- shima__shima
- 9203
- 36
- 8
- 4
シミュレーション世界から実世界への転移学習、なんとなく夢がある気がするぞ(ぶれーくするーの一つになり得る気がする)(てきとー)
2016-08-18 01:02:35Creativity: infinite use of finite means #Humboldt @NandoDF #kdd2016
2016-08-18 01:03:47.@NandoDF mentioning @riedelcastro's work on differentiable Forth interpreter arxiv.org/abs/1605.06640 #kdd2016
2016-08-18 01:04:00Folks looking at assembling the network architecture on the fly, based on problem at hand #kdd2016 @NandoDF
2016-08-18 01:04:43.@NandoDF attempts at compositionality for NNs - not there yet, but we know it's important #kdd2016 pic.twitter.com/mhs0W5jcmS
2016-08-18 01:08:27Great talk at #kdd2016 by @NandoDF putting the whole conference in context. pic.twitter.com/f8H2lEj3Fl
2016-08-18 01:08:33.@NandoDF: teaching NNs to program - training data is execution traces, output is generated code #kdd2016 pic.twitter.com/WGNqCT5Zhq
2016-08-18 01:10:14Impressed by @NandoDF's ability to keep talking with no slides due to technical issues. Now fixed #kdd2016 pic.twitter.com/N9CRAPhN1j
2016-08-18 01:12:53.@NandoDF on key attributes of NPI, neural nets with recursion. Let's not forget the KD part of #kdd2016 pic.twitter.com/Xf2Fbz0ogY
2016-08-18 01:18:32The holy grail: @NandoDF on using an NN to learn NNs. Learn to learn to learn (turtles all the way down?) #kdd2016 pic.twitter.com/eaEd2c0i90
2016-08-18 01:19:57The holy grail: @NandoDF on using an NN to learn NNs. Learn to learn to learn (turtles all the way down?) #kdd2016 pic.twitter.com/cEam3CHweA
2016-08-18 01:20:01Using neural nets to train other neural nets. Nando de Freitas. #kdd2016 pic.twitter.com/4zv7oDkMyB
2016-08-18 01:20:25.@NandoDF Tuning two neural nets side by side, optimizer and optimizee #kdd2016 pic.twitter.com/xQnzQTlCeK
2016-08-18 01:21:32NPI: training network to learn progamming. Nando de Freitas #kdd2016 pic.twitter.com/pJ1wJeEb3Q
2016-08-18 01:22:32.@NandoDF: Stop hand engineering neural nets. Works for narrow domain but will fail outside it #kdd2016 pic.twitter.com/eJ73yJxvE1
2016-08-18 01:23:37My Q: what should people new to NNs look at to keep our jobs? @NandoDF: Look at online lecture notes. Societal issues do matter #kdd2016
2016-08-18 01:35:20.@NandoDF clarification, using a NN to train a NN works in narrow domains but fails outside of it. #kdd2016
2016-08-18 01:36:50最後の基調講演はAlpha GO などに関連したde Freitasさん.2014年のIBISにおいでいただいた ibisml.org/ibis2014/detai… ネオコグニトロンの福島とお話されているときキラキラの少年の目になってて,しかも女性研究者からイケメン認定を獲得していた
2016-08-18 02:42:16今の深層学習はレゴブロックを組み立てるようになったという出だしで,前半はCNNすごいとか,seq2seq とか,GANとかの普通の話.フラクタル拡張ルールとかは知らなかった.attentionというのもあったが,理解できなかった.
2016-08-18 02:42:41後半は,文字の○の部分とかを学習したものを組み合わせて学習する方法.NPIというスタックのついたチューリングマシンのようなマシンで問題解決をする方法.最後は効率のいい最適化というのをやってたが,あまりついていけなかった.
2016-08-18 02:42:52Scalable and fine grained recommendations with GLMix at Linkedin #kdd2016 pic.twitter.com/Bi9vGxYKKs
2016-08-18 02:46:16twitter.com/ShirKi/status/… 本会議2日目のVCの Papadopoulos さんの講演まとめ
2016-08-18 02:54:01My notes incl slide photos from yesterday's #VC talk by Gregory Papadopoulos are available here evernote.com/shard/s101/sh/… #kdd2016 #startup
2016-08-18 00:39:56#KDD2016 what an explanation could look like pic.twitter.com/zvtotssHwq
2016-08-18 03:01:23