mxahan.github.io

When is NCD Possibles

Background

How the DNN works: Filters to learn features,

Fundamental difference with visual systems: synthesis novel features, contextual understanding. IID in data

What is NCD: learning novel features from unlabeled data.

NCD Conditions and their Impacts

The current conditions that helps to solve the problem

Details of the assumptions and their necessity

Counter examples: Human bias in annotation? and fuzzy decision

Importance of the conditions

Causality connections

Learning vs Filtering

Open Questions