Data as Prior/Innate knowledge for Deep Learning models

  1. Use an existing rule-based expert system to generate data
  2. Inject noise to the generated data to make learned system more resilient
  3. Combine synthetic data from real-world data
  4. Train a deep learning system
  1. It allows to introduce innate domain-specific knowledge
  2. It allows to control the balance of innate vs. real-world learned knowledge by simply controlling the balance between synthetic and real-world data.
  3. Adding noise to the synthetic data allows us to learn a system that is more robust to noisy data in inference time
  4. Introducing real-world data means that we can have an extensible continuous learning system that is better over time than the expert system specified by the domain experts.

--

--

--

Cofounder/CTO at Curai (AI for healthcare). Former Quora VP, Netflix Director. Software, Machine Learning, Data, Recsys... From Barcelona, in the Valley

Love podcasts or audiobooks? Learn on the go with our new app.

Recommended from Medium

How to build a simple CNN based Image classifier using Keras

AI Idiots In NLP

Build Free Object Detection Machine Learning model without code

Introduction to CNN Architecture

Implementing Chargrid: Towards Understanding 2D Documents.

Text Generation with GPT-2 in Action

Gaussian Process Regression for Housing Prices: How different kernels and restart optimizers…

Introduction To Anomaly Detection Methods — Part I

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Xavier Amatriain

Xavier Amatriain

Cofounder/CTO at Curai (AI for healthcare). Former Quora VP, Netflix Director. Software, Machine Learning, Data, Recsys... From Barcelona, in the Valley

More from Medium

Housing Data: a Closer Look

Tanzanian Waterpoints: Ternary classification with three ML models

Build Recommendation Systems with PyTorch Geometric and ArangoDB

What Is MLOps? An Introduction to Machine Learning Operations

What is MLOps?