One of the most difficult tasks to solve when using neural network, is to define the optimal configuration of layers and connections to reach the best results in the resolution of a problem.
In this paper, the authors of the Cornell University, introduce a new graphical framework capable to plot the research space in which the neural network is optimized during the training phase. Thanks to this visual representation it is possible to easily verify if the architecture of the neural network can generate optimal results.
In Humanativa, this tool is really precious. Indeed, it give us the possibility to understand the “quality” of our neural network and then, it allows us to improve the quality of our machine learning solutions.
Info on: https://arxiv.org/pdf/1712.09913.pdf