Commit 22bde7f3 authored by Nathan Rebiscoul's avatar Nathan Rebiscoul
Browse files

Add notes

parent 37286307
......@@ -273,3 +273,4 @@ going to quick into conclusion. Maybe because smoker died more quickly
there is more non smoker old people thus statistics are changed. So we
cannot conclude maybe we need another kind of study to show the
dangerousness of smoking.
......@@ -797,6 +797,128 @@ look at slide 3-5.
- Screening design
* Course 10
** Bandit problem
- Find the most valuable bandit
- problem of choosing randomly things
- same as target advertisement (which add it should show)
- Tests in stastical science Surrogates Gaussian process modeling
design and optimisation for the applied science
* Course 11
** DoE exercice
- In original experiment there was a bias depending on the
time/seasons where you do the experiment. It was to show how
important is to spread the measurement and work of data.
** Ethics, Scientific Integrity And Deontology
- OFIS (Office Français de l'intégrité scientifique), HCERES, CNRS
COMETS, Inria COERLE, CERNA
Three Pillars :
- *Ethics* : Question raised by science progress in society, Moral
values and Consequences. Regular discussion connection with
society. Retrospective (law, approval) vs prospective (what will
happened) view.
- *Scientific Integrity* : Professional rules conduct and research
practices for trustable research. Universal soft law professional
code.
- *Deontology* : Independance of research, equity, private activities of
civil servants. Conflict of interest connection with industry,
harassment, intellectual property... Law deontology referent. (In
company it's the deontology of the company)
Opening of science is important (National plan 2018-2021)
Doctoral training in Ethics, SI and Deontology
** Deontology
- "Obligation de dignité" : must be a decent guy, do not insult and
humiliate people.
- "Obligation de probité et d'intégrité" : using your position to
obtain personal benefit, steal credit of another one...
-"Obligation d'impartiabilité" : treat anyone in a fair and objective
way
- "Obligation de neutralité" : Must not depend on
political/reglious/philosophical opinion. No conflict with freedom
of speech...
- "Obligation the secret/discretion professionel" : confidential data,
economic and industrial espionage.
- "Obligation d'exécution des fonctions et d'obéissance hiérarchique"
*** Conflict of interest
- (tobacco, sugar, oil, chemistry) may fund your research
so that you give conclusion they like or study alternate
explanation. relation between funding source and conclusion known as
the *Funding Bias*.
- Study of long term toxicity of roundup-tolerant GMO on rats by
Séralini et al was retired but theey respond monsanto made study so its a
mess (hard and sensitive topics).
*** Bullying and Sexual Harassment
Surprinsingly despite a high level of education.
Harrasment of E.Bik by D.Raoult (doing doxing things) accusation of
being a charlatan and a fake expert and paid off one.
** Ethics and regulation
- Tuskegee Syphilis study
Question what happened if we do not treat people.
Try it on poor and black people.
600 (400 malade, 200 saint)
They never informed of their disease giave them a placebo disguised as
bad blood treatment.
force them to keep accept treatement with meal and money, nurse
- Unethical clinical studies
- Hide information on diseases and risks
- Fraudulent medical care
- Influence of money
- force consent
- Made of code of conducts with Nuremberg (but not mandatory)
- Helsinki I declaration import the need of free and inform consent
- Helsinki II Mandatory free and inform consent and several other
things.
- Gay anonymous relation research
- No free and informed consent, tracked names, adress...
** Scientific Integrity
- Scientific Misconduct
- Several definition depending on the country
- US NSF definition
- 1 fabrication
- 2 falsification
- 3 plagiarism
- Data Fabrication Examples
- Falsified blood result
- fill himself form
- Building hypothesis as you analyse data
- Data falsification with replication of some digit
- Photomanipulation (zoom, photoshop...) it used to be a
common/requested practice.
- HARKING : Hypothesis after getting result
- P-hacking spurious correlations/multiple comparisons. Systematic
bias toward shiny results.
- Overhacking (hack data until <0.05)
- Selection bias (only report the smallest one)
- Selective debugging (change tests, look for bugs until < 0.05)
- publication bias
- Cherry picking and not reporting failure
- Plagiarism
- Plagiarism is stealing
- Must cite things
- Self plagiarism is when you cite yourself without say it
- Maximise number of publication by publishing only small paper
* Tools
** Binder
Launch docker imag
Launch docker image
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment