pith. sign in

arxiv: 2211.16780 · v4 · submitted 2022-11-30 · 💻 cs.LG · cs.CV

An Optimal Transport-driven Approach for Cultivating Latent Space in Online Incremental Learning

Quyen Tran , Hai Nguyen , Hoang Phan , Quan Dao , Linh Ngo , Khoat Than , Dinh Phung , Dimitris Metaxas
show 1 more author
Trung Le
This is my paper

Pith reviewed 2026-05-24 10:15 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords online incremental learningoptimal transportmixture modellatent spacecatastrophic forgettingcontinual learningclass similarity estimation
0
0 comments X

The pith

An optimal transport mixture model evolves class centroids incrementally to handle multimodal data streams in online learning.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces MMOT, an online mixture model grounded in optimal transport theory, to represent classes in the latent space during continual data arrival. Centroids update with incoming samples rather than remaining fixed or limited to a single adaptive point, allowing better capture of complex, multimodal distributions within each class. A dynamic preservation mechanism is added to regulate the space and sustain separability across tasks. This setup is claimed to yield more accurate similarity estimates for new samples at inference time while reducing forgetting on benchmark datasets.

Core claim

We introduce an online Mixture Model learning framework grounded in Optimal Transport theory (MMOT), where centroids evolve incrementally with new data. This approach offers two main advantages: (i) it provides a more precise characterization of complex data streams, and (ii) it enables improved class similarity estimation for unseen samples during inference through MMOT-derived centroids. Furthermore, to strengthen representation learning and mitigate catastrophic forgetting, we design a Dynamic Preservation strategy that regulates the latent space and maintains class separability over time.

What carries the argument

MMOT: an optimal transport-grounded online mixture model whose centroids evolve incrementally with new data, paired with a dynamic preservation strategy that regulates the latent space.

If this is right

  • Centroids update incrementally to match the evolving distribution of each class.
  • MMOT-derived centroids improve similarity-based inference for samples from unseen tasks.
  • Dynamic preservation maintains separability and reduces catastrophic forgetting across sequential tasks.
  • The method outperforms single-adaptive-centroid and multiple-fixed-centroid baselines on standard benchmarks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The incremental centroid evolution may reduce reliance on replay buffers in other streaming classification settings.
  • The same transport-driven update rule could be tested on regression or density estimation tasks with shifting multimodal targets.
  • If the preservation strategy scales, it might support longer task sequences without explicit memory management.

Load-bearing premise

Evolving centroids via optimal transport can characterize multimodal class streams more precisely than fixed or single-centroid methods while the preservation rule keeps classes separable without replaying old samples.

What would settle it

A controlled experiment on a dataset engineered with known multimodal classes per label, comparing final accuracy and forgetting rates of MMOT against a fixed-centroid baseline under identical online arrival schedules.

Figures

Figures reproduced from arXiv: 2211.16780 by Dimitris Metaxas, Dinh Phung, Hai Nguyen, Hoang Phan, Khoat Than, Linh Ngo, Quan Dao, Quyen Tran, Trung Le.

Figure 1
Figure 1. Figure 1: The intuitions and motivations of our CLOT. Dynamic preservation inspires the classes in the old and new tasks to be more [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: t-SNE visualization on MNIST: Motivation of OT-MM. Left: the test latent representation of CoPE [8] with one centroid (i.e., visualized by digits) per class. Right: the test latent representation of our CLOT with four centroids per class (i.e., visualized by digits). We observe that there exists a shift between the test and train representations. Therefore, centroids learned on the training set might misma… view at source ↗
Figure 3
Figure 3. Figure 3: Average Accuracy by different number of centroids per [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Average accuracy through tasks. where ε > 0 is a small number, φ is the Kantorovich network, z˜ c = PK k=1 yk [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Features on latent space of our method (a and b) and CoPE (c). It can be observed that 4 centroids is better than 1 centroid. CLOT [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Average forgetting through tasks. Lower is better. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: t-SNE visualization of learned features and prototypes. The features of different classes are assigned different colors. The prototypes are located in the position of the red ”X” signs. MNIST CIFAR10 CIFAR100 0 20 40 60 80 100 Avg Accuracy 22.35 54.45 93.71 20.73 51.12 91.55 MNIST CIFAR10 CIFAR100 Dataset 0 20 40 60 80 100 Avg Forgetting 47.59 40.25 6.27 50.25 42.15 8.03 Adjust proto Not adjust proto [PIT… view at source ↗
Figure 8
Figure 8. Figure 8: Performance of CLOT when adjusting prototypes [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
read the original abstract

