Lamnoo
Lamnoo Inc.
LaMOO (Multi-objective Optimization by Learning Space Partitions) is a software framework and research project, not a formal company, originating from institutions like Nanjing University. It provides a novel multi-objective optimizer designed to efficiently solve complex problems with multiple conflicting objectives, such as finding optimal neural network architectures. The open-source framework aims to significantly speed up the search for trade-off solutions, known as the Pareto frontier, in large and costly search spaces.
Products & Team
LaMOO Software Framework
The primary offering is the LaMOO software framework, an open-source multi-objective optimizer. It functions as a meta-algorithm that can enhance existing solvers by using a classifier to learn from observed samples, partition the search space, and focus computational effort on the most promising regions to find optimal solutions more quickly.
LaMOO solves the problem of high computational cost and slow convergence in multi-objective optimization. It provides a more sample-efficient method for finding a set of optimal trade-off solutions for complex design problems in AI and computational science.
The target audience struggles with the inefficiency and high computational expense of traditional methods when trying to optimize for multiple, often conflicting, objectives simultaneously, such as balancing model accuracy with latency and size in Neural Architecture Search.