Profile

Cover photo
Michael Nielsen
Lives in New York City
4,972 followers|1,343,920 views
AboutPostsPhotosYouTube

Stream

Michael Nielsen

Shared publicly  - 
 
My new general audience essay on the question: "Is AlphaGo really such a big deal [for artificial intelligence]?"

"In 1997, IBM’s Deep Blue system defeated the world chess champion, Garry Kasparov. At the time, the victory was widely described as a milestone in artificial intelligence. But Deep Blue’s technology turned out to be useful for chess and not much else. Computer science did not undergo a revolution. Will AlphaGo, the Go-playing system that recently defeated one of the strongest Go players in history, be any different?"

(Short answer: a qualified yes.)

https://www.quantamagazine.org/20160329-why-alphago-is-really-such-a-big-deal/
44
25
Tau-Mu Yi's profile photoMihail Etropolski's profile photoMichael Nielsen's profile photoKaj Sotala's profile photo
6 comments
 
Good article! One nitpick:

> For instance, AlphaGo learned from 150,000 human games. That’s a lot of games! By contrast, human beings can learn a great deal from far fewer games. 

It's a lot of games, but not necessarily that much more than the amount of games that a professional human Go player would have familiarized themselves with: http://lesswrong.com/r/discussion/lw/n8b/link_alphago_mastering_the_ancient_game_of_go/d2o5

(That said, I agree with your general point; Lee Sedol adapted his strategy against AlphaGo after each match and learned from his mistakes, while AlphaGo couldn't have done anything similar - I remember the DeepMind team mentioning that individual games have basically no impact on the system's gameplay.)
Add a comment...

Michael Nielsen

Shared publicly  - 
 
New essay: "Toward an exploratory medium for mathematics"

A small excerpt, from the conclusion:

"In discussions of systems of reasoning it is sometimes assumed that the informal, intuitive systems used by humans are things to be “fixed up”, turned into so-called proper, rigorous reasoning. If the purpose of reasoning were merely verifying correctness, then that would be a reasonable point of view. But if the purpose of reasoning is exploration and discovery, then it is wrong. Exploration and discovery require a logic that is different to, and at least as valuable as, conventionally “correct” reasoning. The idea of semi-concrete reasoning is a step toward media to support such exploration and discovery, and perhaps toward new ways of thinking about mathematics.

Alan Kay has asked “what is the carrying capacity for ideas of the computer?” We may also ask the closely related question: “what is the carrying capacity for discovery of the computer?” In this essay we've made progress on that question using a simple strategy: develop a prototype medium to represent mathematics in a new way, and carefully investigate what we can learn when we use the prototype to attack a serious mathematical problem. In future, it'd be of interest to pursue a similar strategy with other problems, and with more adventurous interface ideas. And, of course, it would be of interest to build out a working system that develops the best ideas fully, not merely prototypes.

To conclude, a personal observation. I began thinking about the design of cognitive media just a few years ago. I've repeatedly found that interface design is deeper than I suspected. A powerful medium reifies the deepest ideas we have about a subject: it becomes an active carrier for those ideas. And to the extent it is successful in reifying those ideas, mastering the medium becomes the same as mastering the subject. In this view, designing exploratory media is about designing tools which can transform and extend our ability to think, create, and discover."


http://cognitivemedium.com/emm/emm.html
The world's oldest complete piece of surviving music is the 3400-year-old Hurrian cult hymn. The hymn was found inscribed on clay tablets excavated from the ancient city of Ugarit, in modern-day Syria. Sometime after the hymn was composed, the idea of using notation to record music was lost.
23
15
Boris Borcic's profile photoJean-Luc Delatre's profile photo
2 comments
 
Though much more ambitious this reminds of Barwise /Etchemendy:
"Computers, visualization, and the nature of reasoning"
https://scholar.google.com/scholar?cluster=13739102424462661059&hl=fr&as_sdt=0,5
Also of Barwise/Shimojima:
"Surrogate Reasoning."
http://citeseerx.ist.psu.edu/showciting?cid=2613641
The main import of these is that actual, real human reasoning isn't at all "logical", that's why it's so hard to get a proof out of an intuition. :-)
Add a comment...

Michael Nielsen

Shared publicly  - 
 
+Dan Piponi's one-line review of "The Martian": "It's rare that my only complaint about a movie is that it I think it over/underestimates gas pressures in various scenarios."

https://twitter.com/sigfpe/status/650486904476053504
18
Michelle Vanderbilt-Perry's profile photo
Add a comment...

