NAVER LABS Europe: Internship  

Start date: Fall 2019 or early 2020

Duration; 5-6 months

We are opening a research internship on Reinforcement Learning
techniques for applications to controlled text generation.  Under
conditions where training data is limited, standard end-to-end
training of seq2seq models may generalize poorly and produce
inadequate results at test time. A possible remedy is to augment
models with rewards that control the quality of the outputs. These
rewards can address two complementary goals: (i) taking into account
global characteristics of observed sequences that go beyond standard
local teacher-forcing training techniques (observation bias problem),
and (ii) moving the generation process towards desired properties of
the output (e.g. favoring shorter sentences or performing style
transfer).

Supervisors: Marc Dymetman and Hady Elsahar.

We are looking for a motivated intern to help us develop methods and
algorithms for addressing this general problem, both in theory and in
practice. Experiments will be conducted on selected text generation
tasks (NLG, Summarization or Machine Translation).

The successful candidate should be enrolled in a graduate program, at
the Master or (preferably) PhD level, with experience (ideally) in
Deep Learning, Reinforcement Learning and Natural Language Processing.

Publication of results in major conferences/journals will be strongly
encouraged.

REQUIRED SKILLS :

Strong mathematical and programming skills as well as familiarity with
one of the major current deep learning toolkits (PyTorch preferred but
not compulsory) are a requirement.

For more information and for applying, please visit the link below:

https://europe.naverlabs.com/job/reinforcement-learning-for-controlled-text-generation/