap of the noisy label distributions induced by the kernel, independent of other properties such as the means and variances of the noise. Our main technique is based on a novel reduction to an online comparison scheme of two hypotheses, along with a new conditional version of Le Cam-Birgé testing suitable for online settings. Our work provides the first comprehensive characterization for noisy online classification with guarantees with respect to the grou
Online learning has turned out to be effective for improving tracking performance. However, it could be simply applied for classification branch, but still remains challenging to adapt to regression branch due to its complex design and intrinsic requirement for high-quality online samples. To tackle this issue, we present the fully convolutional online tracking framework, coined as FCOT, and focus on enabling online learning for both classification and regression branches by using a target filter based tracking paradigm. Our key contribution is to introduce an online regression model generator (RMG) for initializing weights of the target filter with online samples and then optimizing this target filter weights based on the groundtruth samples at the first frame. Based on the online RGM, we devise a simple anchor-free tracker (FCOT), composed of a feature backbone, an up-sampling decoder, a multi-scale classification branch, and a multi-scale regression branch. Thanks to the unique design of RMG, our FCOT can not only be more effective in handling target variation along temporal dimension thus generating more precise results, but also overcome the issue of error accumulation during
Class-incremental learning (CIL) aims to train a classification model while the number of classes increases phase-by-phase. An inherent challenge of CIL is the stability-plasticity tradeoff, i.e., CIL models should keep stable to retain old knowledge and keep plastic to absorb new knowledge. However, none of the existing CIL models can achieve the optimal tradeoff in different data-receiving settings--where typically the training-from-half (TFH) setting needs more stability, but the training-from-scratch (TFS) needs more plasticity. To this end, we design an online learning method that can adaptively optimize the tradeoff without knowing the setting as a priori. Specifically, we first introduce the key hyperparameters that influence the trade-off, e.g., knowledge distillation (KD) loss weights, learning rates, and classifier types. Then, we formulate the hyperparameter optimization process as an online Markov Decision Process (MDP) problem and propose a specific algorithm to solve it. We apply local estimated rewards and a classic bandit algorithm Exp3 to address the issues when applying online MDP methods to the CIL protocol. Our method consistently improves top-performing CIL met
Not forgetting old class knowledge is a key challenge for class-incremental learning (CIL) when the model continuously adapts to new classes. A common technique to address this is knowledge distillation (KD), which penalizes prediction inconsistencies between old and new models. Such prediction is made with almost new class data, as old class data is extremely scarce due to the strict memory limitation in CIL. In this paper, we take a deep dive into KD losses and find that "using new class data for KD" not only hinders the model adaption (for learning new classes) but also results in low efficiency for preserving old class knowledge. We address this by "using the placebos of old classes for KD", where the placebos are chosen from a free image stream, such as Google Images, in an automatical and economical fashion. To this end, we train an online placebo selection policy to quickly evaluate the quality of streaming images (good or bad placebos) and use only good ones for one-time feed-forward computation of KD. We formulate the policy training process as an online Markov Decision Process (MDP), and introduce an online learning algorithm to solve this MDP problem without causing much
In this book, I introduce the basic concepts of Online Learning through the modern view of Online Convex Optimization. Here, online learning refers to the framework of regret minimization under worst-case assumptions. I present first-order and second-order algorithms for online learning with convex losses, in Euclidean and non-Euclidean settings. All the algorithms are clearly presented as instantiation of Online Mirror Descent or Follow-The-Regularized-Leader and their variants. Particular attention is given to the issue of tuning the parameters of the algorithms and learning in unbounded domains, through adaptive and parameter-free online learning algorithms. Non-convex losses are addressed through convex surrogate losses and randomization. The bandit setting is also briefly discussed, touching on the problem of adversarial and stochastic multi-armed bandits. Finally, I also cover advanced topics, including black-box reductions, saddle-point optimization, sequential investment, and non-stationary forms of regret analysis. The book concludes with a selection of applications of online learning to domains far from it, such as generalization theory and concentration inequalities. I t
We initiate the study of two-sided online resource allocation with costly cancellations. Our focus is on edge-weighted online bipartite matching (and several of its extensions), where nodes arrive online and request offline resources. In contrast to the classic literature, any fraction of an offline resource that was preallocated to an earlier online node can be reclaimed, resulting in the loss of the previously allocated edge-weight plus an additional penalty equal to a non-negative constant factor $f$ times the edge-weight. Parameterizing the problem by the buyback factor $f$, our main result is the development of optimal competitive algorithms for \emph{all possible values} of $f$ through a novel primal-dual family of algorithms in the fractional (or equivalently, large capacity) setting, and establishing their optimality by deriving matching lower bounds. Interestingly, our results reveal a phase transition: for the small buyback regime ($f < \frac{e-2}{2}$), the optimal competitive ratio is $\frac{e}{e-(1+f)}$, and for the large buyback regime ($f \geq \frac{e-2}{2}$), the competitive ratio is $-W_{-1}\left(\frac{-1}{e(1+f)}\right)$, where $W_{-1}$ is the non-principal bran