Lisa_seminaires Septembre 2015

lisa_seminaires@iro.umontreal.ca

6 participants
6 discussions

Fwd: Colloque DIRO, jeudi 10 sept. Conférencier: Yoshua Bengio
by Yoshua Bengio 10 Sep '15

10 Sep '15

I will give a talk in French, tomorrow 15h30, for an audience of computer scientists. May be especially good for new students. ---------- Message transféré ---------- De : Neil Stewart <stewart(a)iro.umontreal.ca> Date : 4 septembre 2015 13:08 Objet : Colloque DIRO, jeudi 10 sept. Conférencier: Yoshua Bengio À : seminaires(a)iro.umontreal.ca Intelligence artificielle et apprentissage profond par *Yoshua Bengio* DIRO, Université de Montréal *Jeudi 10 septembre, 15:30-16:30*, Salle 3195, Pavillon André-Aidenstadt Université de Montréal, 2920 Chemin de la Tour Café avant 15:00-15:30 *Résumé*: Une petite révolution est en cours dans le monde de l'intelligence artificielle, autant dans les universités que dans les entreprises qui ont commencé à en profiter. Intelligence artificielle ? C'était devenu jusqu'à récemment un terme galvaudé, presque tabou, sans doute à cause des promesses excessives des années 80 et 90. Mais petit à petit, depuis cette époque, a grandi une approche pour construire des machines intelligentes, basée sur des algorithmes d'apprentissage : au lieu de dicter explicitement son comportement à l'ordinateur à l'aide de règles, de faits et de programmes, on le laisse apprendre à partir d'exemples. Et l'une des approches de l'apprentissage automatique qui a connu une percée fulgurante ces dernières années est justement inspirée (de façon grossière) de connaissances sur le cerveau : les réseaux de neurones, et, ce qui est plus nouveau, l'apprentissage profond, qui permet d'apprendre des tâches plus complexes en faisant en sorte que l'ordinateur découvre des représentations de haut niveau de l'information traitée. Ainsi, ces techniques ont battu des records et sont devenues des standards pour ce qui est de la vision par ordinateur, la reconnaissance de la parole, et de plus en plus, la compréhension du langage naturel et la robotique. Les grandes compagnies des technologies de l'information se sont investies massivement dans ces techniques depuis environ deux ou trois ans, d'où une explosion de "start-ups" qui se vendent à des prix mirobolants, et une fuite des cerveaux du monde universitaire vers le monde industriel. Après l'allégresse, arrivent maintenant les oiseaux de mauvais augure, hantés par la crainte d'une intelligence artificielle qui éliminerait la race humaine, telle qu'on le voit dans les films de science-fiction. Nous en sommes encore loin... mais les retombées technologiques, elles, sont bien réelles et à portée de main. La présentation sera en français. *Abstract:* A small revolution is going on in the world of artificial intelligence (AI), in universities as well as in corporations, many of which have started to take advantage of this progress in their products. AI? It had become an almost taboo term in some communities, probably because of excessive promises of the 80's and 90's. However, a little at a time, since these years has grown an approach to AI based on machine learning: instead of explicitly stating how a computer should behave through rules, facts and programs, we let the computer learn from examples and data. One of the machine learning approaches that is particularly attracting attention and investment is one inspired from knowledge of brains: neural networks, and more recently, the deep learning approaches that can be used to train more sophisticated systems where the computer discovers high-level representations. These techniques have beaten many records, reaching human-level performance in several cases (especially in computer vision), with impressive advances in speech recognition, computer vision, natural language processing and robotics. Information technology companies have massively invested in the area and many startups have emerged (or been bought by the IT giants at surprisingly high prices), creating a brain drain from academia to industry. On top of this enthusiasm, though, some are raising the spectre of science-fiction inspired AIs which could wipe out humanity. We are still far from these nightmarish visions, while technological fallouts are quite real and close by. The talk will be given in French.

