d****7
发帖数: 109 | | m*****g
发帖数: 12253 | | l***y
发帖数: 4671 | 3 哈哈，赞！
【在 d****7 的大作中提到】
: 好玩！
| b*******r
发帖数: 361 | 4
【在 l***y 的大作中提到】
: 哈哈，赞！
| m***e
发帖数: 2 | | h*****u
发帖数: 204 | 6 interesting
【在 d****7 的大作中提到】
: 好玩！
| c****t
发帖数: 19049 | 7 from scipy import SVM ... 强啊 | l**********e
发帖数: 336 | 8 ft, this is not ppl working in ML, this is just some ppl use ML
this figure is a joke for CS ML/DM PhD
【在 d****7 的大作中提到】
: 好玩！
| d****n
发帖数: 12461 | 9 So here is the dilemma:
如果算法的名字大家从来没听说过，那你肯定不在做machine learning;
如果算法的名字大家都听说过，那你是在用machine learning，也不是在做machine
learning
【在 l**********e 的大作中提到】
: ft, this is not ppl working in ML, this is just some ppl use ML
: this figure is a joke for CS ML/DM PhD
| l**********e
发帖数: 336 | 10 ... this is simply not true
for instance, do u know deep learning networks 5 years ago? probably not,
and on that time, some machine learning researchers work on DL, and now, ...
, everybody know this
frankly, for ppl come from non-CS background, they often only know some
tools, and do not know
(1) who originally invented the algorithm, and
(2) who proposed the efficient algorithms for the optimization, and
(3) who successfully apply these algorithms to some important problems
【在 d****n 的大作中提到】
: So here is the dilemma:
: 如果算法的名字大家从来没听说过，那你肯定不在做machine learning;
: 如果算法的名字大家都听说过，那你是在用machine learning，也不是在做machine
: learning
| | | l*******m
发帖数: 1096 | 11 Most top Kaggle players don't have CS degrees.
..
【在 l**********e 的大作中提到】
: ... this is simply not true
: for instance, do u know deep learning networks 5 years ago? probably not,
: and on that time, some machine learning researchers work on DL, and now, ...
: , everybody know this
: frankly, for ppl come from non-CS background, they often only know some
: tools, and do not know
: (1) who originally invented the algorithm, and
: (2) who proposed the efficient algorithms for the optimization, and
: (3) who successfully apply these algorithms to some important problems
| l**********e
发帖数: 336 | 12 if u think these top kaggle players are top ML expert ... this is another
evidence u need know more about the community
【在 l*******m 的大作中提到】
: Most top Kaggle players don't have CS degrees.
:
: ..
| l*******m
发帖数: 1096 | 13 top kaggle players may not be top ML experts. However, ML experts should
easily win prizes by using their powerful algorithms. Why not get the easy
money?
【在 l**********e 的大作中提到】
: if u think these top kaggle players are top ML expert ... this is another
: evidence u need know more about the community
| l**********e
发帖数: 336 | 14 top ML experts work in G/F/etc, or Wall Street, or stay in university like
CMU/Stanford/MIT, so ...
also, actually the prize is not big enough, we can recall the $1M Netflix
prize, this is a serious contest and have big industry impact
who got this one? ppl come from AT&T Labs & Research
【在 l*******m 的大作中提到】
: top kaggle players may not be top ML experts. However, ML experts should
: easily win prizes by using their powerful algorithms. Why not get the easy
: money?
| l*******m
发帖数: 1096 | 15 For the Netflix prize, the SVD algorithm was initially invented by Simon
Funk, who was kindly sharing his idea online at that time @http://sifter.org/~simon/journal/20061211.html, which inspired ATT guys. He didn't give a beautiful name. ATT guys presented the idea and developed it nicely. They also cited this post in their papers. Netflix described the algorithm like "One of the most interesting findings during the Netflix Prize came out of a blog post".
People usually don't know how these idea were generated.
【在 l**********e 的大作中提到】
: top ML experts work in G/F/etc, or Wall Street, or stay in university like
: CMU/Stanford/MIT, so ...
: also, actually the prize is not big enough, we can recall the $1M Netflix
: prize, this is a serious contest and have big industry impact
: who got this one? ppl come from AT&T Labs & Research
| l**********e
发帖数: 336 | 16 ... please, basic logic
(1) no one said purely origin algorithms are needed for these contest,
contest is not like publish milestone academic papers (e.g., CRF in imcl
2001), also AT&T team cited simon's results, so?
(2) Simon Funk is also a CS guy, computer Science degree from UCSD!