Michael Nielsen

Shared publicly  - 
 
A stimulating article on The Stanford Encyclopedia of Philosophy as a model for online encyclopediae.

qz.com/480741/this-free-online-encyclopedia-has-achieved-what-wikipedia-can-only-dream-of/
17
2
JC Cabrejas's profile photoDavid Eppstein's profile photoMatthew Leifer's profile photoTimothy Gowers's profile photo
10 comments
 
Yes, someone else posted it and I read it in full. I tend to regard things like the SEP as isolated miracles -- great when they happen but unlikely to be scalable up in a big way. I find that Wikipedia is surprisingly good for mathematics -- it is nearly always helpful when I want to find out about a concept I don't know about, though I also find that it is nearly always not quite as helpful as I would ideally like. 

To step into fantasyland for a moment, what I would love to see is a very well-organized website that was something like a textbook for the whole of mathematics. The articles would not be like encyclopaedia entries but more like the very best textbooks -- explaining motivation, including well-chosen exercises, and so on -- and they would link together nicely (not in the Wikipedia way where sometimes it feels as though every other word is a link to an article that you need to read in order to understand the current article). 

Something a bit more achievable would be a guide to the most useful mathematics literature that is available on the web. So it would be a bit like what I described above, except that a typical page would consist of a broad overview combined with annotated links to the very best explanations that are available online. 
Add a comment...

Michael Nielsen

Shared publicly  - 
 
I just added a donation button to my free online book "Neural networks and deep learning": http://neuralnetworksanddeeplearning.com/  (See the sidebar for the button).  If you've benefited, I'd appreciate a donation!
36
12
John Baez's profile photoAndrei Lopatenko's profile photoArnaud TOURNIER's profile photo
3 comments
 
Thanks for your book, it had deepened my understanding of back propagation.
I thought backpropagation was about finding which neuron has the biggest error. But now after having thought and thought again, i understand that backpropagation is to find which weight has the highest impact to lower the so called error.
Am i on the way ?
Add a comment...

Michael Nielsen

Shared publicly  - 
 
Excellent!  See also +Peter Suber's post from yesterday, which provides an excellent overview: https://plus.google.com/+PeterSuber/posts/j7EWjpuRjyb
 
Senate committee approves FASTR by unanimous vote!

From SPARC (emphases mine): "The Senate Homeland Security & Governmental Affairs Committee (HSGAC) today passed S. 779, the Fair Access to Science and Technology Research (FASTR) Act, unanimously by voice vote and moved it to the full Senate for consideration. This marks the first time the Senate has acted on a government-wide policy ensuring public access to the results of publicly funded research, and is an important step towards codifying the progress made by the 2013 White House OSTP Directive...."
http://www.sparc.arl.org/news/sparc-applauds-senate-committee-action-on-public-access-legislation

For more detail on the bill, see my longer post from yesterday.
https://plus.google.com/+PeterSuber/posts/j7EWjpuRjyb

#oa #openaccess #fastr   #movefastr
1 comment on original post
2
Add a comment...

Michael Nielsen

Shared publicly  - 
 
 
Moving FASTR in the US Senate.

FASTR will go to markup tomorrow in the Senate Homeland Security and Governmental Affairs Committee (HSGAC).

Here's a recap of my recent call-to-action post on FASTR, with some new details and background.
https://plus.google.com/+PeterSuber/posts/G2uebVhVtBv

FASTR is the strongest bill ever introduced in Congress requiring open access to federally-funded research. 

We already have the 2008 NIH policy, but it only covers one agency. We already have the 2013 Obama directive requiring about two dozen federal agencies to adopt OA mandates, but the next President could rescind it.

FASTR would subsume and extend the NIH policy. FASTR would solidify the Obama directive by grounding these agency policies in legislation. Moreover, FASTR would strengthen the NIH policy and Obama directive by requiring reuse rights or open licensing. It has bipartisan support in both the House and the Senate. 

FASTR has been introduced in two sessions of Congress (February 2013 and March 2015), and its predecessor, FRPAA (Federal Research Public Access Act), was introduced in three (May 2006, April 2010, February 2012). Neither FASTR nor FRPAA has gotten to the stage of markup and a committee vote. That's why tomorrow's markup is so big.

For the reasons why FASTR is stronger than the Obama directive, see my 2013 article comparing the two.
http://dash.harvard.edu/handle/1/10528299

FASTR is stronger than the NIH policy and the Obama directive on reuse rights and open licensing. Here's why, in an excerpt from the article above:

"In three separate places, FASTR calls for agency policies to permit 'productive reuse' and 'computational analysis by state-of-the-art technologies'. I like the phrase "computational analysis by state-of-the-art technologies" better than 'text-mining'. FASTR makes agency policies evolve to permit new kinds of reuse, not just the kind represented by 'text-mining' circa 2013....The White House directive takes steps in the same direction but doesn't go as far. It encourages 'public-private collaboration to...maximize the potential for...creative reuse to enhance value to all stakeholders'.... 'Creative reuse' is essentially equivalent to FASTR's 'productive reuse'....The snag is that the directive doesn't actually require policies to maximize that potential. Instead it 'encourages' it, and it encourages it through 'public-private collaboration' rather than directly through open-licensing terms....FASTR requires agencies to study 'whether [deposited OA] research papers should include a royalty-free copyright license that is available to the public and that permits the reuse of those research papers, on the condition that attribution is given to the author or authors of the research and any others designated by the copyright owner'....Agencies that don't require CC-BY licenses must essentially explain why, and they must do so every year....That's why I've been telling colleagues that FASTR tacitly recommends CC-BY...."

