A bugs life

There is another paper about a simple learning mechanism in an insect brain, retasked to learn the MNIST digits.
Another paper:

Ok, you can use dimension reducing random projections to denoise by effectively averaging over sensory inputs while not losing too much information. And you can also denoise by using dimension increasing random projections followed by a sparsity inducing operator. Winner takes all competitive responses are present. The system is put into overdrive by a reward system which triggers Hebbian learning of currently active neurons.