Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Evolution Is the New Deep Learning

sentient.ai

Post reply on HN
Print view View original

151–160 of 242 posts

Re: Evolution Is the New Deep Learning

#151

Post by Houshalter »

inverse_pi wrote: post #103
mav3r1ck wrote: post #15

Having studied this extensively back when they were called Genetic Algorithms, I would like to offer a few insights. 1) One of the biggest reasons they fell out of favor for more "mathematical" approaches was that no one could really explain why exactly they worked. It makes sense on the surface that "survival of the fittest" and doing something akin to multiple stochastic gradient descents would work, but no one has…

My thesis was on Generic Algorithm. I stopped and started working on Deep Learning mainly because like you said, GAs don't really have a strong mathematical foundation. Ironically, no one could really explain why CNNs work mathematically either. I've heard a lot of hand-wavy arguments about local search, local sensitivity, etc. However, no one could really prove anything meaningful. There are some papers around certa…

Why should we expect there to be any mathematical foundation to this stuff? It's quite possible to imagine an alternate universe where GAs and neural nets don't work. Because they have different datasets that don't fit the structure of NNs well. Or problems that happen to not be solvable by the search strategies of GAs.

In fact we have many such problems in our own universe. I can give many examples of things NNs and GAs don't work well on right now. Those are just ignored by the research.

Re: Evolution Is the New Deep Learning

#152

Post by jcora »

pas wrote: post #73
resu_nimda wrote: post #64

Earlier quoted context omitted.

1) One of the biggest reasons they fell out of favor for more "mathematical" approaches was that no one could really explain why exactly they worked. Kind of like how nobody can really explain how the brain works, or life in general. My gut feeling is that it is hubris to think that we are going to "figure out" intelligence with increasingly sophisticated mathematical models anytime soon. We are not giving proper cre…

> We are not giving proper credit to how complex it is, and the multi-billion year developmental process that it took. Or we are simply not ready to accept that it's simply a big book of heuristics fine-tuned over biological eons. It's just big. We have too many interwoven, interdependent, synergistic faculties. Input, output, and a lot of mental stuff for making the right connections between the ins and the outs. Th…

I'd call this pulling a Dennett: trivializing complexity to something that cannot or just doesn't have to be explained. Being unable to conceive consciousness at this moment doesn't mean there's nothing to conceive of: even if we never get to the final satisfactory answer, there is undoubtedly much more room left for useful concepts we don't have yet, around or inside this idea

Re: Evolution Is the New Deep Learning

#153

Post by vanderZwan »

Houshalter wrote: post #149
guest12268 wrote: post #132

Earlier quoted context omitted.

Yes there has been some very interesting recent work. In particular, how evolvability emerges and is harnessed in evolutionary computation. A few papers come to mind: 1. Evolvability is Inevitable: http://journals.plos.org/plosone/article?id=10.1371/journal.... 2. Extinction Events can Accelerate Evolution (2015): http://journals.plos.org/plosone/article?id=10.1371/journal.... 3. Evolvability Search: Directly selecti…

Evolvability can also evolve away too. For instance, a gene that decreases the mutation rate to 0. Most mutations are harmful, so any organism with the gene will be more likely to have successful children. And eventually the gene will become dominant and there will be no more mutations. And evolution will stop.

What you say is fairly hypothetical, since the way genetics works at a fundamental level prevents this possibility.

You are also oversimplifying by not distinguishing local and global success. Locally, this gene will be successful. Globally, some mutations will be beneficial, and the children with those mutations will be more successful.

The gene may become "dominant" (and then gain the beneficial mutations through sexual reproduction), but it will not achieve a monopoly and not remove the existence of evolution.

It is somewhat comparable to why some people are left-handed[0].

However, what you allude to is basically the necessity of evolution of forms of "error correction" against mutations for complicated organisms. Sex plays a large part in that as well[1]. And interestingly, DNA repair, another involved mechanism, may actually help evolvability[2].

[0] https://www.youtube.com/watch?v=TGLYcYCm2FM

[1] https://www.quantamagazine.org/missing-mutations-suggest-a-r...

[2] https://www.quantamagazine.org/beating-the-odds-for-lucky-mu...

Re: Evolution Is the New Deep Learning

#154

Post by jcora »

dorgo wrote: post #120
resu_nimda wrote: post #64

Earlier quoted context omitted.

1) One of the biggest reasons they fell out of favor for more "mathematical" approaches was that no one could really explain why exactly they worked. Kind of like how nobody can really explain how the brain works, or life in general. My gut feeling is that it is hubris to think that we are going to "figure out" intelligence with increasingly sophisticated mathematical models anytime soon. We are not giving proper cre…

> My gut feeling is that it is hubris to think that we are going to "figure out" intelligence with increasingly sophisticated mathematical models anytime soon. We did it already. Compter understand language, translate it, react to it. They can recognize items on a picture. Is there a task left which can't be done by computers better and faster than by humans? >Almost by definition, if we can analytically understand i…

You should study ML a bit more, to see just how wrong you are about it

Re: Evolution Is the New Deep Learning

#155

Post by LMSchmitt »

Schmitt, Lothar M (2001), Theory of Genetic Algorithms, Theoretical Computer Science 259: 1–61

Schmitt, Lothar M (2004), Theory of Genetic Algorithms II: models for genetic operators over the string-tensor representation of populations and convergence to global optima for arbitrary fitness function under scaling, Theoretical Computer Science 310: 181–231

Re: Evolution Is the New Deep Learning

#156

Post by evolutioner »

nrub wrote: post #141
evolutioner wrote: post #124

Earlier quoted context omitted.

