The Kavli Institute for Cosmology, Cambridge (KICC) would like to invite cosmologists to join official Kavli Visitors Benjamin Wandelt and Justin Alsing for a workshop on Likelihood-free Inference.
The Likelihood forms the cornerstone of all traditional astrophysical inference and statistical techniques (both Bayesian and frequentist). However, directly computing an efficient likelihood is only possible for a subset of astrophysical datasets (such as Planck cosmic microwave background data). Likelihood-free inference (LFI) is an emerging paradigm in cosmology, acting as a natural successor to approximate Bayesian computation (ABC). Far from removing likelihoods from the pipeline, LFI is an approach for machine-learning an unknown likelihood from physical simulations. This proxy likelihood can then be used to apply the existing arsenal of inference techniques to a much wider class of problems.
In this project-driven workshop, researchers can learn the theory and practice of LFI from the world experts. By the end of the workshop, attendees should understand the key concepts of massive compression and density estimation for likelihood learning, will have access to and be able to run the industry-standard codes and be equipped to apply these techniques to modern astronomical problems.
The workshop will extend over three days, beginning in the morning of Tuesday 31st March and ending midday Thursday 2nd April and is appropriate for PhD or post-doc level researchers.
Attendance and lunch are free, but we are limited to up to 25 attendees, so please register your interest ASAP.
The intention is to prioritise local participants, with external candidates considered on a 'first come, first served' basis.
Non local attendees should arrange their own accommodation. Please see here for a list of recommended places to stay.
Further information
- pydelfi on GitHub: https://github.com/justinalsing/pydelfi
- Fast likelihood-free cosmology with neural density estimators and active learning: https://arxiv.org/abs/1903.00007
- Sequential Neural Likelihood: Fast Likelihood-free Inference with Autoregressive Flows: https://arxiv.org/abs/1805.07226
- Likelihood-free inference with emulator networks: https://arxiv.org/abs/1805.09294
- Massive optimal data compression and density estimation for scalable, likelihood-free inference in cosmology: https://arxiv.org/abs/1801.01497
Registration will close on Friday 13th March
Successful applicants will be notified by Monday 16th March
IMPORTANT:
We are closely monitoring the situation with regard to coronavirus. Should it become necessary to cancel the workshop, we will notify you as soon as possible.
In the meantime, further information and guidance is available here
