Machine Unlearning: Model-Data Co-Design
Invitation Code: RFP-24-04

As machine learning models become increasingly sophisticated and deployed in dynamic environments, the need for efficient adaptation mechanisms has become paramount. Traditional approaches focus on learning from data, but there is a growing recognition of the importance of unlearning outdated or irrelevant knowledge to maintain model relevance and accuracy. This RFP seeks proposals to explore machine unlearning techniques and their applications in enhancing model adaptability and performance across various domains.

Scope of Work: Proposals are invited for research projects that address the following objectives:

  • Investigate novel machine unlearning methodologies: Develop cutting-edge approaches for selectively updating or removing learned knowledge from machine learning models, considering factors such as privacy preservation, bias mitigation, model fairness, harmful content elimination, and resource efficiency.
  • Study the impact of unlearning on model performance: Conduct empirical evaluations to assess how unlearning techniques influence model adaptability, robustness, and predictive accuracy across different datasets and application scenarios.
  • Investigate how machine unlearning and the progression in the realm of security and privacy (SP) is interconnected. Explore the emerging SP applications propelled by unlearning, particularly in trustworthy machine learning.
  • Tackle the unlearning challenges in the era of big data: Improve the current unlearning algorithms in the practical implementation perspectives, including but not limited to data dependency, model complexity, huge computational cost, dynamic environments in continue learning, and privacy leaks. Research should also consider how unlearning can adapt to and evolve with the data scaling laws prevalent in big data contexts.
  • Develop benchmark datasets and evaluation metrics: Create standardized benchmarks and evaluation protocols to measure the effectiveness and efficiency of machine unlearning algorithms, facilitating comparison and reproducibility across research studies.
  • Explore applications of unlearning in real-world settings: Apply developed unlearning techniques to practical problems, in domains like computer vision, natural language processing, federated learning, graph neural networks, and multi-modality models.

Proposal Submission:

After a preliminary review, we may ask you to revise and resubmit your proposal.

RFPs may be withdrawn as research proposals are funded, or interest in the specific topic is satisfied.

Researchers should plan to submit their proposals as soon as possible.

General Requirements for Consideration, Proposal Details, FAQs

You can find the information by scrolling down to the bottom of the webpage: Research Gifts. If your questions are not answered in the FAQs, please contact research@cisco.com.

Constraints and other information

IPR will stay with the university. Cisco expects customary scholarly dissemination of results and hopes that promising results would be made available to the community without limiting licenses, royalties, or other encumbrances.