Anyone know of a paper that shows f-I/I-V curves and EPSPs/Spikes for *synaptic* input, e.g. via presynaptic axonal stimulation (or glutamate uncaging, in which case EPSP/Spikes as a function of laser intensity)?
Toviah Moldwin
- 257 Followers
- 483 Following
- 17 Posts
Computational neuroscientist @ELSC @Segev_Lab interested in neural learning algorithms. Singer and guitarist for the rock band Synfire Chain. Dualist.
It will take a while for the preprint to come out, so here's a video of my work with Li Azran at the @Segev_Lab about the Calcitron, a simple linear neuron model that can implement many learning rules using Calcium-based plasticity.
The Calcitron
Although Mastodon's ability to create longer posts is appealing to the writer, it is unpleasant for the reader, who often prefers not to scroll through long blocks of text. I suspect this will prevent mass adoption of this platform.
When people say consciousness 'emerges from' or 'reduces to' computation and/or the brain, are they making the claim that consciousness belongs to the same ontological category as computational processes and/or materials?
Sam GershmanThis is so interesting:
https://www.cell.com/cell-reports/fulltext/S2211-1247(21)01802-7
Synaptically connected neurons send proteins to each other packaged in exosomes. Previous studies showed protein transfer but I think this is the first large-scale screen, identifying 200 proteins (likely an underestimate because they only looked at ones passing through axons).
More evidence that our conventional models of neural signaling are radically incomplete!
Lucas C ParraLearning without backpropagation is really taking off in 2022
First, @BAPearlmutter et al show in "Gradients without Backpropagation" that a single forward pass with perturbed weights is enough to compute unbiased estimate of gradients:
https://arxiv.org/abs/2202.08587
Then, Mengye Ren et al show in "Scaling Forward Gradient With Local Losses" that the variance of doing this is high, but can be reduced by doing activity perturbation (as in Fiete & Seung 2006), but more importantly, having many "local loss" functions:
https://arxiv.org/abs/2210.03310
Then Jeff Hinton takes the "local loss" to another level in "Forward-Forward Algorithm", and connects it to a ton of other ideas e.g. neuromorphic engineering, one shot learning, self supervised learning, ...: https://www.cs.toronto.edu/~hinton/FFA13.pdf
It looks like #MachineLearning and #Neuroscience are really converging.
Gradients without Backpropagation
Using backpropagation to compute gradients of objective functions for optimization has remained a mainstay of machine learning. Backpropagation, or reverse-mode differentiation, is a special case within the general family of automatic differentiation algorithms that also includes the forward mode. We present a method to compute gradients based solely on the directional derivative that one can compute exactly and efficiently via the forward mode. We call this formulation the forward gradient, an unbiased estimate of the gradient that can be evaluated in a single forward run of the function, entirely eliminating the need for backpropagation in gradient descent. We demonstrate forward gradient descent in a range of problems, showing substantial savings in computation and enabling training up to twice as fast in some cases.
Most plasticity protocols I've seen reported (e.g. STDP, frequency-dependent) involve multiple (e.g. 50) repetitions of the stimulation protocol. Has anyone shown what happens if you do fewer repetitions or systematically investigated the relationship between induction repetitions and magnitude of plasticity?