XAI-lecture 1

• Artifical intelligence


  is the broad mandate of creating machines that can think intelligently
• Machine learning 


  is one way of creating machines that can think inteligently, by using algorithms to glean insights from data
• Deep Learning 


  is a way of using algorithms to glean insights of data, using a specific algorithm (Neural network)
• Various Purposes

  • Interpretability & Explainability
  • Debugging & Improving Models
  • Comparing & Selecting Models
  • Teaching & Understanding Concepts
• Explaining through projections with help of dimensionality reduction techniques --> PCA, UMAP, t-SNE and visualization of embeddings
• embeddings : a vector representation of a feature vector that is used to represent a feature
• The 6 questions of visual analytics for deep learning : why, what , when, who ,how where
• intrinsic interpretability : the ability to understand the meaning of the model without the need for additional information
• Post-hoc explanation : An explanation that is made after the fact, based on the results of the experiment.
• Dependancy model agnostic : visualisation based on multiple models, can be used to visualise any model
• dependancy model specific : visualisation technique can only be used on 1 type of model
• Different types of visualization principles :

  • tables
  • Networks
  • node-link-diagrams
  • multidim table
  • trees