2 / 0

6th April

The Mercury Neuron Mk 2 uses two different temperatures, the top one tracks the recent stimulation level and is used for feedback to the bottom temp which controls electrical signal transfer. Free to use, if you do then pls tag any saves using it WMN2.

brain learn intelligence learning wmn2

Comments

DRZOIDBERG 6th April

I think that this is an indication that you should increase the value of the reward
NotAChair 6th April

Something interesting I noticed is that even if I put a long DLAY on the positive feedback, it still eventually figures it out. On the other hand, if I add an option that causes one negative feedback and than a bunch of positive feedback, it will avoid it almost as much as just negative feedback.
NotAChair 6th April

This would also raise the top section temperature so you might want to swap the METL for NTCT/PTCT to keep it averaging around 100C (the point where feedback switches(negative feedback cools the bottom if the top is hot and heats the bottom if the top is cold, while positive feedback heats the bottom if the top is hot(more than 100C) and cools the bottom if the top is cold(less than 100C)))
NotAChair 6th April

If you want to increase the influential effects of rewards and punishment, then I would increase the temperature of the hot ARAY and decrease the temperature of the 22C ARAY.
DRZOIDBERG 6th April

for this i would reccomend reducing the amount of pixels and insl to decrease the thermal capacity allowing for more influential effect from rewards and punishments
DRZOIDBERG 6th April

although for things like this, this is completly negligable and inconcequential so it can be left as it is
DRZOIDBERG 6th April

since it found the path and it proivided it with the lowest deficit, discoraging them to find a better path as the memory can only be slowly altered and any increase in negetive reward would cause it to snap back to it previos route. whereas with short term it finds a local optimun and it is refreshed fast enough that a small deficit can be ignored if it finds another more optimal peth through it.
DRZOIDBERG 6th April

yes this is true, but you would also need short term learning aswell, as that allows a system to alter it's coarse more effectivly, since only have long term memory can result in it becoming locked to a path that may not be optimal and can sometimes even be a net negetive reward.
NotAChair 6th April

In this specific case, the COAL may be slowing down learning, but in a real situation, you would likely want a longer learning attention span, since the reward would likely take longer to arrive.
NotAChair 6th April

The INST is just a feature of the test network. There is no coal in the bottom/electrical transfer/feedback(reward & anti-reward) system, only as filler in the top/stimulation measurement/feedback control section.

The Powder Toy

WMN-02 Feedback Neuron Prototype by NotAChair

Comments