We are searching data for your request:
Upon completion, a link will appear to access the found materials.
How can spike-timing-dependent plasticity and homeostatic plasticity both be right? If spike-timing-dependent plasticity consistently tries to strengthen connections, but homeostatic plasticity consistently tries to weaken strong connections, wouldn't the two cancel each other out, causing no net plasticity to occur and consequently brains couldn't learn?
I have a pretty good understanding of spike-timing-dependent plasticity, but I just recently came across homeostatic plasticity and I'm trying to understand how both theories can be right. What am I misunderstanding? Can someone help me understand homeostatic plasticity better and how it applies to spike-timing-dependent plasticity?
Homoestatic plasticity can be used with spike-dependent plasticity given they have two different goals and aren't applied uniformly across neural population. To support this argument, I'm going to use a computation model taken from "Simultaneous unsupervised and supervised learning of cognitive functions in biologically plausible spiking neural networks" by Bekolay et al.
In the model, an ensemble of neurons with input and output weights , is trying to learn how to change their output weights to approximate a given function. PES (STDP rule) changes the output weights of the neurons to approximate. The BCM (homeostatic rule) increases the sparsity of the input weights. When used together as hPES, they allow for less parameter sensitivity of the ensemble and a lower overall spiking rate, allowing the group of neurons to conserve energy.
 For a visual explanation of what this means, see this blog post.
The key concept is that STDP is synapse specific, whereas homeostatic plasticity is more global. (As an aside, it is incorrect to say that STDP strengthens connections. STDP strengthens connections with pre-before-post spiking relationships, but weakens connections with post-before-pre spiking relationships. You man be thinking of Hebbian plasticity. The following explanation applies to either.) Let's say you have 100 synapses onto your neuron, all of which have a weight of 1. Let's say 10 of them undergo strengthening through Hebb's rule or STDP to achieve a weight of 10, so the total strength of all synapses into the neuron is now 190. Then, when homeostatic plasticity is applied so that the total input to the neuron returns to a value of 100, the 10 strengthened synapses will have a new weight of 10*100/190 = 5.3, and the other 90 synapses will have a new weight of 1*100/190 = 0.53. Therefore, even though the total input to the neuron is held constant, the distribution of those weights changes, so information gained through the initial Hebbian plasticity or STDP is retained.