On the Soft-Subnetwork for Few-Shot Class Incremental Learning

Authors: Haeyong Kang, Jaehong Yoon, Sultan Rizky Hikmawan Madjid, Sung Ju Hwang, Chang D. Yoo

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide comprehensive empirical validations demonstrating that our Soft Net effectively tackles the few-shot incremental learning problem by surpassing the performance of state-of-the-art baselines over benchmark datasets. The public code is available at https://github.com/ihaeyong/ Soft Net-FSCIL.
Researcher Affiliation Academia Haeyong Kang, Jaehong Yoon, Sultan Rizky Madjid, Sung Ju Hwang, and Chang D. Yoo Korea Advanced Institute of Science and Technology (KAIST) 291 Daehak-ro, Yuseong-gu, Daejeon {haeyong.kang,jaehong.yoon,suulkyy,sjhwang82,cd yoo}@kaist.ac.kr
Pseudocode Yes Algorithm 1 Soft-Subnetworks (Soft Net)
Open Source Code Yes The public code is available at https://github.com/ihaeyong/ Soft Net-FSCIL.
Open Datasets Yes To validate the effectiveness of the soft-subnetwork, we follow the standard FSCIL experimental setting. We randomly select 60 classes as the base class and the remaining 40 as new classes for CIFAR-100 and mini Image Net.
Dataset Splits No During the training stage in the base session, we select top-c% weights at each layer and acquire the optimal soft-subnetworks with the best validation accuracy. However, the specific split (e.g., percentages or counts) for a dedicated validation set is not explicitly provided in the text.
Hardware Specification Yes The experiments are conducted with NVIDIA GPU RTX8000 on CUDA 11.0.
Software Dependencies Yes The experiments are conducted with NVIDIA GPU RTX8000 on CUDA 11.0.
Experiment Setup Yes In each incremental few-shot learning session, the total number of training epochs is 6, and the learning rate is 0.02. We train new class session samples using a few minor weights of the soft-subnetwork (Conv4x layer of Res Net18 and Conv3x layer of Res Net20) obtained by the base session learning.