Substrates of neural plasticity
At the brain level
The hippocampus as a model to study learning and memory
The Hippocampus si much larger in mice than in humans. The main structures and pathways however are preserved
Henry Gustave Molaison (H.M.)
His Amygdala, hippocampal gyrus and anterior two thirds of the hippocampus were removed after having many dangerous seizures. He then suffered a Severe anterograde amnesia. Even though he had normal STM, and normal LTM prior to surgery, he could not transfer from STM to LTM, therefore he could not consolidate new declarative knowledge. He was still capable of acquiring implicit knowledge. They concluded that the Hippocampus is not a permanent storage area for explicit knowledge. The HC is involved in consolidation, a longerterm process taking months to years. Consolidation is understood to involve biological changes taking place in those other areas of cortex, once this has fully taken place, the hippocampus is not required for retrieval
At the circuit/neuron levels
McCulloch & Pitts neuron (perceptron)
It is very well known that the most fundamental unit of deep neural networks is called an artificial neuron/perceptron. But the very first step towards the perceptron we use today was taken in 1943 by McCulloch and Pitts, by mimicking the functionality of a biological neuron.
- It implements a linear descision boundary (hyperplane)
- The weights and biaises definfe the boundary
- Can be trained on labeled datasets
The difference between the McCulloch& Pitts Neuron and the Perceptron is that in the McCulloch& Pitts Neuron the Inputs can only be 0 or 1 and in the Perceptron they can be anything.
The neuron just sums the inputs and activates or not based on a threshold.
We can then implement decision boundaries
Example:
But only linear classification can be performed (xor is unsolvable)
Unsolvable:
Perceptron learning algorithm
Difference with biological neurons
similarities:
- Both can be active or inactive.
- The input/output is directed.
- The activation of a neuron is dependent on a weighted function of other neurons.
differences: - Real neurons exist in continuous time, whereas McCulloch-Pitts neurons operate in discrete time.
- Real neurons have degrees of activation, not just on/off.
- The activation as a function of the inputs of real neuron is typically not linear or threshold linear