TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Exploring Weight Agnostic Neural Networks

179 pointsby lamchobover 5 years ago

13 comments

elamjeover 5 years ago
Unpopular quote from my image and video processing professor - “The only problem with machine learning is that the machine does the learning and you don’t.”<p>While I understand that is missing a lot of nuance, it has stuck with me over the past few years as I feel like I am missing out on the cool machine learning work going on out there.<p>There is a ton of learning about calculus, probability, and statistics when doing machine learning, but I can’t shake the fact that at the end of the day, the output is basically a black box. As you start toying with AI you realize that the only way to learn from your architecture and results is by tuning parameters and trial and error.<p>Of course there are many applications that only AI can solve, which is all good and well, but I’m curious to hear from some heavy machine learning practitioners - what is exciting to you about your work?<p>This is a serious inquiry because I want to know if it’s worth exploring again. In the past university AI classes I took, I just got bored writing tiny programs that leveraged AI libraries to classify images, do some simple predictions etc.
评论 #20828708 未加载
评论 #20827220 未加载
评论 #20827437 未加载
评论 #20827283 未加载
评论 #20827333 未加载
评论 #20827314 未加载
评论 #20827694 未加载
评论 #20827061 未加载
评论 #20831983 未加载
评论 #20827888 未加载
baylearnover 5 years ago
Previous discussion (about the actual research article at <a href="https:&#x2F;&#x2F;weightagnostic.github.io&#x2F;" rel="nofollow">https:&#x2F;&#x2F;weightagnostic.github.io&#x2F;</a> rather than the blog post):<p><a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=20160693" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=20160693</a>
antplsover 5 years ago
How is it different than pruning a neural network?<p>It seems you could train the weights of a state of the art NN, then quantizite it, then prune it. It will remove some weights of the NN, then all the remaining weights are set to the same value. Isn&#x27;t training then pruning more efficient than using an architecture search algorithm ?
评论 #20849433 未加载
评论 #20827300 未加载
phaedrusover 5 years ago
I wrote a series of Markov chat simulators as a teenager. Often I used a simpler algorithm which ignored the probability weight (all out-links, once learned, given equal probability). These version performed subjectively as well as, if not better, than the versions which tracked the weight of links. I&#x27;m not surprised therefore that weight agnostic neural networks can work, too.
评论 #20827263 未加载
jangidover 5 years ago
The analogy given in the article is interesting. Some organisms perform certain actions even before they start to learn. I myself have seen some animals start running immediately after birth. Less number of parameters (shared parameters) could also be thought of as less complexity and hence less processing power requirements; which implies faster training. Phew! too much similarity.
nurettinover 5 years ago
<a href="https:&#x2F;&#x2F;github.com&#x2F;google&#x2F;brain-tokyo-workshop&#x2F;tree&#x2F;master&#x2F;WANNRelease&#x2F;prettyNEAT" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;google&#x2F;brain-tokyo-workshop&#x2F;tree&#x2F;master&#x2F;W...</a> to me, this is the really interesting part of the article. NEAT (neuro-evolution of augmenting topologies) is an algorithm for GANN. For those who are looking to implement the algorithm from scratch, see <a href="http:&#x2F;&#x2F;nn.cs.utexas.edu&#x2F;downloads&#x2F;papers&#x2F;stanley.ec02.pdf" rel="nofollow">http:&#x2F;&#x2F;nn.cs.utexas.edu&#x2F;downloads&#x2F;papers&#x2F;stanley.ec02.pdf</a> for hours of fun.
ilakshover 5 years ago
This seems like it has the potential for massive efficiency gains and maybe could help with better generalization if the much simpler networks could more easily be reused or recursed or something.
kolarover 5 years ago
How is this different from genetic programming?
评论 #20828730 未加载
评论 #20828630 未加载
评论 #20829026 未加载
s_Hoggover 5 years ago
I&#x27;m pretty sure this was posted a while back (maybe a month or two)?
评论 #20829935 未加载
scribuover 5 years ago
Discussion from 3 months ago: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=20160693" rel="nofollow">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=20160693</a>
TekMolover 5 years ago
Is each architecture given <i>one</i> set of random weights?<p>Or is the architecture of the net tested against a bunch of random weights so that it performs well independently of the weights?
评论 #20828873 未加载
