Because the appreciation of nature is diverse does not mean it does not fulfil anything meaningful in one's life. One's life simply grows in the direction toward which one lay his eyes. What is to live if not the expansion of our conscious experience through goals (the unknown), effort (the knowing) and achievement (the known, the experience itself)?

Note that it also does provide an evolutionary advantage to the society as a whole.

While "climbing the status ladder" is an easy way to tarnish one's drive for the power process. What can be said of someone who's power process is to understand and appreciate the world?

Tis a bad suggestion. There is such a thing as wanting more of the good things while they last.

This "appeal to nature" argument is easily contradicted by nature itself. Our lives depend on whether or not our crops will freeze, on whether or not a cosmic ray burst will eradicate life on earth. As long as an individual is part of an environment which is distinct from him/her, this sense of powerlessness will arise.

The Empty Boat by Chuang Tzu is pretty relevant here.

"If a man is crossing a river And an empty boat collides with his own skiff, Even though he be a bad-tempered man He will not become very angry. But if he sees a man in the boat, He will shout at him to steer clear. If the shout is not heard, he will shout again, And yet again, and begin cursing. And all because there is somebody in the boat. Yet if the boat were empty. He would not be shouting, and not angry."

I would argue that in both cases, the boat is empty. It is always and only conducted by nature and necessity.

This position is hard to argue against. While it relies heavily on the assumption that surrogate activities do not fulfil real goals (differentiating them from pathological activities), much of one's life is indeed constrained by the societal norms required for to achieve a functioning and stable industrial society - thereby voiding the power process.

While this is obviously the case, given the competitive advantage of having technological tools regarding the fulfilment of basic needs; should we necessarily consider the societal evolutionary pressure to evolve along technology as something really worse than the natural environmental evolutionary pressure?

Again, here it is more or less implied that human nature has an unchanging core and while (at the moment) this is certainly true for our biological functions, it can be argued that much of what constitute our humanity is our culture and the psychology that comes with it. How flexible and adaptative can it be?

Same can be said for pretty much everything in existence.

The artificiality of those goals is valid only under the supposition that a "real" goal is one that is set in accordance to "real", fundamental needs of the human nature - which are presumed to be finite and definite. If we otherwise suppose the human nature is something to be evolved and constructed in a way that is always changing relative to its environment, in the end, not much really is a surrogate activity. I would argue that the need to create and evolve is as fundamental as to eat, sleep and drink.

I believe that a more precise designation for the tone of what is here called "surrogate activities" would be "pathological activities"; patterns that are repeated in search of some vacuous goal.

As mentioned, you may encounter several issues while trying the code below on Theano 0.9.0 (especially while trying to run on GPU).

Not all errors are obvious to fix. For instance, if your CUDA installation is not properly set up and you still try to run the code on GPU, you may experience very high memory usage with no special error message and the network will still be performing some learning until the memory runs out.

For this, I highly recommend, post Theano installation, to run the following test:

http://deeplearning.net/software/theano/tutorial/using_gpu.html#testing-theano-with-gpu

Another weird case is when trying to use the convolutional layers on GPU without the CUDNN lib installed (still on theano 0.9.0). The Theano function compilation will complain it failed to use CUDNN but the code will proceed without any error message and some abysmal 10% learning accuracy.

I have annotated the code below with relevant changes.

On Theano 0.9.0, replace with: pool.pool_2d(input=conv_out, ws=self.poolsize, ignore_border=True)

You may encounter several issues while trying to run the code experiments from network3 against version 0.9.0 and up of Theano. I have annotated the full code print out below to highlight the needed modifications.

On Theano 0.9.0, this line should be: from theano.tensor.signal import pool

On Theano 0.9.0, replace with: conv_out = theano.tensor.nnet.conv2d(input=self.inpt, filters=self.w, filter_shape=self.filter_shape, input_shape=self.image_shape)

As previously mentioned, if you try running this code on GPU, install CUDNN. Some relevant links:

https://askubuntu.com/questions/767269/how-can-i-install-cudnn-on-ubuntu-16-04 https://developer.nvidia.com/rdp/cudnn-download

A note from Theano API for the conv2d function says:

If cuDNN is available, it will be used on the GPU. Otherwise, it is the CorrMM convolution that will be used “caffe style convolution”

I am unsure if this is a bug with Theano but when I did not had cuDNN installed, the code proceeded but the learning accuracy was stuck at around 10%.