Autonomous Navigation Ben McElroy 21 st April 2011 University of Essex 1 Part-financed by the European Regional Development Fund
Part-financed by the European Regional Development Fund RAM-based Weightless Networks One Shot Learning Arbitrary Mappings From Inputs to Outputs Easier Direct Hardware Implementation 2
Part-financed by the European Regional Development Fund 3 Input MatrixSample Layer
Part-financed by the European Regional Development Fund Address
Part-financed by the European Regional Development Fund 5 Problems faced with WNNs Number of Layers Number of Neurons Per Layer Connectivity Map per Layer Architecture
Part-financed by the European Regional Development Fund 6 Meta-Network Input ‘Design’ Create population Crossover function Rank and prune this generation Test each architecture Mutation
Part-financed by the European Regional Development Fund 7 Paper Differing number of neurons per layer Using Sonar Data from ISEN
Part-financed by the European Regional Development Fund 8 Other Variables Number of Inputs/Outputs – Not all scenarios require the same set of outputs – Not all outputs will work out
Part-financed by the European Regional Development Fund 9 How it will fit together Modularised Global/Local Goal
Part-financed by the European Regional Development Fund 10 Next Further Study into Modifying the inputs and outputs Split input data into multiple iterations of WNNs Work closely with control