1 2

Re: [Lisa_seminaires] [Lisa_labo] Fwd: Connectionists: The ReScience journal
by Yoshua Bengio 08 Sep '15

08 Sep '15

Unless (a) the research is about things like the computational efficiency brought by some distributed training scheme ;-) or (b) the result was obtained on a very large dataset and model, and would take too much time to replicate in SlowpokeML. -- Yoshua 2015-09-07 23:02 GMT-04:00 Guillaume Alain < guillaume.alain.umontreal(a)gmail.com>: > I've read the thread and some of the links provided, and my two cents to > add to this story would be the following. > > Replicability doesn't have to involve an algorithm that runs at the same > speed. An experiment conducted with 4 GPUs, running for one hour, might be > replicated with a CPU running for a week. > > That doesn't provide an easy solution to the replicability problem, but > it's one aspect to keep in mind. Replicability doesn't mean that you *need* > to duplicate the whole pipeline in an efficient way. > > > You can read a paper about linked lists from 30 years ago, use the code > (in C), and run it on data that's 100+ times bigger, and it'll still work. > You can read the classic SICP course in Scheme, work through all the > exercises, and ignore the fact that it ran insanely more slowly on the > machines that MIT had a the time that they created the course. > > If we had machine learning toolkit called "snail" or "slowpoke" or > "escargot" or whatever, and it was relatively easy to replicate experiments > in that language, but it ran slow as hell, then it might satisfy the > replicability criterion. No vectorization (unless you want to) for > mini-batches. No cluster computation. No fancy GPU convolutions. Your code > just has to be correct. > > I'm taking as hypothesis here that re-implementing your favorite algorithm > in SlowpokeML would be easy, but whether such a language exists or not is > something that is a whole other issue. > > On Sun, Sep 6, 2015 at 1:45 AM, Kyle Kastner <kastnerkyle(a)gmail.com> > wrote: > >> Important to distinguish replication and reproduction. Replication is (in >> my opinion) when experiments in similar or same environments work. >> Reproduction is more independent - implementing the same ideas in a very >> different setting. This uncovers much more subtle (but important) issues >> than basic replication. And we need both in ML - but replication doubly so. >> >> Also dwf's point is a strong reason to minimize dependencies in research >> and library development. Its a personal opinion but one that has helped >> greatly in the past. >> On Sep 5, 2015 7:18 PM, "Jörg Bornschein" <bornj(a)iro.umontreal.ca> wrote: >> >>> >>> I agree and I was basically thinking the same thing (+that Docker or >>> other container virtualization techniques could ease the pain [*]). >>> >>> But for reproducability in the scientific sense it is already a huge win >>> when an independent implementation ran only once or twice on the machines >>> of the reproducing authors (and the editors). From that point of view it >>> would not be devastating if the code was not in a runnable state a few >>> months later. >>> >>> >>> j >>> >>> [*] I actually wonder why they did't make the dependencies more explicit >>> in their submission format. >>> >>> >>> >>> >>> On Sat, Sep 5, 2015 at 7:00 PM David Warde-Farley < >>> d.warde.farley(a)gmail.com> wrote: >>> >>>> Very encouraging to see this happening and that other people are >>>> concerned about it. >>>> >>>> I would add that reproducibility in machine learning looks simple >>>> compared to other scientific domains, but looks are deceiving. Every >>>> "simple Python script" is built upon a broad and deep tower of library >>>> dependencies, leading to an exponential number of ways that your >>>> computing environment can conspire against you (nevermind hardware >>>> differences...). >>>> >>>> On Sat, Sep 5, 2015 at 6:35 PM, Yoshua Bengio >>>> <yoshua.umontreal(a)gmail.com> wrote: >>>> > >>>> > Very interesting! >>>> > Reproducibility is VERY weak in the machine learning community, and >>>> needs to >>>> > be improved. >>>> > >>>> > ---------- Forwarded message ---------- >>>> > From: Nicolas P. Rougier <Nicolas.Rougier(a)inria.fr> >>>> > Date: 2015-09-03 8:57 GMT-04:00 >>>> > Subject: Connectionists: The ReScience journal >>>> > To: Connectionists group <connectionists(a)cs.cmu.edu> >>>> > >>>> > >>>> > >>>> > It's our great pleasure to announce the creation of "ReScience" which >>>> is a >>>> > peer-reviewed journal that targets computational research and >>>> encourages the >>>> > explicit replication of already published research, promoting new and >>>> > open-source implementations in order to ensure that the original >>>> research is >>>> > reproducible. >>>> > >>>> > To achieve such a goal, the whole editing chain is radically >>>> different from >>>> > any other traditional scientific journal. ReScience lives on GitHub >>>> where >>>> > each new implementation is made available together with comments, >>>> > explanations and tests. Each submission takes the form of a pull >>>> request >>>> > that is publicly reviewed and tested in order to guarantee that any >>>> > researcher can re-use it. >>>> > >>>> > Students are strongly encourage to submit to ReScience. Even if the >>>> > publishing model is a bit different from other academic journals, >>>> this will >>>> > give them a first experience at peer-reviewed publishing where they >>>> have to >>>> > use a rigorous and scientific approach. >>>> > >>>> > • More on the journal website: >>>> > https://github.com/ReScience/ReScience/wiki >>>> > • Current issue: >>>> > https://github.com/ReScience/ReScience/wiki/Current-Issue >>>> > • FAQ: >>>> > >>>> https://github.com/ReScience/ReScience/wiki/Frequently-Asked-Questions >>>> > • Follow us on twitter (@ReScienceEds): >>>> > https://twitter.com/rescienceeds >>>> > >>>> > And if you're familiar with Git and GitHub, you can also become a >>>> reviewer: >>>> > just contact us. >>>> > >>>> > >>>> > Konrad Hinsen & Nicolas Rougier >>>> > >>>> > >>>> > _______________________________________________ >>>> > Lisa_labo mailing list >>>> > Lisa_labo(a)iro.umontreal.ca >>>> > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >>>> > >>>> _______________________________________________ >>>> Lisa_labo mailing list >>>> Lisa_labo(a)iro.umontreal.ca >>>> https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >>>> >>> >>> _______________________________________________ >>> Lisa_seminaires mailing list >>> Lisa_seminaires(a)iro.umontreal.ca >>> https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_seminaires >>> >>> >> _______________________________________________ >> Lisa_labo mailing list >> Lisa_labo(a)iro.umontreal.ca >> https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >> >> > > _______________________________________________ > Lisa_labo mailing list > Lisa_labo(a)iro.umontreal.ca > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo > >

