Related work

The foundational work on continual learning, 1988 to 2026: methods, theory, benchmarks, surveys, and the neuroscience of memory. Updated weekly; some carry our notes.

8 papers of 6,984Sort Recent · Most cited
  1. 2021
    Always Be Dreaming: A New Approach for Data-Free Class-Incremental LearningJames Smith, Yen-Chang Hsu, Jonathan Balloch … Zsolt KiraICCV · Georgia Institute of Technology · Samsung (United States) · +1
    PDF ↗
  2. 2021
    Unsupervised Progressive Learning and the STAM ArchitectureJames Smith, Cameron Taylor, Seth Baer, Constantine DovrolisIJCAI · Georgia Institute of Technology
    PDF ↗
  3. 2021
    Memory-Efficient Semi-Supervised Continual Learning: The World is its Own Replay BufferJames Smith, Jonathan Balloch, Yen-Chang Hsu, Zsolt KiraIJCNN · Georgia Institute of Technology · Samsung (United States)
    PDF ↗
  4. 2021
    Unsupervised Class-Incremental Learning Through ConfusionShivam Khare, Kun Cao, James M. RehgarXiv · Georgia Institute of Technology
    PDF ↗
  5. 2021
    Continual Learning of Knowledge Graph EmbeddingsAngel Daruna, Mehul Gupta, Mohan Sridharan, Sonia ChernovaRA-L · Georgia Institute of Technology · University of Birmingham
    PDF ↗
  6. 2021
    Continual Learning for Text Classification with Information Disentanglement Based RegularizationYufan Huang, Yanzhe Zhang, Jiaao Chen … Diyi YangNAACL · Georgia Institute of Technology · Google (United States) · +1
    PDF ↗
  7. 2021
    Lifelong Learning of Hate Speech Classification on Social MediaJing Qian, Hong Wang, Mai ElSherief, Xifeng YanNAACL · Georgia Institute of Technology · University of California, Santa Barbara
    PDF ↗
  8. 2021
    Principal Gradient Direction and Confidence Reservoir Sampling for Continual LearningZhiyi Chen, Tong LinSpringer LNCS · Georgia Institute of Technology · Peking University · +1
    PDF ↗
About this index

We keep this list because we read the field and wanted one place to see it. It covers work on continual learning itself, in the core areas of machine learning, and leaves out papers that apply it inside another field, such as medical imaging or fault diagnosis. It lists only papers we have a reason to trust: published at a venue like NeurIPS, ICML, ICLR, CVPR or TPAMI, or led by someone who has published there, or cited a few hundred times. Preprints that later get accepted, and authors who later publish at those venues, are picked up by the weekly run. It is seeded from the community lists kept by ContinualAI and by Xialei Liu, then filled out from OpenAlex, and every week a script looks for new papers on OpenAlex and arXiv. A model reads each candidate and decides whether it belongs; a person reviews the additions before they go live. Authors and affiliations come from OpenAlex, so a recent preprint can lack its institutions for a week or two.

Missing something, or filed under the wrong venue? Write to hello@unify.ai with the arXiv id or DOI.