In online incremental learning, data continuously arrives with substantial distributional shifts, creating a significant challenge because previous samples have limited replay value when learning a new task. Prior research has typically relied on either a single adaptive centroid or multiple fixed centroids to represent each class in the latent space. However, such methods struggle when class data streams are inherently multimodal and require continual centroid updates. To overcome this, we introduce an online Mixture Model learning framework grounded in Optimal Transport theory (MMOT), where centroids evolve incrementally with new data. This approach offers two main advantages: (i) it provides a more precise characterization of complex data streams, and (ii) it enables improved class similarity estimation for unseen samples during inference through MMOT-derived centroids. Furthermore, to strengthen representation learning and mitigate catastrophic forgetting, we design a Dynamic Preservation strategy that regulates the latent space and maintains class separability over time. Experimental evaluations on benchmark datasets confirm the superior effectiveness of our proposed method.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript proposes an online Mixture Model learning framework grounded in Optimal Transport theory (MMOT) for incremental learning under distributional shifts. It claims that allowing centroids to evolve incrementally with new data yields a more precise characterization of inherently multimodal class streams than single-adaptive or multiple-fixed-centroid baselines, while also improving class-similarity estimation for unseen samples via the MMOT-derived centroids. A Dynamic Preservation strategy is introduced to regulate the latent space and preserve class separability, with benchmark experiments asserted to demonstrate superior effectiveness.

Significance. If the OT objective, incremental update rules, and preservation mechanism can be shown to deliver the claimed advantages without introducing hidden parameters or circularity, the work would supply a principled, transport-based mechanism for balancing plasticity and stability in continual representation learning, potentially benefiting applications that encounter multimodal data streams.

major comments (1)
  1. The submission consists solely of the abstract; no sections, equations, algorithm pseudocode, derivations of the MMOT objective, incremental centroid update rule, or experimental results are provided. Consequently, the two stated advantages and the preservation mechanism cannot be checked for internal consistency or empirical support.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their comments. We agree that the current submission is limited to the abstract and will revise to include the full manuscript details.

read point-by-point responses

  1. Referee: The submission consists solely of the abstract; no sections, equations, algorithm pseudocode, derivations of the MMOT objective, incremental centroid update rule, or experimental results are provided. Consequently, the two stated advantages and the preservation mechanism cannot be checked for internal consistency or empirical support.

    Authors: We acknowledge that the provided text consists only of the abstract. The complete manuscript containing sections, equations, algorithm pseudocode, derivations of the MMOT objective and incremental update rules, as well as experimental results, is available on arXiv:2211.16780. In the revised submission we will include all of these elements so that the claimed advantages and the Dynamic Preservation strategy can be verified for consistency and empirical support. revision: yes

Circularity Check

0 steps flagged

No derivation chain or equations present; circularity cannot be assessed

full rationale

The provided document consists exclusively of the abstract, which offers a high-level description of the MMOT framework and Dynamic Preservation strategy without any equations, update rules, objective functions, or claimed derivations. No load-bearing steps, predictions, or self-citations are exhibited that could reduce to inputs by construction. Per the analysis rules, circularity is only flagged when specific reductions can be quoted and exhibited from the paper's own content; absent any such content, the finding is no significant circularity (score 0) with an empty steps list.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.0 · 5688 in / 1130 out tokens · 29061 ms · 2026-05-24T10:15:58.980940+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 2 internal anchors

  1. [1]

