From N to N+1: Multiclass Transfer Incremental Learning

By Ilja Kuzborskij, Francesco Orabona, Barbara Caputo. 2013

151 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Representation & Self-Supervised 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 Representation & Self-Supervised Learning

Citations in the corpus

Unbounded cache model for online language modeling with open vocabulary (2017, background): “Kuzborskij et al. [39] proposed to reduce the loss of accuracy when adding new classes by partly retraining the existing classiﬁer.”
iCaRL: Incremental Classifier and Representation Learning (2016, background): “[17] showed that a loss of accuracy can be avoided when adding new classes to an existing linear multi-class classifier, as long as the classifiers can be retrained from at least a small amount of data for all classes.”
Doodle It Yourself: Class Incremental Learning by Drawing a Few Sketches (2022, background): “Incremental Learning: Incremental Learning (IL) [38, 25] is a machine learning paradigm where a model adapts itself to learn new tasks sequentially while retaining the previously learnt knowledge.”
Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima (2021, background): “Rooted from real-world applications, this topic has attracted a significant amount of interest in recent years [5, 30, 31, 40, 26].”
Subspace Regularizers for Few-Shot Class Incremental Learning (2021, background): “As noted by past work (Kuzborskij et al., 2013), a more realistic evaluation of models’ ability to rapidly acquire new concepts should consider their ability to discriminate among both new concepts and old ones, a problem usually referred to as few-shot class-incremental learning (FSCIL)2 (Tao et al”
Distilling Causal Effect of Data in Class-Incremental Learning (2021, background): “Incremental learning aims to continuously learn by accumulating past knowledge [2, 19, 24].”
Incremental Learning in Online Scenario (2020, methodology): “Other techniques [19, 17, 13] use ensemble of weak classifiers and nearest neighbor classifier.”
Mnemonics Training: Multi-Class Incremental Learning Without Forgetting (2020, background): “Incremental learning has a long history in machine learning [3, 21, 14].”
ScaIL: Classifier Weights Scaling for Class Incremental Learning (2020, background): “DeeSIL [3] is an adaptation of transfer learning to a class IL context [13, 15].”
Large Scale Incremental Learning (2019, background): “Incremental learning has been a long standing problem in machine learning [3, 17, 16, 12].”
Incremental Classifier Learning with Generative Adversarial Networks (2018, background): “13] which does not use any exemplars on CIFAR-100 [11], Flower-102 [18] and MS-Celeb-1M-Base [5, 6] datasets. 2. Related Work Incremental learning has been a long standing problem in machine learning [3, 20, 17, 12]. Before the deep learning took off, people had been developing incremental learning ”
A Strategy for an Uncompromising Incremental Learner (2017, methodology): “In [13], Kuzborskij et al. develop a least squares SVM approach to incrementally update a N-category classiﬁer to recognize N+1 classes.”
Recent Advances in Open Set Recognition: A Survey (2018, background): “not scale gracefully with the number of classes. On the other hand, though new classes (UUCs) are assumed to appear incremental in class incremental learning (C-IL) [125], [126], [127], [128], [129], [130], these studies mainly focused on how to enable the system to incorporate later coming training”
Bilevel Programming for Hyperparameter Optimization and Meta-Learning (2018, background): “Yet another choice would be to consider g w j ,λ ( x ) = (cid:104) w + λ, x (cid:105) , in which case λ represents a common model around which task speciﬁc models are to be found (see e.g. Evgeniou et al., 2005; Finn et al., 2017; Khosla et al., 2012; Kuzborskij et al., 2013, and reference therein).”
Adding New Tasks to a Single Network with Weight Trasformations using Binary Masks (2018, background): “ntelligence is the ability to adapt an initial, pre-trained model to novel, unseen scenarios. This is crucial for increasing the knowledge of an intelligent system and developing eective incremental [43,24], life-long learning [52,53,49] algorithms. While fascinating, achieving this goal requires fa”
Learning to Learn: Model Regression Networks for Easy Small Sample Learning (2016, methodology): “In addition, adaptive SVM and its variants present SVM-based model adaptation by combining classifiers learned on related categories [47, 102, 23, 2, 3, 53, 88, 97].”
Learning with Augmented Class by Exploiting Unlabeled Data (2014, background): “In (Fink et al. 2006; Kuzborskij, Orabona, and Caputo 2013), the binary classiﬁers of each new class is added incrementally and trained by sharing the hypothesis of the existing classes.”
Multi-Instance Learning With Emerging Novel Class (2021, methodology): “Incremental learning is one branch of methods dealing with the changing environments, which includes example-incremental learning [23], [24], attri-bute-incremental learning [25], class-incremental learning [18], [26], etc.”
Class-Incremental Learning: A Survey (2023, background): “However, most of them address the incremental learning within two tasks, i.e. , the model is only updated with a single new stage [12], [13].”
Incremental few-shot learning via vector quantization in deep embedded space (2021, unknown): “Incremental learning is a learning paradigm that allows the model to continually learn new tasks on novel data, without forgetting how to perform previously learned tasks (Cauwenberghs & Poggio, 2001; Kuzborskij et al., 2013; Mensink et al., 2013).”