Feeling like God – understanding the design in detail, but not the outcomes: Neural Networks


(First of all, to clarify any potential for misunderstandings: I’m not religious, I do not believe in an almighty God – for that to happen, the teodice problem should have a clear and understandable solution, among other things…)

Ok, back to programming, technology and artificial intelligence:

To beat the by now almost dead horse of experimenting with neural networks and image recognition even further, I fine tuned my neural network filters, in an attempt to improve performance from the previous ‘lowly’ 76% target recognition (which as it happens is way better than expected…)

By some tuning of the Sigmoid filters, and more importantly, adding about 50% more images (from ~100 to ~150) to the training set, I was able to get the total target recognition correctness percentage from 76% to 84%.

The target recognition of the training set itself is just above 90%, while the target recognition of images outside of the training set is about 81%, that is, the program is able to classify 4 out of 5 previously unseen images by gender…! From 100 pixels…!

My attempts to understand how this is possible is really starting to hurt my brain…

What’s really disturbing is that I have absolutely no idea on how the neural network program manages to solve the problem… my head is burning after hours and days of attempts to figure out how the heck the evolution actually operates, but to no avail… While I do fully understand how the evolutionary process works (after all, I designed it, feeling quite God-like!) I feel quite uncomfortable with getting the results I was hoping for, but without understanding how those results come about…

I’m feeling like God understanding in perfect detail the design of his work, but not having any understanding for the the actual outcome… Looking at the evolved coefficients just makes me dizzy…

If I’d belive in a God – which I don’t – I’d feel symphathetic to Him: this is how the Almighty must feel when observing the state of the world He has created…. so clean design, so messy and incomprehensible outcome…

So, I’m starting to feel a bit like Dave in the Kubrick movie:

Dave Bowman: Open the pod bay doors, HAL. 
HAL: I’m sorry, Dave. I’m afraid I can’t do that. 

Perhaps this type of exercise is the ultimate proof for the world we inhabit being non-linear, complex, non-deterministic and stochastic: regardless of whether we understand the design, the outcomes will still confuse us.

About swdevperestroika

High tech industry veteran, avid hacker reluctantly transformed to mgmt consultant.
This entry was posted in Complex Systems, development, Math, Non Linear Dynamic Systems, software, Systems, Technology and tagged , , , , , , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s