stem/AI/Neural Networks/Learning/Tasks.md

Pattern Association

• Associative memory
  • Learns by association
• Autoassociation
  • Store a set of patterns by repeatedly presenting them in the network
    • Then presented partial or distorted stored pattern
    • Recall intended
  • Input and output data spaces are same dimensionality
• Heteroassociation
  • Arbitrary set of input patterns paired with another arbitrary set of output patterns
    • Supervised instead of unsupervised
  • No required relationship between input/output dimensionality
• Stages
  • Storage
  • Recall

Pattern Recognition

• Received pattern/signal is assigned to one of a prescribed number of classes

Function Approximation

• System Identification
• Inverse System

Control

• Learn to control a process or critical part of a system

Filtering

• Filtering
  • Extraction of information about a quantity of interest at discrete time $n$ by using data from time up to n
• Smoothing
  • Use information past time n
    • Expect smoother result
  • Delay in processing
• Prediction
  • Predict later data using current and previous

Beamforming

• Spatial filtering
• Distinguish spatial properties of a target signal and background noise
  • Similar to bats
  • Used in radar and sonar