Split-and-Bridge: Adaptable Class Incremental Learning within a Single Neural Network

Authors: Jong-Yeong Kim, Dong-Wan Choi8137-8145

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In our thorough experimental analysis, our Split-and-Bridge method outperforms the state-of-the-art competitors in KD-based continual learning.
Researcher Affiliation Academia Jong-Yeong Kim, Dong-Wan Choi Department of Computer Science and Engineering, Inha University, South Korea kjy93217@naver.com, dchoi@inha.ac.kr
Pseudocode Yes Algorithm 1: Split-and-Bridge Incremental Learning
Open Source Code Yes 2https://github.com/bigdata-inha/Split-and-Bridge
Open Datasets Yes In our experiments, we train two benchmark datasets, CIFAR-100 (Krizhevsky, Hinton et al. 2009) and Tiny-Image Net (Le and Yang 2015)
Dataset Splits Yes Tiny-Image Net includes 100K training images and 10K validation images for 200 classes
Hardware Specification Yes We implement all the methods2 in Py Torch, and train each model on a machine with an NVIDIA TITAN RTX and Intel Core Xeon Gold 5122.
Software Dependencies No The paper mentions 'Py Torch' but does not specify a version number or other software dependencies with versions.
Experiment Setup Yes In the split phase, we divide two last residual blocks along with the final fully-connected (FC) layer of Res Net-18 into two disjoint partitions, i.e., θo and θn, implying S = 13 and L = 18. Full details are covered in our supplementary material.