    Online continual learning with maximal interfered retrieval

    Rahaf Aljundi, Eugene Belilovsky, Tinne Tuytelaars, Laurent Charlin, Massimo Caccia, Min Lin, and Lucas Page-Caccia. Online continual learning with maximal interfered retrieval. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alch´e-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Infor- mation Processing Systems 32, pages 11849–11860. Curran A...

    work page 2019

  2. [2]

    Incremental multi-domain learning with network latent tensor factorization

    Adrian Bulat, Jean Kossaifi, Georgios Tzimiropoulos, and Maja Pantic. Incremental multi-domain learning with network latent tensor factorization. In The Thirty-Fourth AAAI Confer- ence on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on Educational Ad- ...

    work page 2020

  3. [3]

    New insights on reducing abrupt representation change in online continual learning

    Lucas Caccia, Rahaf Aljundi, Nader Asadi, Tinne Tuyte- laars, Joelle Pineau, and Eugene Belilovsky. New insights on reducing abrupt representation change in online continual learning. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022. OpenReview.net, 2022. 3

    work page 2022

  4. [4]

    Efficient lifelong learning with a-GEM

    Arslan Chaudhry, Marc’Aurelio Ranzato, Marcus Rohrbach, and Mohamed Elhoseiny. Efficient lifelong learning with a-GEM. In International Conference on Learning Represen- tations, 2019. 2, 12

    work page 2019

  5. [5]

    Continual learning with tiny episodic memo- ries

    Arslan Chaudhry, Marcus Rohrbach, Mohamed Elhoseiny, Thalaiyasingam Ajanthan, Puneet K Dokania, Philip HS Torr, and M Ranzato. Continual learning with tiny episodic memo- ries. 2019. 2

    work page 2019

  6. [6]

    Online continual learning from imbalanced data

    Aristotelis Chrysakis and Marie-Francine Moens. Online continual learning from imbalanced data. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event , volume 119 of Proceedings of Machine Learning Research, pages 1952–

    work page 2020

  7. [7]

    A continual learning survey: Defying forgetting in classification tasks.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(7):3366–3385, 2021

    Matthias De Lange, Rahaf Aljundi, Marc Masana, Sarah Parisot, Xu Jia, Aleˇs Leonardis, Gregory Slabaugh, and Tinne Tuytelaars. A continual learning survey: Defying forgetting in classification tasks.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(7):3366–3385, 2021. 1

    work page 2021

  8. [8]

    Continual prototype evolution: Learning online from non-stationary data streams

    Matthias De Lange and Tinne Tuytelaars. Continual prototype evolution: Learning online from non-stationary data streams. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 8250–8259, October 2021. 2, 3, 5, 7, 12, 13

    work page 2021

  9. [9]

    Stochastic optimization for large-scale optimal trans- port

    Aude Genevay, Marco Cuturi, Gabriel Peyr ´e, and Francis Bach. Stochastic optimization for large-scale optimal trans- port. Advances in neural information processing systems, 29,

    work page

  10. [10]

    Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron C

    Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron C. Courville, and Yoshua Bengio. Generative adversarial nets. In Zoubin Ghahramani, Max Welling, Corinna Cortes, Neil D. Lawrence, and Kilian Q. Weinberger, editors, Advances in Neural Information Processing Systems 27: Annual Confer- ence on Neural Infor...

    work page 2014

  11. [11]

    Not just selection, but exploration: Online class-incremental contin- ual learning via dual view consistency

    Yanan Gu, Xu Yang, Kun Wei, and Cheng Deng. Not just selection, but exploration: Online class-incremental contin- ual learning via dual view consistency. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7442–7451, June 2022. 3, 7, 13

    work page 2022

  12. [12]

    Improved schemes for episodic memory-based lifelong learn- ing

    Yunhui Guo, Mingrui Liu, Tianbao Yang, and Tajana Rosing. Improved schemes for episodic memory-based lifelong learn- ing. Advances in Neural Information Processing Systems, 33,

    work page

  13. [13]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016, pages 770–778. IEEE Computer Society, 2016. 1

    work page 2016

  14. [14]

