Word
Be a part of us to work on reinventing data-science practices and instruments to
produce sturdy evaluation with much less knowledge curation.
It’s well-known that knowledge cleansing and preparation are a heavy burden to
the information scientist.
Soiled knowledge analysis
Within the soiled knowledge mission, we
have been conducting machine-learning analysis to see how higher
statistical fashions might readily ingest non-curated knowledge, and cut back the
want of knowledge preparation for knowledge science. We now have a rising
understanding of the issues, theoretical and sensible, which lie
throughout statistical and database matters.
Machine studying results in completely different tradeoffs than conventional
inferential statistics (as a result of it may well depend on extra highly effective mannequin). For
occasion, we now have an excellent understanding of the case of lacking values:
in Le Morvan et al, we confirmed that
with conventional strategies, ignorable missingness and “good”
imputation are necessary, nevertheless it seems for prediction, versatile
predictors are what issues and so they can work on any missingness
mechanism.
Equally, we now have made good progress on tolerating normalization errors
and typos. We discover that reasonably to try to deduplicate the entries or
repair the typos, it’s best to signify similarities and ambiguities to
a versatile studying algorithm. The only and most dependable strategies are
carried out within the dirty-cat library, to
facilitate the lifetime of data-scientists
Reinventing knowledge science
With this understanding (and much more thrilling on-going analysis), we
wish to revisit knowledge science. Machine-learning can present versatile
fashions for a lot of usages of knowledge science. Our aim is to make use of it to assist
assembling and analyzing datasets whereas minimizing human efforts. For
this, we want instruments that may reply typical data-science questions utilizing
machine studying and ranging from the uncooked knowledge, typically unfold in a number of
information or a number of tables of a databases. Constructing these instruments requires
data-science analysis, a brand new imaginative and prescient of data-science APIs, and cautious
software program crafting.
Be a part of us on this journey
We now have an superior staff,
with an ideal combine of individuals of various seniority, completely different experience
(statistics, machine studying, databases, software program engineering), sharing
workplaces with the scikit-learn at Inria. However we now have too many
thrilling concepts, so we’re rising this staff.
An information-science engineer: new software program with new concepts
We’re in search of somebody with a background in knowledge science or numerical
Python programming to hitch us, to assist with designing a brand new data-science
library, evolving from dirty-cat, and
to assist with data-science experimentation for the analysis.
We like individuals who care about knowledge, designing good instruments, and have imaginative and prescient
about knowledge science. We’re glad to think about completely different stage of
expertise. Apply on the job supply.
A post-doc researcher: science becoming a member of knowledge engineering to deep studying
We’ll quickly be saying a post-doc place to hitch the staff for
analysis on this scope. We’re all in favour of questions round studying on
relational or tabular knowledge, or studying knowledge integration. We now have loads
of concepts to discover round embeddings in databases, studying to
combination, studying on units, graph neural networks for databases, or
distributional matching for entity and schema alignment.
We anticipate to be borrowing instruments (conceptual and sensible) from deep
studying, however to mixing them with strategies from knowledge integration,
information graphs, and databases.
The job posting shall be out quickly, however I’m operating out of the workplace
proper now for holidays (work-life steadiness additionally issues to us).
Variety is necessary
Our staff is just not as
various as I would love it to be (although in all probability doing higher than
typical computer-science staff). We love various candidates. Don’t
hesitate.
