Local Patch AutoAugment With Multi-Agent Collaboration

Data augmentation: A comprehensive survey of modern approaches

“…For automated image augmentation (e.g., [31], [32], [33]) the search space typically consists of a combination of spatial image transformation operations such as rotation, sheer, vertical and horizontal flipping, cropping and scaling, as well as photometric transformations like smoothing, solarization, sharpening, blurring brightness and contrast adjustment, noise addition, and color space conversion operations. Input data samples are augmented by different policies selected according to specific search criteria.…”

“…Many AutoML techniques (e.g., [17], [27], [28], [31], [33], [77], [78], [79] employ a fixed set of pre-defined transfor-mation operations to perform data augmnetation. In these works, the transformations themselves are fixed and can only be varied by learning tunable parameters that control hyperparameters such as augmentation intensities and application probabilities.…”

“…Optimizing DA on validation data is particularly useful as it allows automatically leaned augmentations to generalize well to unseen data. Many subsequent works (e.g., [27], [28], [32], [78], [80]) adopt a similar formulation of augmentation operations in the search space as (in) AA [17] but utilize [17], FastAA [78], FasterAA [27], PAA [33], DADA [28] and BDA [79]-employ this or similar search space construction.…”

“…While most AutoML-based DA methods typically learn effective data augmentation operations for fixed model architectures, an alternative approach to automated data augmentation is to explore the search space for possible neural network architectures that perform well on the given data. AA [17], DDAS [77], [78],FasterAA [27], PAA [33] GA3N [93], AdvAA [80], OnlineAugment [90],…”

“…It is a classi-cal optimization technique for solving ill-formed problems and has wide applications in machine learning. Indeed, a very large number of automated data augmentation ap-proaches (e.g., AA [17], FastAA [78], AWS [115], PAA [33] Learn2Augment [116], Ref. [89] and RTS [64]) utilize RL for the optimization task.…”