https://github.com/aaltoml/sequential-gp

Code for 'Memory-based dual Gaussian processes for sequential learning' (ICML 2023)

https://github.com/aaltoml/sequential-gp

Science Score: 10.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
  • codemeta.json file
  • .zenodo.json file
  • DOI references
  • Academic publication links
    Links to: arxiv.org
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (13.2%) to scientific vocabulary

Keywords

continual-learning gaussian-process icml-2023
Last synced: 6 months ago · JSON representation

Repository

Code for 'Memory-based dual Gaussian processes for sequential learning' (ICML 2023)

Basic Info
  • Host: GitHub
  • Owner: AaltoML
  • License: mit
  • Language: Jupyter Notebook
  • Default Branch: main
  • Homepage:
  • Size: 14.2 MB
Statistics
  • Stars: 4
  • Watchers: 4
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Topics
continual-learning gaussian-process icml-2023
Created almost 3 years ago · Last pushed over 2 years ago
Metadata Files
Readme License

README.md

Memory-based dual Gaussian processes for sequential learning

This repository is the official implementation of the methods in the publication:

  • P.E. Chang, P. Verma, S.T. John, A. Solin, and M.E. Khan (2023). Memory-based dual Gaussian processes for sequential learning. In International Conference on Machine Learning (ICML). [arXiv]

Sequential learning with Gaussian processes (GPs) is challenging when access to past data is limited, for example, in continual and active learning. In such cases, errors can accumulate over time due to inaccuracies in the posterior, hyperparameters, and inducing points, making accurate learning challenging. Here, we present a method to keep all such errors in check using the recently proposed dual sparse variational GP. Our method enables accurate inference for generic likelihoods and improves learning by actively building and updating a memory of past data. We demonstrate its effectiveness in several applications involving Bayesian optimization, active learning, and continual learning.

Environment

We recommend setting up a conda environment for running the experiments. The code base is tested on a machine with a Ubuntu 22.04 distribution, CUDA11.6, and conda 23.1.0. shell conda create -n sequential-gp python==3.8 conda activate sequential-gp

Within the virtual environment, install the dependencies by running shell pip install -r requirements.txt

(Note that the hotspots experiment has its own environment and setup instructions.)

Experiments

There are a series of experiments which are organized inside the experiments folder as separate sub-folders. Each experiment sub-folder has their respective readme files with instructions on how to run the particular experiment.

Data sets

The datasets used for banana, UCI, and magnetometer experiments are available in experiments/data/ directory. The original source of the data sets are:

  • Banana: https://github.com/thangbui/streamingsparsegp
  • UCI: https://archive.ics.uci.edu/datasets
  • Magnetometer: https://github.com/AaltoML/magnetic-data

Contributing

For all correspondence, please contact paul.chang@aalto.fi or prakhar.verma@aalto.fi.

License

This software is provided under the MIT license.

Owner

  • Name: AaltoML
  • Login: AaltoML
  • Kind: organization
  • Location: Finland

Machine learning group at Aalto University lead by Prof. Solin

GitHub Events

Total
  • Watch event: 2
Last Year
  • Watch event: 2