[18] AI

• Purpose of a graph:

  Structure used to model relationships between objects
• Use of graph to aid AI (7):

  · Artificial neural networks represented on graphs · Graph provides structure for relationship between nodes · AI problems solved as finding path in graph · Graphs analysed by algorithms · A*/ Dijksta’s algorithm · Used in machine learning · E.g. back propagation of errors
• A* algorithm description (1)

  A computer method used to find optimal path between two mapped locations
• Artificial neural network description (1)

  computer system modelled on brain
• Machine Learning description (1)

  Computer program that improves its performance at certain tasks with experience
• Deep Learning description (4)

  · Uses artificial neural networks · Contain high number of hidden layers modelled on human brain · Uses many layers to progressively extract higher level features from raw input · Specialised form of machine learning
• Reasons for using deep learning (5)

  · Good use of unstructured data · Outperforms other methos if data size large · Process data non linear approach · Effective identify hidden patterns – humans cannot see · More accurate outcome with higher number of hidden layers
• Supervised learning (4)

  · Allows data to be collected.

  Data output produced

  · Known input and association outputs are given

  · Labelled input data

  · Able to predict future outcomes based on past data
• Unsupervised learning (3)

  · Unsupervised machine learning helps all kinds of unknown pattern in data to be found · Only requires input data to be given · Uses any data /unlabelled input data
• Reasons for having multiple hidden layers in an artificial neural network (4)

  · Enables deep learning to take place · Problem with higher complexity to solve requires for levels · Enable neural network to learn + make decisions on its own · Improve result accuracy
• How artificial neural networks enable machine learning (7)

  · Aim to replicate how human brain works · Weights assigned for each connection · Data input at input layer + passed down system through hidden layers · At hidden layers, data analysed + outputs calculated · Learning repeated many times to produce optimum results · Output layer provides results · Back propagation of errors used to correct any errors made
• Reinforcement Learning (3)

  · Uses large number of tasks with unknown outcomes · Use of feedback · Computer program improves performance in accomplishing similar tasks
• Regression


  Predict the value of a dependent variable based upon another explanatory variable.
• Non-linear regression


  Used where there is a correlation but it is not linear
• Unsupervised learning why useful

  Useful for categorising many different objects Identifying hidden trends or patterns Anomaly detection(e.g. fraudulent transactions, spotting skin cancer, crime detection)
• Linear regression

  Used where there is a straight line correlation between variables.
• Artificial Neural Network

  Component of AI - meant to simulate a functioning brain

  Key component of machine learning

  Self learning capabilities that enables them to produce better results as data becomes available

  Can be layered - interconnected layers - some might be hidden

  Weights are assigned between nodes