Stochastic Numerics and Statistical Learning: Theory and Applications Workshop

Event Start
Event End
KAUST Campus

About the event

This scientific meeting will concentrate on stochastic algorithms and their rigorous numerical analysis for various problems, including statistical learning, optimization, and approximation. Stochastic algorithms are valuable tools when addressing challenging computational problems. 

For instance, Machine Learning, Stochastic Optimal Control, Data Assimilation, and Bayesian Statistics are hot research areas where these algorithms exhibit their strength. The realm of applications is immense and of great interest to KAUST and the Kingdom. We are flexible in having contributions that either offer mathematical foundations to algorithmic analysis or showcase relevant applications. As a part of this workshop, we will offer two minicourses:

i) Deep Neural Network Robustness, by Prof. Bernard Ghanem (KAUST). This minicourse is a one-day event on May 19

Abstract: Despite their remarkable performance on many learning tasks, Deep Neural Networks (DNNs) behave unexpectedly when exposed to small perturbations in their inputs. This phenomenon is observed even in state-of-the-art DNNs, as they fail to recognize an image when meaningless modifications are applied to it.  For instance, a DNN can provide different predictions for an image and its rotated version, output absurd predictions when few pixels are modified with some predetermined patterns, or fail to recognize a person’s face when wearing certain types of eyeglasses. This pervasive vulnerability inspired researchers to study how these perturbations work and how to train DNNs that are not only accurate but also robust. In this mini-course, we will cover the basics of DNN robustness. In particular, we will give an overview of (1) approaches that provide theoretical guarantees for building robust models, (2) attacks and defenses that work by poisoning the training data of a DNN, and (3) how these perturbations can work in the real-world and how to defend against them.

ii) Machine Learning methods in Computational Finance: from signatures to reinforcement learning, by Christian Bayer (WIAS Berlin), Christoph Belak (TU-Berlin), Blanka Horvath (TU-Munich), and Paul Hager (Humboldt University). This minicourse runs on May 22, 23, and 24.

Abstract: Machine learning methods become more and more prominent in financial engineering, with applications ranging from portfolio optimization and other types of optimal control problems, hedging and risk management, model calibration, to scenario generation and beyond. In this minicourse, we will provide methods for two specific classes of applications, namely optimal control and scenario generation. From a methodological point of view, we will especially concentrate on approaches based on the path signature as input features. Path signatures provide efficient parameterizations of arbitrary, multi-dimensional paths, such as time series. Thus, they are an ideal tool in order to formulate model-free numerical and learning approaches, beyond classical paradigms such as diffusion or even Markov processes. After a gentle, yet comprehensive introduction to path signatures, we will discuss machine learning approaches for optimal stopping and scenario generation based on path signatures. As a final topic for the minicourse, we will introduce reinforcement learning as a prominent alternative for solving stochastic optimal control problems.

This event is organized by Stochastic Numerics PI Professor Raul Tempone (Chair) and Computational Probability PI Professor Ajay Jasra (Co-Chair) with financial support from KAUST.


View Workshop's Agenda.


Recordings of the talks can be found at the youtube channel of the workshop.

ePoster Presentation (Library session on May 23)

Poster presentation

Registration (closed)

For your ePoster presentation, you need to fill in two forms: The Library form and the Registration form. We know that it is not optimal to fill in two separate forms but this is a formal procedure required by the university. Thank you for your understanding.

Deadline for the Library form:

  • Wednesday, May 16

Registration form: Register for presenting an ePoster. Make sure to submit the form even if you only have some of the files/information. You can edit your response later. To edit your answers use the link sent to your email after submitting the form.

Deadlines for the Registration form:

  • Abstract: Wednesday, May 16
  • PDF and LaTeX files: Monday, May 16
  • Pre-recorded poster presentation 5 - 15 min: Monday, May 16
Creation of your ePoster

These instructions are written for PowerPoint but are applicable to any other software you may wish to use. Use PowerPoint 2007 or newer, and set the dimensions of your poster to:

LANDSCAPE: 140cm width x 73cm height (width: 55.0in x height: 28.5in).

 The number of pages (slides): one. 

  • Make sure your text and background have a large contrast (dark lettering on a light background or the reverse)
  • The monitors are full of high definition, and therefore the letters on-screen look nice and smooth. It’s generally advised, though, that for electronic posters to be read on-screen, a minimum 24-28 point size (or bigger) for body text fonts is used to ensure optimal legibility from the usual distance of 3-5 ft. A little larger text (e.g., up to 32 pt size) might also be a good idea as it will provide a comfortable reading from an even larger distance, such as up to 8 ft.
  • For embedded images prefer .jpeg or .png file formats in a resolution of 72 or 96 dpi.
  • Do not use animated effects, “animations,” and animated videos
  • Before submitting, save your poster as a PDF file. All recent versions of PowerPoint and most other software applications allow you to save your poster as a PDF file from the "File > Save as" menu or through the "File > Print > as .PDF."
  • The maximum allowed file size of 5MB


Template for ePoster

You can use this template for creating your ePoster.

Printed poster presentation (Both sessions, Library on May 23 and MOSTI on May 16)

Registration (closed)

Register for presenting a poster on-site. Make sure to submit the form even if you only have some of the files/information. You can edit your response later. To edit your answers use the link sent to your email after submitting the form.

Deadlines for the Registration form:

  • Abstract: Sunday, April 24
  • PDF and LaTeX files: Friday, May 6
  • Pre-recorded presentation 5 - 15 min: Friday, May 13
Creation of your on-site poster

Please make your poster in the size A0, Portrait: 83.96cm width x 118.82cm height (width: 33.06in x height: 46.78in).

Template for poster

You can use this template for creating your poster to be presented on-site.


We will print the posters at KAUST. If you wish to print your poster yourself and bring it, be sure to follow the guidelines specified above regarding its dimensions.

Contact Person