1 0

Re: [Lisa_seminaires] [Lisa_labo] Fwd: Connectionists: The ReScience journal
by Jörg Bornschein 06 Sep '15

06 Sep '15

I agree and I was basically thinking the same thing (+that Docker or other container virtualization techniques could ease the pain [*]). But for reproducability in the scientific sense it is already a huge win when an independent implementation ran only once or twice on the machines of the reproducing authors (and the editors). From that point of view it would not be devastating if the code was not in a runnable state a few months later. j [*] I actually wonder why they did't make the dependencies more explicit in their submission format. On Sat, Sep 5, 2015 at 7:00 PM David Warde-Farley <d.warde.farley(a)gmail.com> wrote: > Very encouraging to see this happening and that other people are > concerned about it. > > I would add that reproducibility in machine learning looks simple > compared to other scientific domains, but looks are deceiving. Every > "simple Python script" is built upon a broad and deep tower of library > dependencies, leading to an exponential number of ways that your > computing environment can conspire against you (nevermind hardware > differences...). > > On Sat, Sep 5, 2015 at 6:35 PM, Yoshua Bengio > <yoshua.umontreal(a)gmail.com> wrote: > > > > Very interesting! > > Reproducibility is VERY weak in the machine learning community, and > needs to > > be improved. > > > > ---------- Forwarded message ---------- > > From: Nicolas P. Rougier <Nicolas.Rougier(a)inria.fr> > > Date: 2015-09-03 8:57 GMT-04:00 > > Subject: Connectionists: The ReScience journal > > To: Connectionists group <connectionists(a)cs.cmu.edu> > > > > > > > > It's our great pleasure to announce the creation of "ReScience" which is > a > > peer-reviewed journal that targets computational research and encourages > the > > explicit replication of already published research, promoting new and > > open-source implementations in order to ensure that the original > research is > > reproducible. > > > > To achieve such a goal, the whole editing chain is radically different > from > > any other traditional scientific journal. ReScience lives on GitHub where > > each new implementation is made available together with comments, > > explanations and tests. Each submission takes the form of a pull request > > that is publicly reviewed and tested in order to guarantee that any > > researcher can re-use it. > > > > Students are strongly encourage to submit to ReScience. Even if the > > publishing model is a bit different from other academic journals, this > will > > give them a first experience at peer-reviewed publishing where they have > to > > use a rigorous and scientific approach. > > > > • More on the journal website: > > https://github.com/ReScience/ReScience/wiki > > • Current issue: > > https://github.com/ReScience/ReScience/wiki/Current-Issue > > • FAQ: > > https://github.com/ReScience/ReScience/wiki/Frequently-Asked-Questions > > • Follow us on twitter (@ReScienceEds): > > https://twitter.com/rescienceeds > > > > And if you're familiar with Git and GitHub, you can also become a > reviewer: > > just contact us. > > > > > > Konrad Hinsen & Nicolas Rougier > > > > > > _______________________________________________ > > Lisa_labo mailing list > > Lisa_labo(a)iro.umontreal.ca > > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo > > > _______________________________________________ > Lisa_labo mailing list > Lisa_labo(a)iro.umontreal.ca > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >

