Evaluation of Representation Models for Text Classification with AutoML Tools

Towards robust real-world historical handwriting recognition

“…This is not tenable in a highly diverse and large manuscript collection which is in a flux due to the ongoing labeling by a user community. Although efforts have been spent in this area, such as Google's 'AutoML' [160]- [164] service, the problem of autonomous training has not yet been solved satisfactorily [165], [166]. Therefore, as the final topic in the 'robustness theme', we will need to explore methods that allow an overfit of neuralnetwork models, without relying on a label-greedy validation scheme in k-fold evaluation.…”