Researchers have developed new Hessian-free algorithms, MOMEHA and MB-MOMEHA, to address multi-objective bilevel optimization problems, particularly those with nonconvex lower levels. These methods utilize the Moreau envelope to transform the problem into a single-level optimization with an envelope constraint. The algorithms maintain computational efficiency by being single-loop and Hessian-free, incorporating a smooth weighted Tchebycheff scalarization. Experiments on few-shot meta-learning and neural architecture search indicate that these new approaches outperform existing methods in terms of Pareto front quality. AI
IMPACT These algorithms could improve efficiency in AI applications like meta-learning and neural architecture search.
RANK_REASON The cluster contains a research paper detailing new algorithms for a specific optimization problem. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- few-shot meta-learning
- MB-MOMEHA
- MOMEHA
- Moreau envelope
- Neural architecture search
- Tchebycheff scalarization
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
