Week 10 Flashcards
Static vs dynamic problems
NN is useful for learning from data that is static
Many problems are dynamic or have a temporal or sequential component e.g. stock market prediction, speech recognition
These can be processed using a NN with a shifting time window to predict future from recent past, or recurrent neural networks.
What is a RNN
Similar structure to MLP but have connections from activity in previous time step (recurrent link)
Short term memory
Many possible types.
What is the aim of unsupervised learning
Find regularities in inputs
What is a self organising map
Aims to learn to map points from high-D space to a low-D space in a way to preserve topological properties (spatial relations)
Use: visualisation and discover regularities in data
Self organised: map emerging from local interactions (competition and cooperation between data points).
SOM Architecture
Many input nodes as ‘features’ in the data
Input nodes are connected to a ‘map’ of interconnected nodes
Every node in the output map is connected to every input node via weighted edges
How does SOM work
Every input pattern is a point in high-D space
Every input is made to correspond to a node in output map via a competitive process among nodes on the output space.
The winner is the node whose weights have smallest Euclidean distance to input pattern.
Applications of SOM
Originally used for speech recognition
Network learnt to classify phonemes from processed input waveforms of speech
The SOM learned to map any input phoneme to its corresponding standard/ideal phoneme
The output standard phonemes can then be easily converted into written words in a word processor.
What is artificial life?
The study of man-made systems that exhibit behaviours characteristic of natural living systems.
Why artificial life?
Biology is restricted to study a specific instance of life, life on Earth: all life forms we know are carbon based
A more universal understanding of life: are the essential ingredients of living organisms independent from the material they are made of?
Why bottom-up approach?
Embraces emergence, start with a system of locally interacting elements according to simple rules that spontaneously give rise to emergent properties.
What is the 2 distinct philosophies of Alife?
Weak Alife - computer simulations are just that, simulations and investigations of lie
Strong Alife - life is not just restricted to a carbon-based chemical process. Life can be ‘created’ in silico.
What is cellular automata
Method of representing stylised universe, with rules acting over the entire universe.
Demonstration of emergence, complex global behaviour emerges spontaneously from local interactions following simple rules
What is an automaton?
Consists of a grid/lattice of cells each of which can be a number of states
How do states change in cellular automata?
States of an automaton change over time in discrete timesteps
The state of each cell is modified in parallel at each timestep according to the state transition rules.
These determine the new states of each of the cells in the next timestep from the states of that cells neighbours.
What is a neighbourhood in CA?
Determines the extent of the interaction between cells. (i.e. count all surrounding cells, or only those in a + shape.)
