The Wholesale Banking Advanced Analytics team is a large team of data scientists, data engineers, software developers and many more, that are focused on bringing data, machine learning and statistical modeling into the products that we build for our clients or internal users. The data scientists in WBAA furthermore have a strong desire to keep up with and be part of the latest developments in the fields of AI, tooling and statistics. Which they do by working closely together with master’s students on a variety of topics to solve academic yet practical problems.


Master thesis project in applying Generative Adversarial NetworksThe Wholesale Banking Advanced Analytics team is a large team of data scientists, data engineers, software developers and many more, that are focused on bringing data, machine learning and statistical modeling into the products that we build for our clients or internal users. The data scientists in WBAA furthermore have a strong desire to keep up with and be part of the latest developments in the fields of AI, tooling and statistics. Which they do by working closely together with master's students on a variety of topics to solve academic yet practical problems.The foreign exchange (FX) market is constantly moving. Prices are updated on a sub-second level. ING has stored millions of such prices collected over years. Being able to forecast price distributions ahead of time could be quite beneficial.At ING we have done research in using Generative Adversarial Networks (GAN) to forecast time series (thesis). The GAN is trained to reproduce the distribution of potential futures. With that one can then estimate profits or design policies on how to trade.The first study has shown that it is possible to train a conditional WGAN to reproduce a 1D time series distribution. However, FX prices are not 1D. There are bid and ask prices, and furthermore multiple of them from multiple players, together at any one moment forming an order book. We invite a student to continue the research to forecast multi-dimensional time series and test it on order book data.Learning distributions from data has some theoretical questions: which loss functions to optimize, the effect of regularization, how good the distributions can be expected to be reproduced. Together with your supervisor you will think about these questions.On the practical side, learning with a neural network has its own challenges: adversarial training, architecture, convergence, complexity. You will hopefully overcome these and build a solution that can learn multi-D time series distributions.You will show in laboratory settings (on dedicated synthetic data sets) that you understand your solution next to that it works. Finally you will test it on historical order book data.Are you a master's student looking for a thesis project and are you interested in this one.Do you furthermoreHave solid experience with Python and neural network libraries (Pytorch, Tensorflow, JAX)?Have solid understanding of (adversarial) neural networks, loss functions and their optimization?Have solid skills in statistics?Get at least six months to do your thesis project?Aim to go for a publication?Bring good vibes to your fellow data scientists?Then we offer a master thesis project, a compensation of 600 euros per month, close supervision, and a tight community of data scientists to interact with and learn from.

Wat ga je doen?

The foreign exchange (FX) market is constantly moving. Prices are updated on a sub-second level. ING has stored millions of such prices collected over years. Being able to forecast price distributions ahead of time could be quite beneficial.

At ING we have done research in using Generative Adversarial Networks (GAN) to forecast time series (thesis). The GAN is trained to reproduce the distribution of potential futures. With that one can then estimate profits or design policies on how to trade.

The first study has shown that it is possible to train a conditional WGAN to reproduce a 1D time series distribution. However, FX prices are not 1D. There are bid and ask prices, and furthermore multiple of them from multiple players, together at any one moment forming an order book. We invite a student to continue the research to forecast multi-dimensional time series and test it on order book data.

Learning distributions from data has some theoretical questions: which loss functions to optimize, the effect of regularization, how good the distributions can be expected to be reproduced. Together with your supervisor you will think about these questions.

On the practical side, learning with a neural network has its own challenges: adversarial training, architecture, convergence, complexity. You will hopefully overcome these and build a solution that can learn multi-D time series distributions.

You will show in laboratory settings (on dedicated synthetic data sets) that you understand your solution next to that it works. Finally you will test it on historical order book data.

Wat krijg je er voor terug?

  • a compensation of 600 euros per month,
  • close supervision,
  • and a tight community of data scientists to interact with and learn from.

Wat ga je doen?

The foreign exchange (FX) market is constantly moving. Prices are updated on a sub-second level. ING has stored millions of such prices collected over years. Being able to forecast price distributions ahead of time could be quite beneficial.

At ING we have done research in using Generative Adversarial Networks (GAN) to forecast time series (thesis). The GAN is trained to reproduce the distribution of potential futures. With that one can then estimate profits or design policies on how to trade.

The first study has shown that it is possible to train a conditional WGAN to reproduce a 1D time series distribution. However, FX prices are not 1D. There are bid and ask prices, and furthermore multiple of them from multiple players, together at any one moment forming an order book. We invite a student to continue the research to forecast multi-dimensional time series and test it on order book data.

Learning distributions from data has some theoretical questions: which loss functions to optimize, the effect of regularization, how good the distributions can be expected to be reproduced. Together with your supervisor you will think about these questions.

On the practical side, learning with a neural network has its own challenges: adversarial training, architecture, convergence, complexity. You will hopefully overcome these and build a solution that can learn multi-D time series distributions.

You will show in laboratory settings (on dedicated synthetic data sets) that you understand your solution next to that it works. Finally you will test it on historical order book data.

Wat krijg je er voor terug?

  • a compensation of 600 euros per month,
  • close supervision,
  • and a tight community of data scientists to interact with and learn from.

Wie is ING

Positieve veranderingen realiseren door onze expertise, middelen en maatschappelijke positie op de juiste manier in te zetten. Zo bouwen wij voort op een prachtige geschiedenis. In 1881 begon ING mensen te helpen met sparen. We waren de eerste bank die kleine bedrijven vooruit hielp. En dat is waar onze passie ligt, met een toegankelijke, positieve instelling mensen vooruit helpen. Overal en altijd.

De weergegeven reviews zijn van oud stagiairs.
Nog geen reviews beschikbaar

Vragen over deze vacature?

Team Mediastages