They can actually work together. Thompson Sampling / Multi-armed bandit algorithms are traffic allocation strategies to use the least amount of traffic to find the optimal variant. So say you have version A/B/../N variants of a website, you can split the traffic equally or use a a bandit algorithm and find the best performing variant of it. But that only helps you test N variants. If you want to test for 4 titles and…

The paper linked above does directly address the case of multiple experiments occurring in the same context. They address this with hill-climbing over those 180 different variations. The use of a bayesian linear regression takes place of the exploration found with Thompson sampling.

You're right, the paper linked above is a different way of solving the same problem. In their case they use a model to decide which website variants to show. Their model accounts for independent effects and pairwise dependencies. Evolution allows you to optimize without needing an explicit model.

I don't think they account for potentially changing conversion rates over time or delayed conversions.

Aside from that, I'd be curious to see how these two approaches compare in a real-life situation.

Re: Evolution Is the New Deep Learning

#157

Post by photojosh »

flavio81 wrote: post #136
dorgo wrote: post #120

Earlier quoted context omitted.

> My gut feeling is that it is hubris to think that we are going to "figure out" intelligence with increasingly sophisticated mathematical models anytime soon. We did it already. Compter understand language, translate it, react to it. They can recognize items on a picture. Is there a task left which can't be done by computers better and faster than by humans? >Almost by definition, if we can analytically understand i…

>I'm amazed every time when the training is over and it actually works like intended. Everything which is big enough is more than the sum of it's parts. What about when it doesn't work as intended and fails ridiculously, even though it usually works perfectly well? http://www.labsix.org/physical-objects-that-fool-neural-nets...

Humans do the same thing; not working as intended some of the time. It's just that the failure modes for ML are different, and so we see them as ridiculous.

Re: Evolution Is the New Deep Learning

#158

Post by hyperpallium »

hyperpallium wrote: post #119
mav3r1ck wrote: post #15

Having studied this extensively back when they were called Genetic Algorithms, I would like to offer a few insights. 1) One of the biggest reasons they fell out of favor for more "mathematical" approaches was that no one could really explain why exactly they worked. It makes sense on the surface that "survival of the fittest" and doing something akin to multiple stochastic gradient descents would work, but no one has…

> I felt that at the point you are understanding the problem, you may just be better off with a direct approach I formed a similar impression in my PhD research.

Not that relevant, but wanted to elaborate a little: I exhaustively solved a toy version of my problem on a cluster, but found no discernable gradients. Good solutions were isolated spikes, with no elevations adjacent that could lead to them. So, random sampling would be as good as you'd get.

The thing to do is change the problem, transform the space/dimensions, so better solutions were spatially proximate. But then, I'd be solving the problem.

Another approach is to have the computer do this, seach the space of search spaces. But this higher-level space is even less likely to have informative gradients.

OTOH, the data was in terms of a language, which would have introduced its own artefacts. A better search space would compensate for those, and might have been easy to find.

Re: Evolution Is the New Deep Learning

#159

Post by rpm13 »

rpm13 wrote: post #131
peterstoziek wrote: post #2

As expected, the article seems to be a typical content marketing piece. If you're looking for real insights into evolutionary algorithms, specifically "neuroevolution", I highly recommend to read this article: https://www.oreilly.com/ideas/neuroevolution-a-different-kin... I enjoyed it much more than - what feels like - a quickly thrown together marketing piece with no real value for the reader.

Note that this blog post is not an article per se, but an overview of a research website ( https://sentient.ai/sentient-labs/ea ) built around five new research papers. The website offers demos that illustrate neuroevolution and evolutionary computation concepts at a much more concrete level than the papers can.

For a short intro to neuroevolution there is http://www.scholarpedia.org/article/Neuroevolution

However, these overview articles do not include the newest research on evolving deep learning networks. Three such papers are introduced at https://www.sentient.ai/sentient-labs/ea-1/; there are other recent ones at https://research.googleblog.com/2018/03/using-evolutionary-a... and https://eng.uber.com/deep-neuroevolution/. It is a rapidly developing area.

Re: Evolution Is the New Deep Learning

#160

Post by eafan »

sewercake wrote: post #18
XnoiVeX wrote: post #6

A philosophical tangent: Evolution did not always produce the best outcomes. Would neuroevolution be vulnerable to similar effects? Probably a good research area. https://www.wired.com/2009/07/st-best-5/

Yeah, agreed -- evolution doesn't produce best, just good enough. But for large classes of 'problems', we can't (from a computational complexity standpoint) find best outcomes efficiently. 'Evolutionary' techniques, are, in some sense, 'trial and error'. For many tasks where we already have efficient algorithms, this isn't particularly useful. However, where we DONT have efficient algorithms, or know what the concept…

In a rainforest there is no ‘best’ solution - John Holland
Post reply on HN