Prototype Aggregation

Prototype aggregation is a machine learning technique focusing on representing data classes using prototypes—summarized feature vectors—to improve classification accuracy and model interpretability. Current research emphasizes developing algorithms that dynamically generate and refine these prototypes, often within contrastive learning frameworks, to address challenges like catastrophic forgetting in incremental learning and domain adaptation. This approach is proving valuable in various applications, including few-shot learning, open-set recognition, and improving the robustness and explainability of deep learning models.

Papers

May 24, 2024

Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective
Leyuan Wang, Liuyu Xiang, Yunlong Wang, Huijia Wu, Zhaofeng He
Contrastive Learning Catastrophic Forgetting Incremental Learning Class Incremental Learning Imbalanced Learning Prototype Aggregation

January 24, 2024

Towards Efficient and Effective Deep Clustering with Dynamic Grouping and Prototype Aggregation
Haixin Zhang, Dong Huang
Contrastive Learning High Efficiency Deep Clustering Contrastive Method Level Aggregation Prototype Aggregation Contrastive Deep Clustering

October 12, 2023

Learning Transferable Conceptual Prototypes for Interpretable Unsupervised Domain Adaptation
Junyu Gao, Xinhong Ma, Changsheng Xu
Unsupervised Domain Adaptation Prototype Selection Prototype Aggregation Intra Domain Category Aware Prototype

May 30, 2023

SENet: A Spectral Filtering Approach to Represent Exemplars for Few-shot Learning
Tao Zhang, Wu Huang
Shot Learning Shot Classification Exemplar Image Exemplar Guided Prototype Aggregation Kernel Induced Loss

January 11, 2023

Head-Free Lightweight Semantic Segmentation with Linear Transformer
Bo Dong, Pichao Wang, Fan Wang
Vision Transformer Semantic Segmentation Linear Transformer Lightweight Semantic Segmentation Prototype Aggregation Adaptive Frequency

December 5, 2022

Cross-Domain Few-Shot Relation Extraction via Representation Learning and Domain Adaptation
Zhongju Yuan, Zhenkun Wang, Genghui Li
Contrastive Learning Domain Adaptation Representation Learning Shot Relation Extraction Prototype Aggregation

July 14, 2022

Provably Adversarially Robust Nearest Prototype Classifiers
Václav Voráček, Matthias Hein
Adversarial Perturbation Threat Model Prototype Classifier Prototype Aggregation \Ell_p$ Robustness

May 18, 2022

TTAPS: Test-Time Adaption by Aligning Prototypes using Self-Supervision
Alexander Bartler, Florian Bender, Felix Wiewel, Bin Yang
Deep Neural Network Adaptation Concern Self Supervision Visual Prototype Self Supervised Training Unseen Test Distribution Test Loss Prototype Aggregation

March 16, 2022

PMAL: Open Set Recognition via Robust Prototype Mining
Jing Lu, Yunxu Xu, Hao Li, Zhanzhan Cheng, Yi Niu
Open Set Recognition Prototype Learning Prototype Aggregation Prototype Generation

Prototype Aggregation

Papers

Rethinking Class-Incremental Learning from a Dynamic Imbalanced Learning Perspective

Towards Efficient and Effective Deep Clustering with Dynamic Grouping and Prototype Aggregation

Learning Transferable Conceptual Prototypes for Interpretable Unsupervised Domain Adaptation

SENet: A Spectral Filtering Approach to Represent Exemplars for Few-shot Learning

Head-Free Lightweight Semantic Segmentation with Linear Transformer

Cross-Domain Few-Shot Relation Extraction via Representation Learning and Domain Adaptation

Provably Adversarially Robust Nearest Prototype Classifiers

TTAPS: Test-Time Adaption by Aligning Prototypes using Self-Supervision

PMAL: Open Set Recognition via Robust Prototype Mining