Postdoc in sustainable generative artificial intelligence for media Full-time NEW

Postdoc In Sustainable Generative Artificial Intelligence For Media

The five-year research project FrugalAI is exploring the impacts of using generative artificial intelligence (GenAI) for the creation of media content. Current genAI models, used for generating music, images, text, and other media, rely on vast data sets and require significant computational resources. Training genAI models and employing them on a global scale contributes to substantial energy consumption, resource depletion, and increased carbon emissions, exacerbating environmental harm. Our project aims to develop and promote an innovative, sustainable alternative to existing large and commercial genAI systems. It aims to reduce the environmental footprint of genAI technologies while fostering positive social impacts by prioritizing small, ethically sourced data sets, local models and compute, and encouraging sustainable behavior among users.

The research and development conducted in the project can be regarded as situated within the area of human-centered AI. FrugalAI has the following objectives:

1. Estimate the environmental impact of current big-data genAI.
2. Construct a set of open-source frugal genAI tools based on small, ethical data.
3. Map the impact on social sustainability that emerges from using frugal genAI.
4. Compare the environmental impact of these tools with big-data genAI tools.
5. Propose and apply evaluation metrics that consider resource use.
6. Investigate implications for other genAI applications (e.g., image generation).

The postdoctoral researcher, having a background related to HCI and machine learning, will contribute to all objectives over a period of two years at 100% time. Working with a diverse team of engineers, musicologists, and designers, the job will include the following tasks:

• Contribute to the development of an interactive online broadcasting environment (a prototype is “Stochastic Pirate Radio” KSPR) by designing and maintaining project software, and developing pipelines for creating media programs and the automated monitoring for offensive content, and estimating and tracking energy usage.
• Research and develop frugal AI principles to reduce model sizes, data, and computation using the state of the art in machine learning.
• Develop a multi-objective metric of genAI sustainability (environmental, economic, sociocultural), building from C. Douwes, On the Environmental Impact of Deep Generative Models for Audio, Doctoral dissertation, Sorbonne Université 2023).
• Help organize and run workshops with team members

For more information, contact Bob L. T. Sturm (bobs@kth.se).

KTH shapes the future through education, research and innovation. As a leading international technical university, we play an active role in advancing the transition towards a sustainable society. At KTH, you have the opportunity to grow and develop in a creative and dynamic environment, with good working conditions and attractive benefits. Equality, diversity and equal opportunities are essential to quality and form an integral part of KTH's core values as a university and public authority.

Learn more about our benefits and what it's like to work and grow at KTH.

The position offered is for, at the most, two years. A position as a postdoctoral fellow is a time-limited qualified appointment focusing mainly on research, intended as a first career step after a dissertation.