Information: this page is not translated in english

Petits déjeuners

Une fois par mois : du café, des croissants et un sujet de discussion.

20 october 2016, 10:00 AM   (Room: BAT5 02/022)

Benjamin Linard, "Science and social networking"

Résumé

Social networks are more and more used by scientists to spread
their research (Twitter, LinkedIn, Facebook). But social networks specifically
targeted to scientists are now appearing (Google scholar, Researchgate).
Are these useful and can be considered as a research tool ?
Or are they just a consequence of the "mediatic science" in which we live
more and more, forgetting sometimes the purposes of science ?

24 november 2016, 10:00 AM  (Room: BAT5 01/124)

Fabio Pardi, "SIMONS human diversity project"

Résumé

The SIMONS project completed the 1000 human project by sequencing many
ancestral Asian and Oceania populations. Their results highlights the complexity
of ancestral Homo sapiens cross-breeding which occured at different times
and different continents (sapiens, neanderthalis and denisova).
A nice example of coalescence models application will be discussed.

8 december 2016, 10:00 AM  (Room: BAT5 02/022)

Laurent bréhélin, "Cell biology by the numbers"

Résumé

"One of the central missions of our book is to serve as an entry point
that invites the reader to explore some of the key numbers of cell
biology.  We imagine readers of all kinds with different approaches:
seasoned researchers who simply want to find the best values for some
number of interest or beginning biology students who wish to supplement
their introductory course materials.  In the pages that follow, we
provide several dozen vignettes, each of which focuses on quantities
that help us think about sizes, concentrations, energies, rates,
information content and other key quantities that describe the living
world." http://book.bionumbers.org/

9 february 2016, 10:00 AM  (Room: BAT5 02/022)

Clément Agret, "MARIANA: the cutest deep learning framework"

Résumé

VO :As neural nets increase in complexity they also become harder to write and harder to teach. Our hypothesis is that these difficulties stem from the absence of a language that elegantly describe
neural networks.
Mariana (named after the deepest place on earth, the Mariana trench) is an attempt to create such a language within python. That being said, you can also call it an Extendable Python Machine Learning
Framework build on top of Theano. VF :Plus les réseaux neuronaux augmentent en complexité, plus ils deviennent difficiles à écrire et à enseigner. L’hypothèse des développeurs de Mariana est que ces
difficultés proviennent de l'absence d'un langage qui décrit avec simplicité les réseaux de neurones.
Mariana (du nom de l'endroit le plus profond sur terre) est une tentative de créer un tel langage pour python. Cela étant dit, vous pouvez également l'appeler "Python Extensible Machine Learning
Framework" construite sur Theano.

16 march 2017, 10:00 AM  (Room: BAT5 02/022)

Rodrigo-antonio Canovas-barroso, "Debate about Compressed Structures"

Résumé

A compressed structure is a data structure modified to take up little space, besides providing direct access
and functionalities to answer certain queries on the stored data without need to decompress them.
Nowadays it has become of vitally important to operate on compressed data.
This is mainly because the information handled has grown exponentially and,
as an aggravating circumstance, secondary storage technology has not evolved enough in terms of speed.
Currently, main memory access is estimated at 10 ns, whereas disk access is estimated at 10^7 ns.
The speed difference is so vast that it is preferable to pay a moderate cost of access to a slower compressed structure,
that it fits in main memory, than to make constant access to disk.

In this context compressed structures make more sense, because by using a smaller space,
more information can be handled in main memory by paying a lower operating cost than
managing the same information efficiently on disk. However, still remains the question
for how long these structures will be still needed.

20 april 2017, 10:00 AM  (Room: BAT5 01/124)

Krister Swenson, "Gluing computation experiments together with Snakemake"

Résumé

Do you have a program that operates on data produced by another program?
Snakemake is an efficient way to deal with interactions between a group of programs that depend on each other.
Some benefits of Snakemake are:
  1. python-based syntax (little syntax specific to snakemake)
  2. automatic parallelization of calls
  3. easily rerun only the necessary parts of an experiment after changes to code or input

I will present the basics of Snakemake and go through a simple example.

Last update on 15/03/2017