评论 #20828857 未加载
DoctorOetkerover 5 years ago
how is this different from boring old evolutionary algorithms?<p>In my opinion the big breakthrough that enabled optimization and machine learning was the discovery of reverse mode automatic differentiation, since the space or family of all possible decision-functions is high dimensional, while the goal (survival, reproduction) is low dimensional. Unless I see a mathematical proof that evolutionary algorithms are as efficient as RM AD, I see little future in it, and apparently neither did biology since it decided to create brains.<p>It&#x27;s not an ideological stance I take here (of nature vs nurture).<p>For simplicity, lets pretend humans are single-cellular organisms, what does natural selection exert pressure on? our DNA code: both the actual protein codes and the promotor regions. I claim that variation on the proteins are risky (a modification in a proteinn coding region could render a protein useless) while a variation on the promotor regions is much less risky: altering a nucleotide there would slightly affect the affinity modulating transcription, so the cell would behave essentially the same but with different treshold concentrations, think of continuous parameters that describe our body (assuming same nurture, food, etc) some people are a bit taller, some people a bit stronger, etc... so how many of these continuous parameters do we have? On the order of the same number as the total number of promotor regions in DNA in the fertilized egg: both on human DNA and in one mitochondria (assuming there isn&#x27;t a chemical signals addressing and reading and writing scheme for say 10 mitochondria)...<p>EDIT: just adding that for a certain fixed environment, there are local (and a global) optimum of affinity values for each protein, so that near a local optimum the fitness is roughly shaped like -s(a-a_opt)^2 where s is spread and a_opt the local optimum affinity value. In other words, it is not so that &quot;better affinity&quot;, means fitter, not at all, a collection of genomes from an identical environment will hover around an affinity sweet spot.<p>According to wikipedia [0] that would result in about<p>about 2x 20412 &quot;floats&quot; for just protein-coding genes<p>about 2x 34000 &quot;floats&quot; when also including the pseudo-genes<p>about 2x 62000 &quot;floats&quot; when also including long ncRNA, small ncRNA, miRNA, rRNA, snRNA, snoRNA<p>these &quot;floats&quot; are the variables that allow a species to modulate the reaction constants in the gene regulatory network, since natural selection can not directly modulate the laws of physics and chemistry, and modulating the protein directly instead of the promotor region affinities &#x2F; reaction rates risks disfunctional proteins...<p>so my estimate of <i>an</i> upper limit of the number of &quot;floats&quot; in the genetic algorithm is ~120000 (and probably much less if not each of the above has a promotor region).<p>thats not a lot of information, if we think about the number of synaptic weights in the brain, and many of these are shared <i>in utilization</i> by the other cell types besides neurons.<p>I consider the possibility that: sperm cell, egg cell, or fertilized egg cell performs a kind of POST (power-on-self-test) that checks for some of the genes, although simply reaching the fertilized state may be enough of a selftest so no spontaneous abortion test may be needed (to save time and avoid resources spent on a probably malformed child).<p>[0] <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Human_genome#Molecular_organization_and_gene_content" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Human_genome#Molecular_organiz...</a><p>EDIT2: regarding:<p>&gt;This makes WANNs particularly well positioned to exploit the Baldwin effect, the evolutionary pressure that rewards individuals predisposed to learn useful behaviors, without being trapped in the computationally expensive trap of ‘learning to learn’.<p>The computationally expensive trap of having to &#x27;learn to learn&#x27; could end up being as mundane as a low number of hormones to which neurons in the brain globally or collectively respond, which enables learning by reward or punishment, and from then on anticipating reward or punishment, and our individual end goal stems from this anticipation, and anticipating the anticipation etc...
评论 #20827198 未加载
评论 #20827188 未加载
评论 #20827698 未加载
评论 #20828402 未加载
ianamartinover 5 years ago
Next thing you know, google will be telling us that the fastest websites are server side rendered from templates with minimal JavaScript.<p>How could anyone possibly have known?