2 1

Re: [Lisa_seminaires] [Lisa_labo] Fwd: Connectionists: The ReScience journal
by Hovig Bayandorian 05 Sep '15

05 Sep '15

John Ioannidis - Reproducibility of scientific results | | | | | | | | | | | John Ioannidis - Reproducibility of scientific results | | | | View on www.youtube.com | Preview by Yahoo | | | | | also see https://en.wikipedia.org/wiki/Ben_GoldacreBad Pharma - Wikipedia, the free encyclopedia | | | | | | | | | | | Bad Pharma - Wikipedia, the free encyclopediaBad Pharma: How Drug Companies Mislead Doctors and Harm Patients is a book by British physician and academic Ben Goldacre about the pharmaceutical industry, i... | | | | View on en.wikipedia.org | Preview by Yahoo | | | | | On Saturday, September 5, 2015 9:13 PM, David Warde-Farley <d.warde.farley(a)gmail.com> wrote: I would argue that it's a social problem that is at least partly solved by enforcing community norms on archival publication venues, i.e. Reject anything that doesn't have compelling results on known public benchmarks (if there's other less reproducible stuff too, fine).If it's just people posting on arXiv, there's little to be done about it. We are better off having those unpublished reports in public view than not, though they are clearly a less compelling and less trustworthy form of evidence.On Sep 5, 2015 8:05 PM, "Yoshua Bengio" <yoshua.umontreal(a)gmail.com> wrote: Yes. It's sad. 2015-09-05 19:44 GMT-04:00 Alex Lamb <alex6200(a)gmail.com>: A lot of the papers that I've read recently use a private training set and a public testing set (for example, the Google and Facebook face recognition papers). In areas where the largest datasets are privately held, research is probably going to continue to be impossible to fully reproduce. On Sat, Sep 5, 2015 at 7:03 PM, David Krueger <david.scott.krueger(a)gmail.com> wrote: To what extent can this be remedied by using aws or similar?On Sep 5, 2015 7:00 PM, "David Warde-Farley" <d.warde.farley(a)gmail.com> wrote: Very encouraging to see this happening and that other people are concerned about it. I would add that reproducibility in machine learning looks simple compared to other scientific domains, but looks are deceiving. Every "simple Python script" is built upon a broad and deep tower of library dependencies, leading to an exponential number of ways that your computing environment can conspire against you (nevermind hardware differences...). On Sat, Sep 5, 2015 at 6:35 PM, Yoshua Bengio <yoshua.umontreal(a)gmail.com> wrote: > > Very interesting! > Reproducibility is VERY weak in the machine learning community, and needs to > be improved. > > ---------- Forwarded message ---------- > From: Nicolas P. Rougier <Nicolas.Rougier(a)inria.fr> > Date: 2015-09-03 8:57 GMT-04:00 > Subject: Connectionists: The ReScience journal > To: Connectionists group <connectionists(a)cs.cmu.edu> > > > > It's our great pleasure to announce the creation of "ReScience" which is a > peer-reviewed journal that targets computational research and encourages the > explicit replication of already published research, promoting new and > open-source implementations in order to ensure that the original research is > reproducible. > > To achieve such a goal, the whole editing chain is radically different from > any other traditional scientific journal. ReScience lives on GitHub where > each new implementation is made available together with comments, > explanations and tests. Each submission takes the form of a pull request > that is publicly reviewed and tested in order to guarantee that any > researcher can re-use it. > > Students are strongly encourage to submit to ReScience. Even if the > publishing model is a bit different from other academic journals, this will > give them a first experience at peer-reviewed publishing where they have to > use a rigorous and scientific approach. > > • More on the journal website: > https://github.com/ReScience/ReScience/wiki > • Current issue: > https://github.com/ReScience/ReScience/wiki/Current-Issue > • FAQ: > https://github.com/ReScience/ReScience/wiki/Frequently-Asked-Questions > • Follow us on twitter (@ReScienceEds): > https://twitter.com/rescienceeds > > And if you're familiar with Git and GitHub, you can also become a reviewer: > just contact us. > > > Konrad Hinsen & Nicolas Rougier > > > _______________________________________________ > Lisa_labo mailing list > Lisa_labo(a)iro.umontreal.ca > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo > _______________________________________________ Lisa_labo mailing list Lisa_labo(a)iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo _______________________________________________ Lisa_labo mailing list Lisa_labo(a)iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo _______________________________________________ Lisa_labo mailing list Lisa_labo(a)iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo _______________________________________________ Lisa_labo mailing list Lisa_labo(a)iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo

1 0

Re: [Lisa_seminaires] [Lisa_labo] Fwd: Connectionists: The ReScience journal
by David Krueger 05 Sep '15

05 Sep '15

To what extent can this be remedied by using aws or similar? On Sep 5, 2015 7:00 PM, "David Warde-Farley" <d.warde.farley(a)gmail.com> wrote: > Very encouraging to see this happening and that other people are > concerned about it. > > I would add that reproducibility in machine learning looks simple > compared to other scientific domains, but looks are deceiving. Every > "simple Python script" is built upon a broad and deep tower of library > dependencies, leading to an exponential number of ways that your > computing environment can conspire against you (nevermind hardware > differences...). > > On Sat, Sep 5, 2015 at 6:35 PM, Yoshua Bengio > <yoshua.umontreal(a)gmail.com> wrote: > > > > Very interesting! > > Reproducibility is VERY weak in the machine learning community, and > needs to > > be improved. > > > > ---------- Forwarded message ---------- > > From: Nicolas P. Rougier <Nicolas.Rougier(a)inria.fr> > > Date: 2015-09-03 8:57 GMT-04:00 > > Subject: Connectionists: The ReScience journal > > To: Connectionists group <connectionists(a)cs.cmu.edu> > > > > > > > > It's our great pleasure to announce the creation of "ReScience" which is > a > > peer-reviewed journal that targets computational research and encourages > the > > explicit replication of already published research, promoting new and > > open-source implementations in order to ensure that the original > research is > > reproducible. > > > > To achieve such a goal, the whole editing chain is radically different > from > > any other traditional scientific journal. ReScience lives on GitHub where > > each new implementation is made available together with comments, > > explanations and tests. Each submission takes the form of a pull request > > that is publicly reviewed and tested in order to guarantee that any > > researcher can re-use it. > > > > Students are strongly encourage to submit to ReScience. Even if the > > publishing model is a bit different from other academic journals, this > will > > give them a first experience at peer-reviewed publishing where they have > to > > use a rigorous and scientific approach. > > > > • More on the journal website: > > https://github.com/ReScience/ReScience/wiki > > • Current issue: > > https://github.com/ReScience/ReScience/wiki/Current-Issue > > • FAQ: > > https://github.com/ReScience/ReScience/wiki/Frequently-Asked-Questions > > • Follow us on twitter (@ReScienceEds): > > https://twitter.com/rescienceeds > > > > And if you're familiar with Git and GitHub, you can also become a > reviewer: > > just contact us. > > > > > > Konrad Hinsen & Nicolas Rougier > > > > > > _______________________________________________ > > Lisa_labo mailing list > > Lisa_labo(a)iro.umontreal.ca > > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo > > > _______________________________________________ > Lisa_labo mailing list > Lisa_labo(a)iro.umontreal.ca > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_labo >

3 2

Fwd: Connectionists: The ReScience journal
by Yoshua Bengio 05 Sep '15

05 Sep '15

Very interesting! Reproducibility is VERY weak in the machine learning community, and needs to be improved. ---------- Forwarded message ---------- From: Nicolas P. Rougier <Nicolas.Rougier(a)inria.fr> Date: 2015-09-03 8:57 GMT-04:00 Subject: Connectionists: The ReScience journal To: Connectionists group <connectionists(a)cs.cmu.edu> It's our great pleasure to announce the creation of "ReScience" which is a peer-reviewed journal that targets computational research and encourages the explicit replication of already published research, promoting new and open-source implementations in order to ensure that the original research is reproducible. To achieve such a goal, the whole editing chain is radically different from any other traditional scientific journal. ReScience lives on GitHub where each new implementation is made available together with comments, explanations and tests. Each submission takes the form of a pull request that is publicly reviewed and tested in order to guarantee that any researcher can re-use it. Students are strongly encourage to submit to ReScience. Even if the publishing model is a bit different from other academic journals, this will give them a first experience at peer-reviewed publishing where they have to use a rigorous and scientific approach. • More on the journal website: https://github.com/ReScience/ReScience/wiki • Current issue: https://github.com/ReScience/ReScience/wiki/Current-Issue • FAQ: https://github.com/ReScience/ReScience/wiki/Frequently-Asked-Questions • Follow us on twitter (@ReScienceEds): https://twitter.com/rescienceeds And if you're familiar with Git and GitHub, you can also become a reviewer: just contact us. Konrad Hinsen & Nicolas Rougier

1 0

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

Lisa_seminaires Septembre 2015