【Keras 2 : 生成深層学習 – CycleGAN】Keras Web サイトのドキュメントの翻訳を進めています。今回は Code examples の「生成深層学習」から「CycleGAN」です。このノートブックは、画像-to-画像変換問題を解くことを目標としたモデル CycleGAN の実装を実演します。[Moe]

【Keras 2 : コンピュータビジョン – Reptile による Few-Shot 学習】今回は Code examples の「コンピュータビジョン」から「Reptile による Few-Shot 学習」です。Reptile アルゴリズムはモデル不可知なメタ学習を遂行するために OpenAI により開発されたアルゴリズムです。このアルゴリズムは最小限の訓練 (few-shot 学習) で新しいタスクを遂行するために素早く学習するように設計されました。 [More]

We algorithmically identify label errors in the test sets of 10 of the most commonly-used computer vision, natural language, and audio datasets, and subsequently study the potential for these label errors to affect benchmark results. Errors in test sets are numerous and widespread: we estimate an average of 3.4% errors across the 10 datasets, where for example 2916 label errors comprise 6% of the ImageNet validation set. Putative label errors are found using confident learning and then human-validated via crowdsourcing (54% of the algorithmically-flagged candidates are indeed erroneously labeled). Surprisingly, we find that lower capacity models may be practically more useful than higher capacity models in real-world datasets with high proportions of erroneously labeled data. For example, on ImageNet with corrected labels: ResNet-18 outperforms ResNet-50 if the prevalence of originally mislabeled test examples increases by just 6%. On CIFAR-10 with corrected labels: VGG-11 outperforms VGG-19 if the prevalence of originally mislabeled test examples increases by 5%. Traditionally, ML practitioners choose which model to deploy based on test accuracy -- our findings advise caution here, proposing that judging models over correctly labeled test sets may be more useful, especially for noisy real-world datasets.

Keras, the deep learning library written in Python, has a new release. Version 2.3.0 is now the first release that supports TensorFlow 2.0. This version adds a few breaking changes and API changes and maintains TensorFlow 1.14 and 1.13 compatibility. For those new to the API, a quick introduction: Keras is a deep learning that's user friendly and uses models as a way to organize layers. It allows for fast prototyping and supports convolutional networks and recurrent networks.