Continual Prototype Evolution: Learning Online from Non-Stationary Data Streams

By Matthias De Lange, Tinne Tuytelaars. 2020

265 citations recorded by the source corpus.

Attaining prototypical features to represent class distributions is well established in representation learning. However, learning prototypes online from streaming data proves a challenging endeavor as they rapidly become outdated, caused by an ever-changing parameter space during the learning process. Additionally, continual learning does not assume the data stream to be stationary, typically resulting in catastrophic forgetting of previous knowledge. As a first, we introduce a system addressing both problems, where prototypes evolve continually in a shared latent space, enabling learning and prediction at any point in time. To facilitate learning, a novel objective function synchronizes the latent space with the continually evolving prototypes. In contrast to the major body of work in continual learning, data streams are processed in an online fashion without task information and can be highly imbalanced, for which we propose an efficient memory scheme. As an additional contribution, we propose the learner-evaluator framework that i) generalizes existing paradigms in continual learning, ii) introduces data incremental learning, and iii) models the bridge between continual learning and concept drift. We obtain state-of-the-art performance by a significant margin on eight benchmarks, including three highly imbalanced data streams. Code is publicly available. 1

Subfields: Artificial Intelligence, Meta & Continual Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Meta & Continual Learning

Citations in the corpus

A Comprehensive Survey of Continual Learning: Theory, Method and Application (2023, methodology): “More advanced strategies prioritize the replay of those training samples that are informative (InfoRS [398]), diversified in parameter gradients (GSS [17]) or previously-learned knowledge (ODDL [473]), balanced in class labels (CBRS [75], GDumb [340], CoPE [86]), and beneficial to latent decision bo”
Task-Free Continual Learning via Online Discrepancy Distance Learning (2022, positive): “For instance, memory-based approaches can store a few past samples from the data stream at each training time and replay them later during the training [8, 12].”
Meta-attention for ViT-backed Continual Learning (2022, background): “Existing continual learning methods can be broadly categorized into three schools: replay methods [3,6,19,36,37], regularization methods [21,23, 26, 27, 34, 50, 52] and mask methods [18, 30–32, 41].”
Natural continual learning: success is a journey, not (just) a destination (2021, methodology): “While we enforce this stringent condition in this paper, our approach may be easily combined with memory-based techniques such as coresets or generative replay [8, 10, 13, 31, 33, 35, 37, 41, 42, 44, 47].”
Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (2020, methodology): “This was explored for the tinyImageNet (Stanford, CS231N [128]) dataset in [64], [125], where the authors show that some methods report different results based on different semantics-based class orderings.”
Online Prototype Learning for Online Continual Learning (2023, unknown): “However, they are still inferior to OnPro.”
SDDGR: Stable Diffusion-Based Deep Generative Replay for Class Incremental Object Detection (2024, methodology): “In CIL, we can cluster the main methods into knowledge distillation [16, 33, 38, 61], and re-play [5–7, 12, 17, 22, 47, 50, 57, 58, 66, 67] in general.”
Class-Incremental Learning: A Survey (2023, methodology): “Reference § 3.1 Data Replay Direct Replay [26], [33], [34], [35], [36], [37], [38], [39], [40] Generative Replay [41], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52] § 3.2: Data Regularization [34], [53], [54], [55], [56], [57], [58] § 3.3 Dynamic Networks Neuron Expansion [45],…”
Computationally Budgeted Continual Learning: What Does Matter? (2023, methodology): “Lastly, balanced sampling [15, 16, 37] select samples such that an equal distribution of classes is selected for training.”
Saliency Guided Experience Packing for Replay in Continual Learning (2021, methodology): “(a) ACC(%) comparison in class-incremental learning setup, where results for CIFAR-100 and miniImageNet baselines are taken from [11] and [43] respectively.”
Bring Evanescent Representations to Life in Lifelong Class Incremental Learning (2022, methodology): “Most of the successful CIL methods use exemplars of old classes C old to rehearse past knowledge [1, 3, 4, 7, 8, 12, 18, 20, 25, 27, 32, 36].”
Deep Class-Incremental Learning: A Survey (2023, methodology): “1 Data Replay Direct Replay [35], [39], [40], [41], [42], [43], [44], [45], [46], [47]”
Steering Prototypes with Prompt-tuning for Rehearsal-free Continual Learning (2023, background): “Although existing methodologies attempt to alleviate the impact of semantic drift—either through the preservation and replay of prior samples [46, 12] or by compensating for drifts post-facto [66, 18]—they often necessitate significant memory overhead or become susceptible to cumulative errors as th”