Are you passionate about deep learning (DL) and modern machine learning (ML) techniques? Are you intrigued by understanding how artificial intelligence (AI) models make predictions and inspired by developing to make them fair and responsible? Are you eager to contribute to transforming the Dutch and European industry towards implementing responsible AI solutions? If your answer “yes” to these questions, you might be our next PhD who will develop novel techniques to simplify complex AI models based on deep learning. Your challenge is to simplify the DL models to improve their explainability and ensure the fairness of the decisions.
AI-driven business innovation is crucial in many industries, which, nowadays, often depend on the use of deep learning . Furthermore, it is important to be at the forefront of creating responsible AI solutions that are fair, transparent and societally acceptable. Overparameterization is at the core of the success of deep learning algorithms. For example, Google’s recently developed language model called Switch Transformer has 1.6 trillion free parameters. Although over-parameterization is beneficial to the training and subsequent generalization performance of deep learning algorithms, the resulting models lack transparency, are not robust to adversarial attacks and waste resources due to excessive training times. In this project, you will develop methods to mitigate these problems. The scientific challenge is to develop, evaluate, and validate simplified versions of existing over-parameterized deep learning algorithms, which will be evaluated in terms of transparency, fairness, robustness and resource consumption. A set of guidelines will be defined to identify the most suitable simplification methods and the metrics to apply in practice. A very important part of the work is developing methods for Fairness Quantification to ensure the algorithms’ solutions will not have disparate impact to certain groups, based on attributes like but not limited to gender, race, religion, color, age, and their covariates.
As the PhD working on this challenge, you will become part of the Innovation Center for Artificial Intelligence (ICAI) Lab on Responsible AI in which KPN and JADS scientists work together develop transparent, privacy aware, and personalized AI solutions for businesses. You will also be part of the Computational Intelligence for Decision Support Lab at JADS, collaborating with other PhDs working in the Data Analytics Unit. In this diverse, inclusive and interdisciplinary environment, you are expected to collaborate closely with different stakeholders from KPN, academia and research institutes.
Please find the full job desciption on the website of TU/e
Application can only be done through the online portal of TU/e (see button below). Applications via regular email will not be taken into consideration.
We do cool stuff that matters, with data. The Jheronimus Academy of Data Science (JADS) is a unique cooperation between Eindhoven University of Technology (TU/e) and Tilburg University (TiU). At JADS, we believe that data science can provide answers to society’s complex issues. We provide innovative educational programs, data science research, and support for business and society. With a team of lecturers, students, scientists and entrepreneurs – from a wide range of sectors and disciplines – we work on creating impact with data science. We do this by connecting people, sectors and industries: in the past 5 years we have been working with 300+ organizations on data-related projects. Our main drivers? Doing cool stuff that matters with data. Our location at the former monastery Mariënburg in Den Bosch houses a vibrant campus fully dedicated to data science.
At JADS, you work in an ambitious team of professionals to meet the challenges of tomorrow together. We do cool stuff, that matters. We provide talent opportunities for both scientific staff and support staff. JADS is a human-centered organization; we pay attention to your personal development and value a good work-life balance.