【在 l*******m 的大作中提到】
: For the Netflix prize, the SVD algorithm was initially invented by Simon
: Funk, who was kindly sharing his idea online at that time @http://sifter.org/~simon/journal/20061211.html, which inspired ATT guys. He didn't give a beautiful name. ATT guys presented the idea and developed it nicely. They also cited this post in their papers. Netflix described the algorithm like "One of the most interesting findings during the Netflix Prize came out of a blog post".
: People usually don't know how these idea were generated.
| l*******m
发帖数: 1096 | 17 I don't think with a BA of CS Simon learned ML in school. His is smart. My
point is that IQ determines the technical achievements in ML, which is not
very correlated with CS background. Of course, You can say smarter ppl study
CS.
BTW, the meaning of Kaggle competitions is a proof of themselves rather than
money for ML experts. Geoff Hinton entered one and proved himself.
【在 l**********e 的大作中提到】
: ... please, basic logic
: (1) no one said purely origin algorithms are needed for these contest,
: contest is not like publish milestone academic papers (e.g., CRF in imcl
: 2001), also AT&T team cited simon's results, so?
: (2) Simon Funk is also a CS guy, computer Science degree from UCSD!
| l**********e
发帖数: 336 | 18 (1) Simon is a CS guy, not stats, this is my point (clearly he has high IQ )
(2) AT&T (CS guy for sure), won the Netflix prize (a serious contest), not
stats, this my point too
(3) if Geoff Hinton (CS professor) entered one game in kaggle, probably he
is just for fun
ML experts do not need kaggle to prove themseleves, they need real impact (
industry products like Kinect, or good nips & imcl papers)
btw, can u give some example of stats ppl made significant tech
contributions in G/F/, or any IT firms?
study
than
【在 l*******m 的大作中提到】
: I don't think with a BA of CS Simon learned ML in school. His is smart. My
: point is that IQ determines the technical achievements in ML, which is not
: very correlated with CS background. Of course, You can say smarter ppl study
: CS.
: BTW, the meaning of Kaggle competitions is a proof of themselves rather than
: money for ML experts. Geoff Hinton entered one and proved himself.
| d****n
发帖数: 12461 | 19 这不过是Hinton对NN的改良吧，起了个新名字而已。
..
【在 l**********e 的大作中提到】
: ... this is simply not true
: for instance, do u know deep learning networks 5 years ago? probably not,
: and on that time, some machine learning researchers work on DL, and now, ...
: , everybody know this
: frankly, for ppl come from non-CS background, they often only know some
: tools, and do not know
: (1) who originally invented the algorithm, and
: (2) who proposed the efficient algorithms for the optimization, and
: (3) who successfully apply these algorithms to some important problems
| X******2
发帖数: 5859 | 20
真是想当然，那三个人中两个是统计PhD出身。
Volinsky是Washington，Bell是斯坦福。
而且Volinsky在AT&T也是在统计部门。
【在 l**********e 的大作中提到】
: (1) Simon is a CS guy, not stats, this is my point (clearly he has high IQ )
: (2) AT&T (CS guy for sure), won the Netflix prize (a serious contest), not
: stats, this my point too
: (3) if Geoff Hinton (CS professor) entered one game in kaggle, probably he
: is just for fun
: ML experts do not need kaggle to prove themseleves, they need real impact (
: industry products like Kinect, or good nips & imcl papers)
: btw, can u give some example of stats ppl made significant tech
: contributions in G/F/, or any IT firms?
:
| | | X******2
发帖数: 5859 | 21 看来你还真是无知。
搞出SVM以及开创统计机器学习这个领域的Vapnik就是统计
出身。工业界缺省标准的logistic regression是统计的，
不少机器学习方法的理解也是统计的搞出来的。
你这种区分CS还是统计其实很无聊。
【在 l**********e 的大作中提到】
: (1) Simon is a CS guy, not stats, this is my point (clearly he has high IQ )
: (2) AT&T (CS guy for sure), won the Netflix prize (a serious contest), not
: stats, this my point too
: (3) if Geoff Hinton (CS professor) entered one game in kaggle, probably he
: is just for fun
: ML experts do not need kaggle to prove themseleves, they need real impact (
: industry products like Kinect, or good nips & imcl papers)
: btw, can u give some example of stats ppl made significant tech
: contributions in G/F/, or any IT firms?
