Novel Class Discovery

Novel class discovery (NCD) focuses on identifying new categories within unlabeled data by leveraging knowledge from previously known classes. Current research emphasizes developing robust algorithms and model architectures, such as those based on self-training, knowledge distillation, and prototype learning, to address challenges like imbalanced class distributions and catastrophic forgetting across various data types (images, point clouds, tabular data, graphs). This field is significant because it enables more adaptable and robust machine learning systems capable of handling open-world scenarios where new classes continuously emerge, impacting applications ranging from biomedical concept discovery to object detection in evolving environments.

Papers

September 2, 2022

A Method for Discovering Novel Classes in Tabular Data
Colin Troisemaine, Joachim Flocon-Cholet, Stéphane Gosselin, Sandrine Vaton, Alexandre Reiffers-Masson, Vincent Lemaire
Multi Task Learning Practical Method Tabular Data Novel Class Discovery Classification Datasets Heterogeneous Tabular Data

July 21, 2022

Novel Class Discovery without Forgetting
K J Joseph, Sujoy Paul, Gaurav Aggarwal, Soma Biswas, Piyush Rai, Kai Han, Vineeth N Balasubramanian
Latent Representation Novel Class Discovery Unsupervised Discovery ImageNet LT Unseen Category

July 18, 2022

Class-incremental Novel Class Discovery
Subhankar Roy, Mingxuan Liu, Zhun Zhong, Nicu Sebe, Elisa Ricci
Knowledge Distillation Novel Class Discovery Class Prototype Rehearsal Based Continual Learning

July 5, 2022

June 24, 2022

Mutual Information-guided Knowledge Transfer for Novel Class Discovery
Chuyu Zhang, Chuanyang Hu, Ruijie Xu, Zhitong Gao, Qian He, Xuming He
Knowledge Transfer Long Term Cooperation Unseen Class Novel Class Discovery Joint Latent Space

April 22, 2022

Spacing Loss for Discovering Novel Categories
K J Joseph, Sujoy Paul, Gaurav Aggarwal, Soma Biswas, Piyush Rai, Kai Han, Vineeth N Balasubramanian
Loss Function Latent Space Novel Class Discovery Category Discovery Cifar 10

March 7, 2022

Open Set Domain Adaptation By Novel Class Discovery
Jingyu Zhuang, Ziliang Chen, Pengxu Wei, Guanbin Li, Liang Lin
Novel Class Discovery Open Set Domain Adaptation

December 23, 2021

Dual Path Structural Contrastive Embeddings for Learning Novel Objects
Bingbin Li, Elvis Han Cui, Yanan Li, Donghui Wang, Weng Kee Wong
Contrastive Learning Transfer Learning Feature Representation Object Representation Novel Class Discovery Structural Embeddings Transfer Learning Strategy

December 3, 2021

Novel Class Discovery in Semantic Segmentation
Yuyang Zhao, Zhun Zhong, Nicu Sebe, Gim Hee Lee
Semantic Segmentation COCO Dataset Novel Class Discovery Semantic Segmentation Problem