computation-thru-dynamics
computation-thru-dynamics copied to clipboard
Updated distribution of biases
With this change, biases are ~Uniform(-bval, bval). Previously, biases were drawn from a Gaussian and were not centered at 0.
Thanks @mattgolub! I've been aware of this, but there are two concerns about this. (1) This fix will change the noise distribution from gaussian to uniform distribution. Alternatively we can fix line 34 (i.e., removing -0.5). (2) The asymmetry might be intentional, in order to visualize an asymmetric line attractor manifold in the FixedPointFinder tutorial. @sussillo, could you comment on it?
Thanks @mattgolub! I've been aware of this, but there are two concerns about this. (1) This fix will change the noise distribution from gaussian to uniform distribution. Alternatively we can fix line 34 (i.e., removing -0.5). (2) The asymmetry might be intentional, in order to visualize an asymmetric line attractor manifold in the FixedPointFinder tutorial. @sussillo, could you comment on it?
Definitely not intentional. Oops! I'd suggest getting rid of the 2.0 and the - 0.5. Thank you, -David