Neurons
- Most of my research is with simulated neurons.
- I'm using standard models as part of the
Human Brain Project but have my own neural model.
- Mine is a fatiguing leaky integrate and fire model.
It's a relatively accurate model of biological neurons, but
there are more accurate models.
- One of my big things is the cell assembly. This is a bunch
of neurons that represent concepts in both humans and in
my models.
- We've been working on them for a long time, and can learn them.
Still there are problems with their formation and duration of
persistence.
- We've used simple CAs to create cell assembly robots. These
are agents that move about in a virtual environment.
- They respond to natural language commands, have plans,
view the environment, and learn some simple things.
- Learning in this context involves Hebbian learning rules
for synaptic weight modification. I have a compensatory
learning rule.
- We use these systems to learn categories. The performance
of these is comparable to standard machine learning algorithms.
- The paper I've just submitted is on a spiking half-cognitive model
for classification.
- This is another one of our neuro-cognitive models.
- Cognitive scientists have been making cognitive models, both
box models and programs, for a long time.
- We're doing this, but we're using simulated neurons as the basis
of our models.
- We have neuro-cognitive models for learning a two choice task,
categorisation, and natural language parsing.
- Most of the coding for this has been in Java.