:
| l**********e
发帖数: 336 | 22 we just said industry contributions, ok? not academic
back to academic, there is no (almost) ML groups in any place 30 years ago,
so ppl come from different areas, e.g, Math, Physics (String theory), CS,
Stat
but for now, most of ML groups are in IT firms & Labs /CS dept, etc
btw, for logistic regression, this is a straightforward method and ppl know
this for years, the reason that LR is so popular in industry is because of
the high quality LR implementation (can handle large-scale data, missing
data, etc) , and we all know who code up this
also, GOOG start use DNN as the default classifier to replace LR
lastly, u know nothing for this field and the community
【在 X******2 的大作中提到】
: 看来你还真是无知。
: 搞出SVM以及开创统计机器学习这个领域的Vapnik就是统计
: 出身。工业界缺省标准的logistic regression是统计的，
: 不少机器学习方法的理解也是统计的搞出来的。
: 你这种区分CS还是统计其实很无聊。
| l**********e
发帖数: 336 | 23 if this is 改良 ... then every thing is just 改良 ...
there are many deep contributions of DNN in the past 5 years
in industry, firms like GOOG start use DNN for the default classification
tools (not simple logistic regression any more)
put in this way, feature extraction + linear method is not the preferred
choice now, we are in the age of Revolution of ML and big data
btw, for DNN, besides theory (not much now, tough to get solid theory for
this type of works), most important things are algorithm, infrastructure,
multi-core computing, etc, all hardcore CS issues
【在 d****n 的大作中提到】
: 这不过是Hinton对NN的改良吧，起了个新名字而已。
:
: ..
| g******e
发帖数: 140 | 24 WL
..
【在 l**********e 的大作中提到】
: ... this is simply not true
: for instance, do u know deep learning networks 5 years ago? probably not,
: and on that time, some machine learning researchers work on DL, and now, ...
: , everybody know this
: frankly, for ppl come from non-CS background, they often only know some
: tools, and do not know
: (1) who originally invented the algorithm, and
: (2) who proposed the efficient algorithms for the optimization, and
: (3) who successfully apply these algorithms to some important problems
| X******2
发帖数: 5859 | 25 新年好。
吵架无益身心，老夫承认所知不多。
CS不是世界的全部即使是高科技领域，Deep learning也不是，Google更不是。工业界
呆过的都知道，CS再牛也牛不过business model。
,
know
【在 l**********e 的大作中提到】
: we just said industry contributions, ok? not academic
: back to academic, there is no (almost) ML groups in any place 30 years ago,
: so ppl come from different areas, e.g, Math, Physics (String theory), CS,
: Stat
: but for now, most of ML groups are in IT firms & Labs /CS dept, etc
: btw, for logistic regression, this is a straightforward method and ppl know
: this for years, the reason that LR is so popular in industry is because of
: the high quality LR implementation (can handle large-scale data, missing
: data, etc) , and we all know who code up this
: also, GOOG start use DNN as the default classifier to replace LR
| c****t
发帖数: 19049 | | l*******m
发帖数: 1096 | 27 那给我发个包子吧。我就是个好学的高中生, 假装学统计，就是想吧统计的人拉来吵架。
【在 c****t 的大作中提到】
: 吵架有益流量。加油！
| c****t
发帖数: 19049 | 28 不够给力啊。继续继续
架。
【在 l*******m 的大作中提到】
: 那给我发个包子吧。我就是个好学的高中生, 假装学统计，就是想吧统计的人拉来吵架。
| X******2
发帖数: 5859 | 29 你来添把火吧。
【在 c****t 的大作中提到】
: 不够给力啊。继续继续
:
: 架。
| j****x
发帖数: 943 | 30 Like you said, no theory for DNN. Isn't this where the Math guys come in.
【在 l**********e 的大作中提到】
: if this is 改良 ... then every thing is just 改良 ...
: there are many deep contributions of DNN in the past 5 years
: in industry, firms like GOOG start use DNN for the default classification
: tools (not simple logistic regression any more)
: put in this way, feature extraction + linear method is not the preferred
: choice now, we are in the age of Revolution of ML and big data
: btw, for DNN, besides theory (not much now, tough to get solid theory for
: this type of works), most important things are algorithm, infrastructure,
: multi-core computing, etc, all hardcore CS issues
| | | c****t
发帖数: 19049 | 31 lovelyminnie是deep learning的disciple。小心他跟你拼命。
【在 j****x 的大作中提到】
: Like you said, no theory for DNN. Isn't this where the Math guys come in.
| l*******m
发帖数: 1096 | 32 不是convex，理论很难搞。当然就没理论才能呼吁，因为大家也不知道最好是多好。现
在AI回热，一个根正苗红的deep learning PHD, 都是40＋w的包袱
【在 j****x 的大作中提到】
: Like you said, no theory for DNN. Isn't this where the Math guys come in.
|
|
