Artificial Neural Networks – Introduction | Hacker News Reader

Artificial Neural Networks – Introduction

rzagabe.com

69 pointsrzagabe9 comments

This was a really nice article but I was anticipating more (an implementation example) at the end of it, hope you'll add more to it later!

The following quote from the author: "Our brains behave like monkeys and it can be proven that most of our lives are about "what am I going to have for lunch?"..."

has this been proved ? would be very interested to know who proved this and how they went about doing it.

> has this been proved ?

Yes!

> would be very interested to know who proved this and how they went about doing it.

Well. There's this thing you do. It only takes a few minutes. Then you wait about nine months. At that point you start collecting data. The first year or two are somewhat inconclusive. The data (and the subject) can be very noisy. By the time you've made observations and collected data for about eight years the conclusion is inescapable: Monkeys.

According to my observations across three of these experiments I would add that monkeys might sometimes have an edge over some of the crap we pull.

> Our brains behave like monkeys and it can be proven that most of our lives are about "what am I going to have for lunch?

Keyword being most, however this primitive impulse, like many primitive impulses such as wanting to strangle your boss when they make you upset, can be overcome with cognitive reappraisal. "I am not really that hungry, and I could even make it an entire day without eating and still survive" or "I don't really want to kill her, I would lose my job and go to prison"

That's impossible to prove because it has no concrete meaning -- there is no provable understanding of "what life is about", much less a provable way to actually answer that implicit question, "What is life about?"

Take all arguments that these models resemble our brain with a huge grain of salt. At best they're a metaphor.

There's one critical difference between biological neural networks and the artificial models they're talking about here: the biological neurons produce spikes when they activate, not continuously-variable activation levels.

These artificial models whitewash these features away by modelling only the neuron's firing rate and not the inter-spike timing or spike train behaviours.

There is this interesting paper titled “A Few Useful Things to Know about Machine Learning”, which basically ...

That link leads to the Japanese toilet maker Toto, not to any ML paper. I suspect it was pranksters, not the author.

The brain has no objective function it is trying to minimize

> Keep in mind that ANNs aren't about duplicating human brains and that a lot of aspects already don't correlate with our current, very short, understanding of them (i.e. learning algorithms, backpropagation,...)

> We'll see that artificial neural networks are not that different from our magnificent biological system "in their most basic definitions".

I've tried to make it clear enough that we should avoid analogies with biological neural systems.

Reply on news.ycombinator.com