Functional Regularisation for Continual Learning with Gaussian Processes

Authors: Michalis K. Titsias, Jonathan Schwarz, Alexander G. de G. Matthews, Razvan Pascanu, Yee Whye Teh

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We now test the scalability and competitiveness of our method on various continual learning problems, referring to the proposed approach as Functional Regularised Continual Learning (FRCL).
Researcher Affiliation Collaboration Michalis K. Titsias Deep Mind mtitsias@google.com Jonathan Schwarz Deep Mind & University College London schwarzjn@google.com Alexander G. de G. Matthews Deep Mind alexmatthews@google.com Razvan Pascanu Deep Mind razp@google.com Yee Whye Teh Deep Mind ywteh@google.com
Pseudocode Yes Algorithm 1 Functional Regularised Continual Learning (FRCL) with task boundary detection
Open Source Code No The paper mentions that FRCL methods were implemented using GPflow and that VCL results were obtained using code provided by other authors, but does not provide a link or explicit statement that their own implementation code for FRCL is open-source.
Open Datasets Yes We consider experiments on three established Continual Learning classification problems: Split-MNIST, Permuted-MNIST and sequential Omniglot (Goodfellow et al., 2013; Zenke et al., 2017; Schwarz et al., 2018), described in the Appendix.
Dataset Splits Yes Note that for the MNIST results, we obtain final results after optimising hyperparameters on the validation set and using those values to train on the union of training & validation set.
Hardware Specification No The paper does not provide specific details about the hardware used, such as GPU models, CPU specifications, or memory, for running the experiments.
Software Dependencies No FRCL methods have been implemented using GPflow (Matthews et al., 2017). The paper mentions GPflow as a library but does not provide its version number or version numbers for any other software dependencies.
Experiment Setup Yes Experimental details for all experiments are shown in Tables 4, 5 and 6.