Filters








1,856 Hits in 6.7 sec

Semantic Drift Compensation for Class-Incremental Learning [article]

Lu Yu, Bartłomiej Twardowski, Xialei Liu, Luis Herranz, Kai Wang, Yongmei Cheng, Shangling Jui, Joost van de Weijer
2020 arXiv   pre-print
In addition, we propose a new method to estimate the drift, called semantic drift, of features and compensate for it without the need of any exemplars.  ...  Class-incremental learning of deep networks sequentially increases the number of classes to be classified.  ...  Semantic Drift Compensation Embeddings suffer from drift when learned in a sequential manner.  ... 
arXiv:2004.00440v1 fatcat:fw3e7rh5njgfddkvdtxsyzihse

Incremental Prototype Prompt-tuning with Pre-trained Representation for Class Incremental Learning [article]

Jieren Deng and Jianhua Hu and Haojian Zhang and Yunkuan Wang
2022 arXiv   pre-print
In detail, we incrementally prompt-tune category prototypes for classification and example prototypes to compensate for semantic drift, the problem caused by learning bias at different phases.  ...  Class incremental learning has attracted much attention, but most existing works still continually fine-tune the entire representation model, inevitably resulting in much catastrophic forgetting.  ...  Figure 1 : T-SNE [19] of CIFAR-10 [13] to show semantic drift and its compensation. Note that the black arrows reflect the semantic drift.  ... 
arXiv:2204.03410v3 fatcat:uzx27s3vfnbqtpjhhtinzyvuom

Co-Transport for Class-Incremental Learning [article]

Da-Wei Zhou, Han-Jia Ye, De-Chuan Zhan
2021 arXiv   pre-print
As a result, we propose CO-transport for class Incremental Learning (COIL), which learns to relate across incremental tasks with the class-wise semantic relationship.  ...  However, new classes often emerge in real-world applications and should be learned incrementally.  ...  Recent works focus on compensating for the drift of the incremental model. [46] utilizes the exemplars to build an extra validation set, with which an extra bias correction layer is trained.  ... 
arXiv:2107.12654v1 fatcat:65cocgr6zfhanc2cv3e4l5afwi

SSUL: Semantic Segmentation with Unknown Label for Exemplar-based Class-Incremental Learning [article]

Sungmin Cha, Beomyoung Kim, Youngjoon Yoo, Taesup Moon
2021 arXiv   pre-print
This paper introduces a solid state-of-the-art baseline for a class-incremental semantic segmentation (CISS) problem.  ...  drift of the background class and the multi-label prediction issue.  ...  One of the key additional difficulties of the class-incremental semantic segmentation (CISS) problem lies in the semantic drift of the background class present in the incrementally arriving training data  ... 
arXiv:2106.11562v3 fatcat:rco3ggvsnvhy3jlyurhuqiapvu

Cognitively Inspired Learning of Incremental Drifting Concepts [article]

Mohammad Rostami, Aram Galstyan
2021 arXiv   pre-print
Inspired by the nervous system learning mechanisms, we develop a computational model that enables a deep neural network to learn new concepts and expand its learned knowledge to new domains incrementally  ...  Our model can generate pseudo-data points for experience replay and accumulate new experiences to past learned experiences without causing cross-task interference.  ...  RELATED WORK The problem of continual learning of incremental drifting concepts lies in the intersection of lifelong learning to encode drifting concept classes and incremental learning to incorporate  ... 
arXiv:2110.04662v1 fatcat:55sswxlm3bgjrm5a5n2nipmyhy

Continual Attentive Fusion for Incremental Learning in Semantic Segmentation [article]

Guanglei Yang, Enrico Fini, Dan Xu, Paolo Rota, Mingli Ding, Hao Tang, Xavier Alameda-Pineda, Elisa Ricci
2022 arXiv   pre-print
However, the first incremental learning methods for semantic segmentation appeared only recently.  ...  Finally, we also introduce a novel strategy to account for the background class in the distillation loss, thus preventing biased predictions.  ...  ACKNOWLEDGMENT The authors would like to thank Nvidia Corporation for GPU donations to support our research.  ... 
arXiv:2202.00432v1 fatcat:2woq3xbvazah5e337ge2g4lzru

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

Marc Masana, Xialei Liu, Bartlomiej Twardowski, Mikel Menta, Andrew D. Bagdanov, Joost van de Weijer
2021 arXiv   pre-print
In this paper, we provide a complete survey of existing class-incremental learning methods for image classification, and in particular we perform an extensive experimental evaluation on thirteen class-incremental  ...  For future learning systems incremental learning is desirable, because it allows for: efficient resource usage by eliminating the need to retrain from scratch at the arrival of new data; reduced memory  ...  1 We do not refer to the scenario where each task only contains a single class, but consider adding a group of classes for each task.  ... 
arXiv:2010.15277v2 fatcat:wacloedzxrea3dgcuwm7xknyxe

Learning to Remember: A Synaptic Plasticity Driven Framework for Continual Learning [article]