    Learning a unified classifier incrementally via rebalancing

    Saihui Hou, Xinyu Pan, Chen Change Loy, Zilei Wang, and Dahua Lin. Learning a unified classifier incrementally via rebalancing. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2019. 1, 3

    work page 2019

  15. [15]

    Categorical Reparameterization with Gumbel-Softmax

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144, 2016. 2, 5

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  16. [16]

    Continual learning of a mixed sequence of similar and dissimilar tasks

    Zixuan Ke, Bing Liu, and Xingchang Huang. Continual learning of a mixed sequence of similar and dissimilar tasks. 33, 2020. 1

    work page 2020

  17. [17]

    Overcoming catastrophic forgetting in neural net- works

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. Overcoming catastrophic forgetting in neural net- works. Proceedings of the national academy of sciences , 114(13):3521–3526, 2017. 3

    work page 2017

  18. [18]

    Learning from students: Online contrastive distillation net- work for general continual learning

    Jin Li, Zhong Ji, Gang Wang, Qiang Wang, and Feng Gao. Learning from students: Online contrastive distillation net- work for general continual learning. In Proceedings of the International Joint Conference on Artificial Intelligence, 2022. 7, 13

    work page 2022

  19. [19]

    Learning without forget- ting

    Zhizhong Li and Derek Hoiem. Learning without forget- ting. IEEE transactions on pattern analysis and machine intelligence, 40(12):2935–2947, 2017. 2

    work page 2017

  20. [20]

    Supervised contrastive replay: Revisiting the nearest class mean classifier in online class-incremental continual learning

    Zheda Mai, Ruiwen Li, Hyunwoo Kim, and Scott Sanner. Supervised contrastive replay: Revisiting the nearest class mean classifier in online class-incremental continual learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3589–3599, 2021. 2

    work page 2021

  21. [21]

    Au- tonomous vehicles: theoretical and practical challenges

    Margarita Mart ´ınez-D´ıaz and Francesc Soriguera. Au- tonomous vehicles: theoretical and practical challenges. Transportation Research Procedia, 33:275–282, 2018. 1

    work page 2018

  22. [22]

    Class- incremental learning: survey and performance evaluation on image classification

    Marc Masana, Xialei Liu, Bartlomiej Twardowski, Mikel Menta, Andrew D Bagdanov, and Joost van de Weijer. Class- incremental learning: survey and performance evaluation on image classification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022. 1

    work page 2022

  23. [23]

    Adaptive grasping for a small humanoid robot utilizing force- and electric current sensors

    Heinrich Mellmann, Marcus Scheunemann, and Oliver Stadie. Adaptive grasping for a small humanoid robot utilizing force- and electric current sensors. volume 1032, 09 2013. 1

    work page 2013

  24. [24]

    An efficient domain-incremental learning approach to drive in all weather conditions

    Muhammad Jehanzeb Mirza, Marc Masana, Horst Possegger, and Horst Bischof. An efficient domain-incremental learning approach to drive in all weather conditions. InIEEE/CVF Con- ference on Computer Vision and Pattern Recognition Work- shops, CVPR Workshops 2022, New Orleans, LA, USA, June 19-20, 2022, pages 3000–3010. IEEE, 2022. 1

    work page 2022

  25. [25]

    Linear mode connec- tivity in multitask and continual learning

    Seyed-Iman Mirzadeh, Mehrdad Farajtabar, Dilan G¨or¨ur, Raz- van Pascanu, and Hassan Ghasemzadeh. Linear mode connec- tivity in multitask and continual learning. In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021. 12

    work page 2021

  26. [26]

    Olsson, C.L

    L. Olsson, C.L. Nehaniv, and D. Polani. Sensor adaptation and development in robots by entropy maximization of sensory data. In 2005 International Symposium on Computational Intelligence in Robotics and Automation , pages 587–592,

    work page 2005

  27. [27]

    Learning to remember: A synaptic plasticity driven framework for continual learning

    Oleksiy Ostapenko, Mihai Puscas, Tassilo Klein, Patrick Jah- nichen, and Moin Nabi. Learning to remember: A synaptic plasticity driven framework for continual learning. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11321–11329, 2019. 3

    work page 2019

  28. [28]

