Internship - Temporal Difference Back Propagation - 6 months

Blu Analytics is a systematic event trading fund that uses NLP to anticipate large market moves based on newsand generate investment signals. Further, it uses Deep Reinforcement Learning to infer true market behaviour,
predict patterns, devise quantitative strategies, and optimise portfolio allocation.
These results are then applied to dynamically maximise Risk/Return for private investors and retail market.

The aim of the project is to develop and improve the Temporal Difference Back Propagation (TDBP) models, a particular type of Reinforcement Learning models, to compute conditional expectations.
Some of the tasks to perform during the project include:
  • Build an encoder-decoder module to encapsulate the underlying asset to account for non-Markovian processes
  • Variance reduction: apply control variate to reduce variance of the conditional expectation, improving accuracy and convergence of the TDBP models
  • Improve gradient-descent: true-online TD, corrected momentum, accelerated gradient method (deriving approx- imate second-order methods)
  • Improve and test: convergence, generalisation, sample efficiency

Skills required:
  • Good understanding of machine learning (deep neural networks, recurrent neural networks, queueing networks)
  • Good understanding of reinforcement learning
  • Very good knowledge of Python (Tensorflow and Pytorch)
  • No need to know financial markets

  • the trainee can choose any convenient location (teleworking/full remote)
  • to be discussed, minimum EUR 1.600/month
Starting date:
  • ASAP, for six months with the possibility of employment afterwards

Academic level:
  • Master Degree and over

Apply by email.