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.

Hybrid computing using a neural network with dynamic external memory

153 pointsby idunningover 8 years ago

15 comments

the_deciderover 8 years ago
Some interesting ideas sadly blocked behind a pay-wall journal, all for the purpose of boosting a researcher&#x27;s prestige because they now hold a &quot;Nature&quot; publication. Thankfully, this article is easily accessible via Sci-Hub. <a href="http:&#x2F;&#x2F;www.nature.com.sci-hub.cc&#x2F;nature&#x2F;journal&#x2F;vaop&#x2F;ncurrent&#x2F;full&#x2F;nature20101.html" rel="nofollow">http:&#x2F;&#x2F;www.nature.com.sci-hub.cc&#x2F;nature&#x2F;journal&#x2F;vaop&#x2F;ncurren...</a>
评论 #12695610 未加载
评论 #12703438 未加载
nlover 8 years ago
This is probably the most important research direction in modern neural network research.<p>Neural networks are great at pattern recognition. Things like LSTMs allow pattern recognition through time, so they can develop &quot;memories&quot;. This is useful in things like understanding text (the meaning of one word often depends on the previous few words).<p>But how can a neural network know &quot;facts&quot;?<p>Humans have things like books, or the ability to ask others for things they don&#x27;t know. How would we build something analogous to that for neural network-powered &quot;AIs&quot;?<p>There&#x27;s been a strand of research mostly coming out of Jason Weston&#x27;s Memory Networks research[1]. This extends on that by using a new form of memory, and shows how it can perform at some pretty difficult tasks. These included graph tasks like London underground traversal.<p>One good quote showing how well it works:<p><i>In this case, the best LSTM network we found in an extensive hyper-parameter search failed to complete the first level of its training curriculum of even the easiest task (traversal), reaching an average of only 37% accuracy after almost two million training examples; DNCs reached an average of 98.8% accuracy on the final lesson of the same curriculum after around one million training examples.</i><p>[1] <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1410.3916v11.pdf" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;pdf&#x2F;1410.3916v11.pdf</a>
评论 #12700598 未加载
评论 #12699529 未加载
idunningover 8 years ago
Blog post for the paper: <a href="https:&#x2F;&#x2F;deepmind.com&#x2F;blog&#x2F;differentiable-neural-computers&#x2F;" rel="nofollow">https:&#x2F;&#x2F;deepmind.com&#x2F;blog&#x2F;differentiable-neural-computers&#x2F;</a>
triplefloatover 8 years ago
Very exciting extension of Neural Turing Machines. As a side note: Gated Graph Sequence Neural Networks (<a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1511.05493" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1511.05493</a>) perform similarly or better on the bAbI tasks mentioned in the paper. The comparison to existing graph neural network models apparently didn&#x27;t make it into the paper (sadly).
gallerdudeover 8 years ago
Can someone explain what the full implications of this are? This seems really cool, but I can&#x27;t really wrap my head around it.<p>From what I can tell you can give the DNC simple inputs and it can derive complex answers.
评论 #12695686 未加载
bra-ketover 8 years ago
if you&#x27;re interested in this check out &quot;Reasoning, Attention, Memory (RAM)&quot; NIPS Workshop 2015 organized by Jason Weston (Facebook Research): <a href="http:&#x2F;&#x2F;www.thespermwhale.com&#x2F;jaseweston&#x2F;ram&#x2F;" rel="nofollow">http:&#x2F;&#x2F;www.thespermwhale.com&#x2F;jaseweston&#x2F;ram&#x2F;</a>
footaover 8 years ago
I have a couple questions that I&#x27;m not getting from this, does this memory persist between each &quot;instance&quot; of a task? Or does it get wiped out after each one? Is this something where you might say present the model with some data that is the input (which it might learn to then store in memory) and then ask a question of it?<p>i.e, in the blog post it discusses using the network to find the shortest path between two stations, would the steps to do that look like this?<p>1. Train the NN how to navigate any network, presenting the graph data each time you ask the NN a problem 2. take the trained NN and feed it the London Underground, then ask it to tell you how to get there?
zardoover 8 years ago
Instead of saving the data, you could think of using a memory address as applying the identity function and saving the data.<p>Could it learn to use addresses that perform more interesting functions than f(x)=x?
kylekover 8 years ago
I&#x27;m probably totally off base here (neural networks&#x2F;AI is not my wheelhouse), but is having &quot;memory&quot; in neural networks a new thing? Isn&#x27;t this just a different application of a more typical &#x27;feedback loop&#x27; in the network?
评论 #12695651 未加载
评论 #12695632 未加载
评论 #12695586 未加载
评论 #12696329 未加载
gallerdudeover 8 years ago
Does this mean we could get way better versions of char-rnn?
评论 #12696303 未加载
bluetwoover 8 years ago
One of the examples given is a block puzzle (reorder 8 pieces in a 3x3 grid back into order)<p>Has this been a problem for AI and CNN&#x27;s?
评论 #12696874 未加载
0xdeadbeefbabeover 8 years ago
&gt; a DNC can complete a moving blocks puzzle in which changing goals are specified by sequences of symbols<p>A neural network without memory can&#x27;t do that or can&#x27;t do it as well perhaps?
评论 #12695705 未加载
prats226over 8 years ago
Would love to see if these networks learn concepts of fast retrieval for eg indexing etc
plgover 8 years ago
but why use an ANN for tasks involving symbolic logic? I don&#x27;t get it. It&#x27;s like ANNs are jumping the shark
ktamiolaover 8 years ago
This is remarkable!
评论 #12696305 未加载