    Gdumb: A simple approach that questions our progress in continual learning

    Ameya Prabhu, Philip Torr, and Puneet Dokania. Gdumb: A simple approach that questions our progress in continual learning. In The European Conference on Computer Vision (ECCV), August 2020. 7, 13

    work page 2020

  29. [29]

    Ameya Prabhu, Philip H. S. Torr, and Puneet K. Dokania. Gdumb: A simple approach that questions our progress in continual learning. In Andrea Vedaldi, Horst Bischof, Thomas Brox, and Jan-Michael Frahm, editors, Computer Vision - ECCV 2020 - 16th European Conference, Glasgow, UK, Au- gust 23-28, 2020, Proceedings, Part II , volume 12347 of Lecture Notes in...

    work page 2020

  30. [30]

    Girshick, and Ali Farhadi

    Joseph Redmon, Santosh Kumar Divvala, Ross B. Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. In 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016, pages 779–788. IEEE Computer Society, 2016. 1

    work page 2016

  31. [31]

    Gaussian mixture models

    Douglas A Reynolds. Gaussian mixture models. Encyclope- dia of biometrics, 741(659-663), 2009. 2

    work page 2009

  32. [32]

    Catastrophic forgetting, rehearsal and pseu- dorehearsal

    Anthony Robins. Catastrophic forgetting, rehearsal and pseu- dorehearsal. Connection Science, 7(2):123–146, 1995. 2

    work page 1995

  33. [33]

    Gradient projec- tion memory for continual learning

    Gobinda Saha, Isha Garg, and Kaushik Roy. Gradient projec- tion memory for continual learning. In International Confer- ence on Learning Representations, 2021. 1, 2

    work page 2021

  34. [34]

    Santambrogio

    F. Santambrogio. Optimal transport for applied mathemati- cians. Birk¨auser, 2015. 2, 3

    work page 2015

  35. [35]

    Online class- incremental continual learning with adversarial shapley value

    Dongsub Shim, Zheda Mai, Jihwan Jeong, Scott San- ner, Hyunwoo Kim, and Jongseong Jang. Online class- incremental continual learning with adversarial shapley value. In Thirty-Fifth AAAI Conference on Artificial Intelligence, AAAI 2021, Thirty-Third Conference on Innovative Applica- tions of Artificial Intelligence, IAAI 2021, The Eleventh Sym- posium on Edu...

    work page 2021

  36. [36]

    Online class- incremental continual learning with adversarial shapley value

    Dongsub Shim, Zheda Mai, Jihwan Jeong, Scott San- ner, Hyunwoo Kim, and Jongseong Jang. Online class- incremental continual learning with adversarial shapley value. In Proceedings of the AAAI Conference on Artificial Intelli- gence, volume 35, pages 9630–9638, 2021. 7, 13

    work page 2021

  37. [37]

    Continual learning with deep generative replay

    Hanul Shin, Jung Kwon Lee, Jaehong Kim, and Jiwon Kim. Continual learning with deep generative replay. Advances in neural information processing systems, 30, 2017. 2

    work page 2017

  38. [38]

    Improving and Understanding Variational Continual Learning

    Siddharth Swaroop, Cuong V . Nguyen, Thang D. Bui, and Richard E. Turner. Improving and understanding variational continual learning. arXiv:1905.02099 [cs, stat], 2019. 1

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  39. [39]

    A sur- vey on video streaming over multimedia networks using tcp

    Rahamathunnisa Usuff and Saravanan Ramakrishnan. A sur- vey on video streaming over multimedia networks using tcp. Journal of Theoretical and Applied Information Technology, 53:205–209, 07 2013. 1

    work page 2013

  40. [40]

    Optimal transport: old and new, volume 338

    C´edric Villani. Optimal transport: old and new, volume 338. Springer, 2009. 2, 3

    work page 2009

  41. [41]