While the Obama directive was always weaker than FASTR on reuse rights, the Obama White House has approved agency policies that don't even live up to its weaker standard. See my August 2014 criticism of the Department of Energy for not living up to the White House guidelines. The same could be said about every subsequent agency policy elicited by the White House directive. Of course, the real problem lies in fact that the White House abandoned the reuse provision of its own guidelines.
https://plus.google.com/+PeterSuber/posts/ZHRXEvLoq4n

As originally written, FASTR was also stronger than the NIH policy and the Obama directive on embargoes. The NIH policy permits embargoes up to 12 months, and the Obama directive makes 12 months the default. FASTR originally capped embargoes at 6 months for all covered agencies. However, just this week as the bill was approaching markup, Ron Johnson (R-WI) and Tom Carper (D-DE), the HSGAC chairman and ranking member, introduced an amendment allowing embargoes up to 12 months. Tomorrow's markup will deal with the bill as amended. 

There's no doubt that this amendment is bad news. It weakens the bill and compromises the public interest. 

On the other hand, there's still no doubt that the bill is stronger than the NIH policy and Obama directive, if only for its strong, unambiguous position on reuse rights. 

Moreover, if you're willing to measure small differences, FASTR continues to be better on embargoes. Even as amended, FASTR would encourage embargoes shorter than 12 months, allow petitions to shorten embargoes to come from researchers, agency officials, and members of the public, and let the petitions be decided by the agencies. It's a process that favors the public over publishers.

Yesterday SPARC issued a good memo on the nature of the amendment and political circumstances that led to it.
http://sparc.arl.org/blog/fastr-be-considered-senate-committee

Publishers lobbied fiercely for this amendment. For my most recent short attempt to answer their arguments, see this January 2014 post. 
https://plus.google.com/+PeterSuber/posts/gPRFVdDD8Dg

(Another result of publisher lobbying: John McCain withdrew as a co-sponsor of FASTR.)

Finally, let me point out one more benefit of FASTR over the Obama directive. The Obama directive encouraged agencies to coordinate with one another but allowed their policies to differ. The resulting policies do differ, sometimes significantly. I can say from inside a university gearing up to comply with all these policies that their differences threatened to create huge implementation headaches. FASTR will bring these policies much closer to uniformity. If there any US universities that don't care about OA (and I doubt there are), they should still push hard for FASTR to simplify compliance.

Bonus: FASTR will also nip Elsevier's new, long embargoes in the bud, at least for articles arising from federally-funded research.

If FASTR passes, agencies that wrote policies at the weak end of the spectrum allowed by the Obama directive will have to strengthen them. When the Obama White House issued its directive in February 2013, FASTR had already been introduced in both houses of Congress and I urged agencies to keep an eye on it: "What happens if agencies develop policies to satisfy the directive, and then Congress adopts FASTR?...FASTR would add the requirements on which it is stronger than the directive....If agencies were weak on reuse and open licensing, they'd have to strengthen their libre provisions....[T]he prospect of future revision to conform with FASTR, especially after careful coordination and consultation, is a reason for agencies to set policy in light of FASTR in the first place."
http://dash.harvard.edu/handle/1/10528299

* For more details on FASTR itself, see my reference page.
http://bit.ly/hoap-fastr

* Also see my reference page on FRPAA, the predecessor of FASTR.
http://bit.ly/hoap-frpaa

* For steps you can take to support FASTR, see the action pages from the Electronic Frontier Foundation (EFF) and Scholarly Publishing and Academic Resources Coalition (SPARC).
https://action.eff.org/o/9042/p/dia/action/public/?action_KEY=9061
http://www.sparc.arl.org/advocacy/national/fastr

Please urge your Senators to support FASTR, and spread the word.

#oa #openaccess #fastr #movefastr  
2 comments on original post
4
1
Brian Ahier's profile photo
 
:-)
Add a comment...

Michael Nielsen

Shared publicly  - 
 
This is amazing, well worth watching all the way to the end. I've watched it maybe a half dozen times.
 
