7.2 How to evaluate the model performance?

    (Robert) TODO - 目前 tensor board 沒有加上 f1 score 等 metrics, 待調整後補上

    [ 待補上 metrics image ]

    At each (local) training epoch, a local validation will be processed and the result will be sent to Aggregator (at Gitea server). Summarized validation results will be displayed on the Tensor Board like the following picture. Therefore, PI can evaluate the training results of each edge training in each epoch and choose the best training result.

    If PI is not satisfied the best training result, he can create another training plan trying different aggregation algorithm, training models, and/or different training datasets.

    Taiwan AI Labs (AILabs.tw) Copyright © 2023Powered by Bludit