GP-Tree: A Gaussian Process Classifier for Few-Shot Incremental Learning

Authors: Idan Achituve, Aviv Navon, Yochai Yemini, Gal Chechik, Ethan Fetaya

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our method against other Gaussian process training baselines, and we show how our general GP approach achieves improved accuracy on standard incremental few-shot learning benchmarks.
Researcher Affiliation Collaboration 1Bar-Ilan University, Israel 2Nvidia, Israel.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is publicly available at https://github.com/Idan Achituve/GP-Tree.
Open Datasets Yes We evaluated GP-Tree in this setup on the finegrained classification dataset, Caltech-UCSD Birds (CUB) 200-2011 (Welinder et al., 2010). For evaluating GP-Tree with DKL we used the CIFAR-10 and CIFAR-100 datasets. mini-Imagenet, a 100-class subset of the Imagenet (Deng et al., 2009) dataset
Dataset Splits Yes Since the data splits made public by (Tao et al., 2020) did not include a validation set, we pre-allocate a small portion of the base classes dataset for hyper-parameter tuning of GP-Tree, SDC (Yu et al., 2020), and PODNet (Douillard et al., 2020) on both datasets.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments, only general computing concepts.
Software Dependencies No The paper mentions PyTorch but does not provide specific version numbers for any software dependencies required to replicate the experiments.
Experiment Setup Yes We used Res Net-18 (He et al., 2016) as the backbone NN with an embedding layer of size 1024 and trained the models for 200 epochs.