The Fourier Transform explained with graphical algebra (and MathBox).

http://acko.net/files/gltalks/toolsforthought/
(laptop / desktop recommended)
13 comments on original post
35
16
Add a comment...

Michael Nielsen

Shared publicly  - 
 
More Orwellian than Orwell.
 
Oh, wow. "Authoritarianism gamified", indeed:

> China is launching a comprehensive “credit score” system [...] If that and the little other reporting I’ve seen is accurate, the basics are this:

> * Everybody is measured by a score between 350 and 950, which is linked to their national identity card. While currently supposedly voluntary, the government has announced that it will be mandatory by 2020.
> * The system is run by two companies, Alibaba and Tencent, which run all the social networks in China and therefore have access to a vast amount of data about people’s social ties and activities and what they say.
> * In addition to measuring your ability to pay, as in the United States, the scores serve as a measure of political compliance. Among the things that will hurt a citizen’s score are posting political opinions without prior permission, or posting information that the regime does not like, such as about the Tienanmen Square massacre that the government carried out to hold on to power, or the Shanghai stock market collapse.
> * It will hurt your score not only if you do these things, but if any of your friends do them. Imagine the social pressure against disobedience or dissent that this will create.
> * Anybody can check anyone else’s score online. Among other things, this lets people find out which of their friends may be hurting their scores.
> * Also used to calculate scores is information about hobbies, lifestyle, and shopping. Buying certain goods will improve your score, while others (such as video games) will lower it.
> * Those with higher scores are rewarded with concrete benefits. Those who reach 700, for example, get easy access to a Singapore travel permit, while those who hit 750 get an even more valued visa.
> * Sadly, many Chinese appear to be embracing the score as a measure of social worth, with almost 100,000 people bragging about their scores on the Chinese equivalent of Twitter.
4 comments on original post
16
14
Juan Pablo Carbajal's profile photo
 
I wonder how this sharer would "label" the bombing of a hospital... "Assassination virtualised"?
Add a comment...

Michael Nielsen

Shared publicly  - 
 
A few rough working notes for a discussion group I convened to discussion Engelbart's famous 1962 paper on "Augmenting Human Intellect".
Engelbart: "Augmenting Human Intellect". “By "augmenting human intellect" we mean increasing the capability of a man to approach a complex problem situation, to gain comprehension to suit his particular needs, and to derive solutions to problems… whether the problem situation exists for twenty ...
7
3
Shriram Krishnamurthi's profile photo
 
Sharing this with my group!
Add a comment...

Michael Nielsen

Shared publicly  - 
 
Norms, law enforcement, and mathematics
 
The kid and I talk now and then about laws and social norms and law enforcement and things like that. I didn't expect it to show up in such an unexpected setting.

The kid and I are studying (algebraic) variables. She was making some headway with them, then got stuck on programs, so we went back to paper and algebraic formulas. Progress restored. I give her `x + 2 * y`. She thinks she's got it. Then she stops.

- Daddy, I have a question.
- Sure, what is it?
- I'm not sure what the answer is.
- Why not?
- I'm confused about what this is.
- Can you tell me why you're confused?
- Well, I can't tell if it's x plus 2-times-y or x-plus-2 times y.
- Good!

So we work out what the answers would be, and that they'd be different, so it does matter. I tell her there are three options. I show her (infix) parentheses. I tell her there are different notations we could use and draw her a parse tree. Or, I say, we could just decide that we do the multiplication before the addition.

- Is that a rule?
- Yes, that's a rule.
- I mean, is that a real rule?
- What do you mean?
- I mean, is that just a rule you're making up or do other people also follow that rule?
- No no, other people use it too.
- But we don't have to use it, right?
- No, really, we should. I mean, other people would expect you to use it.

Pause. Eyes get bigger. A slight look of concern crosses her face.

- The police?

#parenting  
23 comments on original post
28
3
Add a comment...

Michael Nielsen

Shared publicly  - 
 
Frank Wilczek just posted a draft book proposal for a "Princeton Companion to Physics" to Twitter.  I hope it eventuates!

Click through to see the proposal.
“Sneak preview: draft proposal / outline for "Princeton Companion to Physics" http://t.co/iB0CNSLpv5”
9
5
Fernando Pereira's profile photoTimothy Gowers's profile photo
3 comments
 
Just had a look at the proposed contents for the Physics volume -- I can certainly imagine reading that from cover to cover if it's done well. 
Add a comment...
Basic Information
Gender
Male
Places
Map of the places this user has livedMap of the places this user has livedMap of the places this user has lived
Currently
New York City
Previously
Toronto, Canada - Brisbane, Australia - Los Angeles, California - Santa Barbara, California - Los Alamos, New Mexico - Albuquerque, New Mexico - Croydon, Australia - Waterloo, Canada