220 likes | 244 Views
Neural Networks. Joost N. Kok Universiteit Leiden. Neural Networks. Book: Introduction to the theory of Neural Computation by Hertz, Krogh, Palmer Website: www.liacs.nl/home/joost/nn08.htm Additional Material: Powerpoint Sheets, Journal Articles, practical exercises
E N D
Neural Networks Joost N. Kok Universiteit Leiden
Neural Networks • Book: Introduction to the theory of Neural Computation by Hertz, Krogh, Palmer • Website: www.liacs.nl/home/joost/nn08.htm • Additional Material: Powerpoint Sheets, Journal Articles, practical exercises • Other recommended books: Haykin, Bishop
The Hungarian-born mathematician, John von Neumann (1903-1957)
Biological computers Five distinguishing properties: • Highly parallel • Robust and fault tolerant • Adaptive • Deals with fuzzy, noisy information • Small, compact
Graceful Degradation performance damage
Neural activity out in
1 f(x) = 1 + e -x/a Input-output function • nonlinear function: a 0 f(e) a e
wAB A B Artificial Connections (Synapses) • wAB • The weight of the connection from neuron A to neuron B
Real Neurons • Nonlinear Summation • Sequences of pulses • No fixed time-delay
History • 1943: McCulloch and Pitts: artificial neuron • 1960: Rosenblatt: perceptrons • 1969: Minsky and Papert: XOR problem • 197?: Associative content-addressable memory • 198?: Hopfield Networks, Boltzmann Networks • 1985: Backpropagation learning rule • 2000+ : Spiking networks, Support Vector Machines
Hopfield McCulloch Pitts Papert Minsky Boltzmann Hebb
Issues • Neurocomputing vs. Neuroscience • Types of Learning: • Supervised • Unsupervised • Reinforcement
Research Questions • Design: what is best architecture? • Learning: find good algorithms • Analysis: what is the power of networks? • Implementation: how should the network be implemented?