The First Cry of Atom Today is the first day of the rest of my life.

Style Transfer in TensorFlow

CS20SI: TensorFlow for Deep Learning Research の課題の1つとしてStyle Transferを実装してみました。 TensorFlow-StyleTransfer Style TransferはCVPR 2016に出されたGatys et al. 2016が基礎となっているDeep Learningのアート系の技術で、元の画像を特定のスタイルで変換するというものです。 上記の様にどこかの町並みをゴッホが書いた様な画像に変換することができます。使うCNNは2014年のILSVRCというコンペで優勝したVGGというモデルから全結合層を除いたモデル。 畳み込み層は通常深い層に行くほどより高次元の特徴を捉える層になります。という... Read more

Machine Learning on ECS

On the other day, I had a chance to try ECS (Amazon EC2 Container Service). ECS is a container management service provides by AWS. Though I heard about it and thought it must be useful, there were some points to be cared as usual. This is the note for the ECS users and of course for me. What we did I tried to run machine learning algorithm on ... Read more

WireMock in Scala

Mock server which handles HTTP request is useful component for testing HTTP client library. Since it emulates the handling logic of HTTP request artificially, we can verify our HTTP client without relying external service. If you write your application in Scala, I’ll recommend to use WireMock for mock server. This is easy to use and flexible. W... Read more

t-SNE visualization by TensorFlow

From TensorFlow 0.12, it provides the functionality for visualizing embedding space of data samples. It’s useful for checking the cluster in embedding by your eyes. Embedding means the way to project a data into the distributed representation in a space. This technique is often used NLP method and famous by word2vec. Although the detail of word... Read more

Tensorflow Ops

Tensorflow is a famous library for creating deep learning models. We can run efficient deep learning algorithms in Google scale. But it’s more GENERAL library than you thought. I have found that in this lessons. So I want to write several useful tips about Tensorflow Ops. Visualize you graph on Tensorboard Tensorboard visualize your graph in f... Read more