some-thoughts-on-machine-learning-with-small-data

Metadata - Author: niklasriewald - URL: https://niklasriewald.com/2022/06/02/some-thoughts-on-machine-learning-with-small-data/ - #data-science #article

• Summary: Techniques for making machine learning work with small data is not discussed upon because companies usually boast about large data-sets. However, lots of the time there are only limited amount of data available
  • Metalearning techniques are useful here to train on adjacent problems
    • Few-shot learning, transfer learning and multi-task learning
    • Augmenting data or create synthetic data
    • https://www.youtube.com/watch?v=0rZtSwNOTQo&list=PLoROMvodv4rMC6zfYmnD7UG3LVvwaITY5
    • https://arxiv.org/abs/2009.14379
  • Inject domain knowledge by determining constraints like monotonically, convexity, and pairwise trust into the model directly
    • Your assumption has to be accurate and holds up over time
• Notes: