<html><body><div style="color:#000; background-color:#fff; font-family:arial, helvetica, sans-serif;font-size:12pt"><div><span>I think we could use a conventional NN package, but we'd have to modify it to label some layers as non changeable and read in a starting model.</span></div><div><span><br></span></div><div><span>i'd been wanting to look into this - it would be awesome. </span></div><div><br></div><div style="font-size: 12pt; font-family: arial, helvetica, sans-serif; "><div style="font-size: 12pt; font-family: 'times new roman', 'new york', times, serif; "><font size="2" face="Arial"><hr size="1"><b><span style="font-weight:bold;">From:</span></b> Mike Schachter <mike@mindmech.com><br><b><span style="font-weight: bold;">To:</span></b> Ron Shigeta <rtshigeta@yahoo.com><br><b><span style="font-weight: bold;">Cc:</span></b> "ml@lists.noisebridge.net" <ml@lists.noisebridge.net><br><b><span style="font-weight:
bold;">Sent:</span></b> Friday, July 22, 2011 8:42 AM<br><b><span style="font-weight: bold;">Subject:</span></b> Re: [ml] Deep Learning ?<br></font><br>Sounds good, maybe we could meet up some time in August and<br>start putting something together. In the meantime I'll look into packages<br>we could use to train deep nets, either deep belief nets or stacked<br>auto-encoders or both. PyBrain claims to have support for deep nets:<br><br><a href="http://pybrain.org/docs/" target="_blank">http://pybrain.org/docs/</a><br><br>However, any neural network library should have support for deep nets,<br>because the pre-training just involves training layers of auto-encoders,<br>and each auto-encoder is just a normal neural network that has been<br>trained to reproduce it's input, with the output layer removed.<br><br> mike<br><br><br>On Fri, Jul 22, 2011 at 11:26 AM, Ron Shigeta <<a ymailto="mailto:rtshigeta@yahoo.com"
href="mailto:rtshigeta@yahoo.com">rtshigeta@yahoo.com</a>> wrote:<br>> I'm just starting to work with ML packages and have read a couple of deep<br>> learning papers, but would love to see some examples. if it just turns into<br>> a working session with some basic data we've done i'd definitely show up for<br>> that too.<br>><br>><br>> ________________________________<br>> From: Mike Schachter <<a ymailto="mailto:mike@mindmech.com" href="mailto:mike@mindmech.com">mike@mindmech.com</a>><br>> To: Ron Shigeta <<a ymailto="mailto:rtshigeta@yahoo.com" href="mailto:rtshigeta@yahoo.com">rtshigeta@yahoo.com</a>><br>> Cc: "<a ymailto="mailto:ml@lists.noisebridge.net" href="mailto:ml@lists.noisebridge.net">ml@lists.noisebridge.net</a>" <<a ymailto="mailto:ml@lists.noisebridge.net" href="mailto:ml@lists.noisebridge.net">ml@lists.noisebridge.net</a>><br>> Sent: Thursday, July 21, 2011 10:07 PM<br>>
Subject: Re: [ml] Deep Learning ?<br>><br>> Hi Ron,<br>><br>> Yes, quite ambiguous! I'm down for a deep net meetup,<br>> maybe even a workshop, but haven't put together any<br>> of the materials and haven't used them for anything yet.<br>><br>> Have you? Anyone else? I've heard of theano for using<br>> them but don't know of any formal frameworks that do<br>> the pre-training. I guess for a workshop or meetup we<br>> could just get an arbitrary neural net package and train<br>> layers of auto-encoders then stack them up? Any ideas?<br>><br>> mike<br>><br>> On Thu, Jul 21, 2011 at 4:11 PM, Ron Shigeta <<a ymailto="mailto:rtshigeta@yahoo.com" href="mailto:rtshigeta@yahoo.com">rtshigeta@yahoo.com</a>> wrote:<br>>> Is there going to be a deep networks session at noisebridge in august?<br>>> would really like to come...but the calendar was ambiguous - at least the<br>>> part I
read<br>>><br>>> _______________________________________________<br>>> ml mailing list<br>>> <a ymailto="mailto:ml@lists.noisebridge.net" href="mailto:ml@lists.noisebridge.net">ml@lists.noisebridge.net</a><br>>> <a href="https://www.noisebridge.net/mailman/listinfo/ml" target="_blank">https://www.noisebridge.net/mailman/listinfo/ml</a><br>>><br>>><br>><br>><br>><br><br><br></div></div></div></body></html>