SNSF supports two projects of the School of Computer Science with CHF 1.6 million

An overview of the two projects of the Institute of Computer Science (ICS-HSG): 

Better data protection: SNSF-funded project in the field of software engineering

Most software systems, including social networks or office and e-commerce applications, are managed centrally. They force users to share their data with third parties, e.g. with a cloud provider. The project ‘Consistency Programming for Local First Software’ by Prof. Guido Salvaneschi, which is being funded by the SNSF with CHF 997,524 for a period of four years, is researching a different approach. In the context of ‘Local First Software’, users would retain control over their data and only share it in a conscious and controlled manner when necessary. A number of technical challenges need to be overcome to achieve this: For example, software developers must ensure that their applications work properly locally and not only when they are connected to a cloud. Data that is available in different locations is no longer synchronised. The project is therefore developing methods to ensure that software applications function smoothly and correctly on different devices without the need for constant access to the data in the cloud. 

The central questions to this research are:

• How can distributed software ensure that users only share the data they actually need with service providers while retaining ownership of their data? 
• How can software tools be developed that enable applications to function efficiently and correctly even without permanent internet access?
• What methods are best suited to ensure that data remains consistent and accurate across different services and user interactions, even without constant access to the cloud, even in regions with poor internet connectivity?

The project promises the following benefits for society: 

• Improved data security and privacy.
• Full functionality of applications even with interrupted or no internet connection – data loss is avoided, and the responsiveness of applications is improved.
• Promote inclusion and equal access to digital resources by improving application performance in regions with limited infrastructure.

Improve trustworthiness of AI: Project in the field of machine learning

Current research in the field of artificial intelligence and machine learning is mainly focused on investigating deep neural networks and their structures – and thus understanding their robustness and behaviour. These questions are important to make deep neural networks – the ‘heart of GenAI’, so to speak – much more reliable and trustworthy. The project ‘Hyper-Representations: Learning from Populations of Neural Networks’ by Prof. Dr. Damian Borth, which the SNSF is supporting with CHF 608,700 for four years, sheds light on these questions from a new perspective. Instead of investigating individual models, the project is focussing on populations of neural networks. This approach should provide a better insight into the structure and behaviour of neural networks than would be possible with a single model. Based on the findings, a basic model will be trained.

Central research questions are: 

• Do populations of neural networks form a structure in weight space?
• Can we learn a low-dimensional representation by learning these weight space structures?
• How can the weights of the neural network be extended for efficient training?
• How can the proposed approach for learning hyper-representations be applied to large models?
• How can we utilise the structure of multiple populations of neural network models to enable efficient transfer learning between populations?

The project promises the following benefits for society: 

• Improving the trustworthiness of artificial intelligence.
• Improving sustainability within artificial intelligence: neural networks should be trained much more efficiently by reducing computational and energy requirements.