    Random sampling with a reservoir

    Jeffrey Scott Vitter. Random sampling with a reservoir. ACM Trans. Math. Softw., 11(1):37–57, 1985. 7, 13

    work page 1985

  42. [42]

    Continual learning with hypernet- works

    Johannes V on Oswald, Christian Henning, Jo˜ao Sacramento, and Benjamin F Grewe. Continual learning with hypernet- works. arXiv preprint arXiv:1906.00695, 2019. 3

    work page arXiv 1906

  43. [43]

    A study of live video streaming system for mobile devices

    Jiushuang Wang, Weizhang Xu, and Jian Wang. A study of live video streaming system for mobile devices. In 2016 First IEEE International Conference on Computer Communication and the Internet (ICCCI), pages 157–160, 2016. 1

    work page 2016

  44. [44]

    Large scale in- cremental learning

    Yue Wu, Yinpeng Chen, Lijuan Wang, Yuancheng Ye, Zicheng Liu, Yandong Guo, and Yun Fu. Large scale in- cremental learning. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019, Long Beach, CA, USA, June 16-20, 2019, pages 374–382. Computer Vision Founda- tion / IEEE, 2019. 1, 3

    work page 2019

  45. [45]

    Lampert, Bernt Schiele, and Zeynep Akata

    Yongqin Xian, Christoph H. Lampert, Bernt Schiele, and Zeynep Akata. Zero-shot learning - A comprehensive evalua- tion of the good, the bad and the ugly. IEEE Trans. Pattern Anal. Mach. Intell., 41(9):2251–2265, 2019. 2

    work page 2019

  46. [46]

    Yi Xiao, Felipe Codevilla, Akhil Gurram, Onay Urfalioglu, and Antonio M. L´opez. Multimodal end-to-end autonomous driving. IEEE Trans. Intell. Transp. Syst. , 23(1):537–547,

    work page

  47. [47]

    General in- cremental learning with domain-aware categorical represen- tations

    Jiangwei Xie, Shipeng Yan, and Xuming He. General in- cremental learning with domain-aware categorical represen- tations. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2022, New Orleans, LA, USA, June 18-24, 2022, pages 14331–14340. IEEE, 2022. 1

    work page 2022

  48. [48]

    Bhattacharyya

    Jiacong Xu, Zixiang Xiong, and Shankar P. Bhattacharyya. Pidnet: A real-time semantic segmentation network inspired from pid controller, 2022. 1

    work page 2022

  49. [49]

    Sola: continual learning with second-order loss approximation

    Dong Yin, Mehrdad Farajtabar, and Ang Li. Sola: continual learning with second-order loss approximation. 2020. 3

    work page 2020

  50. [50]

    Online coreset selection for rehearsal-based con- tinual learning

    Jaehong Yoon, Divyam Madaan, Eunho Yang, and Sung Ju Hwang. Online coreset selection for rehearsal-based con- tinual learning. In International Conference on Learning Representations, 2022. 7, 13

    work page 2022

  51. [51]

    Contin- ual learning through synaptic intelligence

    Friedemann Zenke, Ben Poole, and Surya Ganguli. Contin- ual learning through synaptic intelligence. In International Conference on Machine Learning, pages 3987–3995. PMLR,

    work page

  52. [52]

    Maintaining discrimination and fairness in class incre- mental learning

    Bowen Zhao, Xi Xiao, Guojun Gan, Bin Zhang, and Shu-Tao Xia. Maintaining discrimination and fairness in class incre- mental learning. In 2020 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, CVPR 2020, Seattle, WA, USA, June 13-19, 2020, pages 13205–13214. Computer Vision Foundation / IEEE, 2020. 1, 3

    work page 2020

  53. [53]

    Implementation Detail We implement our proposed method and baselines on the same code base, based on the learner-evaluator framework proposed in [8]

    Supplementary Material 7.1. Implementation Detail We implement our proposed method and baselines on the same code base, based on the learner-evaluator framework proposed in [8]. Our code is available at https://github.com/tranquyenbk173/Streaming WSD GGM 7.1.1 Hyperparameters configuration We shared the same replay memory size (total samples of all classes...

    work page