Postdoc in sustainable generative artificial intelligence for media

Tasks

The five-year research project FrugalAI investigates the effects of using generative artificial intelligence (GenAI) to create media content. Current GenAI models, used to generate music, images, text, and other media, rely on enormous datasets and require extensive computational resources. The training and large-scale use of GenAI models contribute to significant energy consumption, resource consumption, and increased carbon emissions, exacerbating the environmental impact. The project aims to develop and promote an innovative and sustainable alternative to today's large, commercial GenAI systems. The goal is to reduce the environmental impact of GenAI technology while simultaneously promoting positive social effects by prioritizing small, ethically collected datasets as well as local models and computational resources, while encouraging users toward sustainable behavior.

The research and development within the project can be placed within the field of human-computer interaction and AI (human-centered AI). FrugalAI has the following goals:

• Estimate the environmental impact of current GenAI systems that rely on large datasets (big data).
• Develop a set of open-source tools for resource-efficient (”frugal”) GenAI, based on small and ethically collected datasets.
• Map the effects on social sustainability when using resource-efficient GenAI.
• Compare the environmental impact of these tools with GenAI tools based on large datasets.
• Propose and apply evaluation metrics that take resource use into account.
• Investigate the consequences for other GenAI applications (e.g., image generation).

The postdoc, who has a background in HCI and machine learning, will contribute to all goals during a two-year full-time period. The work takes place in an interdisciplinary team consisting of engineers, musicologists, and designers, and includes the following tasks:

• Contribute to the development of an interactive environment for web broadcasting (a prototype is ”Stochastic Pirate Radio”, KSPR) by designing and maintaining the project's software, developing workflows for creating media programs and for automated monitoring of offensive content, as well as estimating and following up on energy consumption.
• Research and develop principles for resource-efficient AI with the aim of reducing model sizes, datasets, and computational needs using the latest technology in machine learning.
• Develop a metric for sustainability within generative AI (environmental, economic, and socio-cultural sustainability) that encompasses multiple goals, based on C. Douwes' thesis *On the Environmental Impact of Deep Generative Models for Audio* (Sorbonne Université, 2023).
• Contribute to organizing and conducting workshops together with team members.

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

Qualifications

Requirements

• Completed doctoral degree or a foreign degree assessed to correspond to a doctoral degree. This eligibility requirement must be met no later than when the employment decision is made.
• The project's working language is English, so you should be proficient in English.
• We would prefer that you have a background in HCI and machine learning.
• You have good collaborative skills.

Meritorious

• At most three years since the doctoral degree or a foreign degree assessed to correspond to a doctoral degree was obtained by the final application date.
• Awareness of diversity and equal treatment issues, with a particular focus on gender equality.
• It is meritorious if you can work independently.

We will place great emphasis on personal qualities.

Become a part of KTH

KTH shapes the future through education, research, and innovation. As a leading international technical university, we play an active role in driving and contributing to the transition to a sustainable society. Here, you are offered the opportunity to grow and develop in a creative and dynamic workplace with good working conditions and benefits. Equality, diversity, and equal opportunities are matters of quality and an inherent part of KTH's values as a university and a government agency.

Read more about our benefits and what it is like to work and develop at KTH.

Union representatives

Contact details for union representatives.

Application

You apply via KTH's recruitment system. As an applicant, you are primarily responsible for ensuring that your application is complete when submitted.

The application must contain:

• CV with details regarding relevant professional experience and competence.
• Copies of degree certificates and transcripts from previous university studies.
• A personal letter where you describe how your background matches the tasks in this position as well as your interest in the overall project. Maximum two pages long.

The application must reach KTH no later than the final application date at midnight, CET/CEST (Central European Time/Central European Summer Time).

About the employment

The employment is permanent, but for a maximum of two years.

An employment as a postdoc is a fixed-term qualifying position primarily focused on research, intended as a first career step after graduation.

Other

For information regarding the processing of personal data in connection with recruitment.

An employment at KTH may be subject to security clearance. If this is the case for this specific employment, a security screening of the applicant will be conducted in accordance with the Security Protection Act (2018:585) after consent. In these cases, a prerequisite for employment is that the applicant is approved after the security screening.

We decline direct contact with staffing and recruitment agencies as well as sellers of job advertisements.