Noisy data learning in deep tech life sciences applications

Project Description

The field of health and life sciences is being revolutionized by ‘deep tech’ applications. With the help of these advanced technologies, often combinations of novel sensing or processing hardware and artificial intelligence, deep tech is facilitating the discovery of new drugs, the development of precision medicine, and the acceleration of medical research. Imec is committed to making a difference in this domain by applying its extensive expertise to develop next-generation and highly specialized sensing technologies that offer an unseen look into the body and its individual components from organs to cells to proteins. 

While the unique hardware emerging from the imec deep-tech activities will generate large amounts of highly valuable data, the scale, resolution, or frequency at which it must operate often results in highly noisy data, which can adversely affect the accuracy and reliability of the models and algorithms developed to analyze the data, leading to poor performance and unreliable results. This challenge is made even more complex by the fact that, given the hardware being developed is often new, not much prior data is available.

The current state of the art of techniques to cope with noisy data encompasses a broad range from unsupervised to (semi)supervised learning methods or approaches like contrastive learning, transfer learning (e.g. domain adaptation techniques) and more. Few shot learning techniques can help to overcome the limited amount of data available at the initial stages of hardware development.

In this project you will investigate and bring together various techniques to cope with noisy (and sometimes limited) data in deep-tech stack sensor development projects. You will work alongside hardware engineers; life sciences domain experts and data scientists. You will help develop and bring together the best techniques to be used at various stages of our hardware roadmaps and so help push the boundaries of innovation in health and life sciences.

The team

You will join the imec AI group (EDiT), which focuses, amongst others, on research and engineering in the domain of Health Analytics. The team is multidisciplinary and international, composed of talent with skills in ML algorithms, sensor fusion techniques, MLOps pipelines and general application development. EDiT is currently a team of around 100 people, operating from the imec offices in Ghent, Antwerp and Leuven. Its main emphasis is on software/hardware co-design and your work will be part of an ongoing effort for disruptive innovation through creative collaboration between the hardware and software departments at imec. 

Initial project plan:

• Explore the various faces of noisy (and limited) data throughout our hardware roadmaps in health and life sciences and how they impact gathering insights from this data
• Explore state-of-the-art techniques for dealing with noisy (often limited) data
• Combine and match these techniques at various stages of hardware development for optimal results
• Contribute to a reusable framework to maximize the impact of this work