[Abishek_Muthian]

>programmers usually have to engage in a laborious process of manually adjusting the initial coefficients, or weights, that will be applied to each type of data point the network processes. Another challenge is to get the software to balance how much it should be trying to explore new solutions to a problem versus relying on solutions the network has already discovered that work well. “All these problems are completely eluded if you have a system that is based on biological neurons to begin with,” Friston said.

Using real neurons avoids hyperparameter tuning? Can someone explain how.

[alex7o]

I suppose neurons are already turned for the task at hand. Although I suppose you can change their environment with different chemicals and that would be very close to hyper parameter tuning, after all happy neurons think better.

[andbberger]

Well... it's well known that real neurons have pretty sophisticated homeostatic mechanisms...

But this is in vitro so all bets are off

[Mithriil]

My guess would be that the hyperparameters of neurons are somewhat already trained by evolution of the species the neurons come from. I assume that neurons are defined by their genetic and environment. Then if one reconstructs the environment in which neurons normally dwell, then everything is set for this neuron to perform as it should be.

But on the other hand, one could interpret the environment variables as the hyperparameters. If you overheat the chips, it is possible the neurons might act differently (as would a brain?) If you overfeed them with, say, creatine, it might be possible that the neurons will perform erratically.

[dr1337]

Are you tuning hyperparameters when learning a task?

[barkingcat]

Yes, the amount of caffine, food (enzymes, proteins, carbs), and if it is a long running learning process, amount of sleep. all parameters are being tuned all the time.