Distance-Based Image Classification: Generalizing to New Classes at Near-Zero Cost

By Thomas Mensink, Jakob Verbeek, Florent Perronnin, Gabriella Csurka. 2013

493 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, 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 Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

In Meta & Continual Learning

Citations in the corpus

Prototypical Networks for Few-shot Learning (2017, methodology): “[21] attempt to extend their approach to also perform non-linear classification, but they do so by allowing classes to have multiple prototypes.”
Meta-Learning With Differentiable Convex Optimization (2019, neutral): “Indeed, nearest class mean classifier [17] is proven to work well under a low-dimensional feature as shown in Prototypical Networks [28].”
Diversity With Cooperation: Ensemble Methods for Few-Shot Classification (2019, methodology): “Simple mean-centroid classifiers have proven to be effective in the context of few-shot classification [5, 21, 31], which is confirmed in the following experiment.”
Towards Open Set Deep Networks (2015, neutral): “While more complex models, such as nearest class multiple centroids (NCMC) [13] or NCM forests [17], could provide more accurate modeling, for simplicity this paper focuses on just using a single mean.”
iCaRL: Incremental Classifier and Representation Learning (2016, methodology): “The choice of the average vector as prototype is inspired by the nearest-class-mean classifier [24] for incremental learning with a fixed feature representation.”
FeTrIL: Feature Translation for Exemplar-Free Class-Incremental Learning (2022, methodology): “The nearest class mean (NCM) [26] was used in [33], linear SVMs [30] were used in [1] and extreme value machines [34] were recently tested by [6].”
A Simple Baseline that Questions the Use of Pretrained-Models in Continual Learning (2022, methodology): “For the classifier, inspired by [21, 17] we use the nearest mean classification strategy.”
Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima (2021, methodology): “In each session, the inference is conducted by a simple nearest class mean (NCM) classification algorithm [36].”
Few-Shot Class-Incremental Learning via Relation Knowledge Distillation (2021, methodology): “They make the base knowledge space and new task knowledge space integrate into one space, namely composite representation space and achieves a very good performance though NCM (Mensink et al. 2013) classifier training the new tasks.”
Supervised Contrastive Replay: Revisiting the Nearest Class Mean Classifier in Online Class-Incremental Continual Learning (2021, methodology): “Although the Nearest-Class-Mean (NCM) classifier [41] is significantly undervalued in the CL community, we demonstrate that it is a simple yet effective substitute for the Softmax classifier as it not only addresses the recency bias but also avoids structural changes in the FC layer when new classes”
Online Continual Learning in Image Classification: An Empirical Survey (2021, methodology): “With a larger memory size, NCM remains very effective, and RV becomes much more helpful.”
Continual Prototype Evolution: Learning Online from Non-Stationary Data Streams (2020, background): “Similar to [33, 37], the class-prototype approximates the center of mass in the latent space, which we formally justify in Section 4.”
MgSvF: Multi-Grained Slow versus Fast Framework for Few-Shot Class-Incremental Learning (2020, background): “; E max do 5 Obtain (cid:2) z ð t Þ ;slow and (cid:2) z ð t Þ ;fast by Equations (5)-(6”
Semantic Drift Compensation for Class-Incremental Learning (2020, methodology): “However, they can also be used for classification when combined with, for example, a nearest class mean classifier [27].”
Memory-Efficient Incremental Learning Through Feature Adaptation (2020, methodology): “The literature for incremental learning prior to the deep-learning era includes incrementally trained support vector machines [7], random forests [42], and metric-based methods that generalize to new classes [36].”
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].”