Robust Learning

• Robust

  • Robust refers to the internal mechanism or nature of a system that prevents it from being affected by disturbances such as stress or environmental changes (Wikipedia)

• Co-teaching learning

  • Sometimes, large noise can cause significant changes in the output.
  • As a tendency, the model learns from data with less noise first.
  • Two learners can become more robust to noise by sharing information with each other.

• Reject option classification

  • Difficult classifications are rejected and left to humans, especially for critical domains such as medical and financial where mistakes are not acceptable.
  • Approach 1: Reject objects with low confidence as they are (with limitations).
  • Approach 2: Train a classifier and a rejector separately, and use the rejector for rejection.
    • This approach works well for binary problem, but not for more complex problems.

Class Classification §

Machine Learning §

(Source: Information Science Expert)