Oleksiy Ostapenko, Mihai Puscas, Tassilo Klein, Patrick Jähnichen, Moin Nabi
2019 arXiv   pre-print
The amount of added capacity is determined dynamically from the learned binary mask. We evaluate DGM in the continual class-incremental setup on visual classification tasks.  ...  In order to tackle these challenges, we introduce Dynamic Generative Memory (DGM) - a synaptic plasticity driven framework for continual learning.  ...  training time and sensitivity to semantic drift.  ... 
arXiv:1904.03137v4 fatcat:kd3bvn263vcz5krnxzstkjwbiq

Essentials for Class Incremental Learning [article]

Sudhanshu Mittal and Silvio Galesso and Thomas Brox
2021 arXiv   pre-print
Moreover, we identify poor quality of the learned representation as another reason for catastrophic forgetting in class-IL.  ...  With these lessons learned, class-incremental learning results on CIFAR-100 and ImageNet improve over the state-of-the-art by a large margin, while keeping the approach simple.  ...  Semantic drift compensation for hilbert space. arXiv:12002.05715, 2020. 6 class-incremental learning.  ... 
arXiv:2102.09517v1 fatcat:li3tvwjwanfbnn52dx6lhi42im

SegMap: Segment-based mapping and localization using data-driven descriptors

Renaud Dubé, Andrei Cramariuc, Daniel Dugas, Hannes Sommer, Marcin Dymczyk, Juan Nieto, Roland Siegwart, Cesar Cadena
2019 The international journal of robotics research  
SegMap exploits a single compact data-driven descriptor for performing multiple tasks: global localization, 3D dense map reconstruction, and semantic information extraction.  ...  We show that the learned SegMap descriptor has superior segment retrieval capabilities, compared with state-of-the-art handcrafted descriptors.  ...  Note that these classes are solely used for training the descriptor and are not related to the semantics presented in Section 4.4.  ... 
doi:10.1177/0278364919863090 fatcat:qllqpzsy7fbllccrt4u5nz2cgm

Table of Contents

2018 2018 IEEE/ACM 5th International Conference on Big Data Computing Applications and Technologies (BDCAT)  
with Active Sampling for MOX Drift Compensation 61 Tamara Matthews (Dublin Institute of Technology), Muhammad Iqbal (National College of Ireland), and Horacio Gonzalez-Velez (National College of Ireland  ...  Dog Breed Detection and Recognition Based on Deep Learning 87 Richard O.  ... 
doi:10.1109/bdcat.2018.00004 fatcat:ykdhdzmqerarpook3yfqlzxfxi

A Survey on Multi-label Data Stream Classification

Xiulin Zheng, Peipei Li, Zhe Chu, Xuegang Hu
2019 IEEE Access  
Contrary to traditional data mining using static datasets, there are several challenges for data stream mining, for instance, finite memory, one-pass and timely reaction.  ...  Secondly, we identify mining constraints on classification for multi-label streaming data, and present a comprehensive study in algorithms for multi-label data stream classification.  ...  Detailed descriptions are listed as follows: • MediaMill 15 : The MediaMill data set is used for video semantic annotation task, where there are 43907 data instances (video frames) with 101 class labels  ... 
doi:10.1109/access.2019.2962059 fatcat:wqws3xkpmzeenatuzftjfshb2a

A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks [article]

Eden Belouadah, Adrian Popescu, Ioannis Kanellos
2020 arXiv   pre-print
and analyze them according to these properties, (2) introduce a unified formalization of the class-incremental learning problem, (3) propose a common evaluation framework which is more thorough than existing  ...  Here, we focus on the latter, place them in a common conceptual and experimental framework and propose the following contributions: (1) define six desirable properties of incremental learning algorithms  ...  In embedding systems, Semantic Drift Compensation ( ) [80] was proposed to estimate the semantic drift of past knowledge while learning new knowledge to compensate for it, to further improve performance  ... 
arXiv:2011.01844v4 fatcat:kw43joxgwzgsfmmi3su2pjfwv4

Self-Promoted Prototype Refinement for Few-Shot Class-Incremental Learning [article]

Kai Zhu, Yang Cao, Wei Zhai, Jie Cheng, Zheng-Jun Zha
2021 arXiv   pre-print
Few-shot class-incremental learning is to recognize the new classes given few samples and not forget the old classes.  ...  To address this problem, we propose a novel incremental prototype learning scheme.  ...  SDC [32] proposes a new method called semantic drift compensation to deal with the drift of the data in incremental learning.  ... 
arXiv:2107.08918v1 fatcat:xflgrzwl3bbwhaditcadipag4u

Deep-SLAM++: Object-level RGBD SLAM based on class-specific deep shape priors [article]

Lan Hu, Wanting Xu, Kun Huang, Laurent Kneip
2019 arXiv   pre-print
The result is an effective object-level RGBD SLAM system that produces compact, high-fidelity, and dense 3D maps with semantic annotations.  ...  Our work extends this idea to environments with unknown objects and imposes object priors by employing modern class-specific neural networks to generate complete model geometry proposals.  ...  It furthermore contains a loop closure module that effectively compensates for eventual drift accumulations. We conclude our exposition with our experimental results in Section 4.  ... 
arXiv:1907.09691v2 fatcat:4p6apklvrnc6hm3yin7mxwmska
« Previous Showing results 1 — 15 out of 1,856 results