Auptimizer

2019: Announcement

In November, 2019 released LG the software tool open source which is intended for optimization of models of artificial intelligence. The innovation under the name Auptimizer developed in division of LG Advanced AI allows developers to configure the AI systems and to keep system accounting, for example, memory distribution.

In the description as it is said to the project that Auptimizer simplifies process of setup of group of models with different configurations.

Released LG the software tool open source which is intended for optimization of the AI models

The same types of models of machine learning can require the different assumptions, weight or the speed of training for different types of data. These parameters are called hyper parameters and they should be configured so that the model could solve a problem of training optimum. For this purpose there is a train of hyper parameters which gives the optimal model optimizing given function of losses on the set independent data.

Like other algorithms of finding of hyper parameters, Auptimizer initializes search space and a configuration before offering values for hyper parameters then trains target model and updates results. Then it repeats stages of the offer, training and updating, will not define best values yet.

Auptimizer requires only several code lines, and it helps users to pass step by step to make the settings connected with the project. The tool supports switching between different algorithms of hyper parameters and computing resources, without requiring from users to rewrite the training script, and is intended for expansion to other algorithms and resources without significant changes.

Source codes of Auptimizer are laid out on the GitHub portal.

^[1][2]

Notes