September 2018 – September 2019

MSc in Machine Learning

Royal Holloway, University of London

Modules include:

  • Machine Learning
  • Online Machine Learning
  • Deep Learning
  • Distributed Systems
  • Data Analysis
  • Programming in Matlab
February 2018 – October 2018

Applied Researcher in NLP with Deep Learning


  • Analysed the result deeply and define the issue arising in the approach.
  • Considered the solution and actively discuss with other team-members.
  • Implementated the algorithms to our products.

Resarch Topics:

  • Intent Extraction
  • Named Entity Extraction
  • Deep Learning
  • Applied Reinforcement Learning in NLP
June 2015 – February 2018

Data Scientist for Customer Acquisition


  • Extracted data from Database using SQL.
  • Analysed the customer behaviour.
  • Created the machine learning applied model to enhance the marketing strategies.

Recent Posts

More Posts

Introduction Hello everyone! It’s been almost more than a few decades since the theoretical importance of DL was academically proposed. Needless to say, but I have been studying theoretical aspects of DL long. But when it comes to deep understanding, we cannot avoid the actual experiment. So, these days Google or Apple or other IT giants put more efforts for contributing the open source DL libraries enhancing the development. Hence, I would like to study and review keras APIs to show the sample usage of them in this article.


Foreword I have long been interested in working for Google and been wondering what kind of people could actually pass their interview.. So I have decided to do some research about their content of the interview and also their expectation for candidates. Google’s Interview Process CV Screening: Everyone struggling with this step. Why can’t my CV pass screening for Google and Facebook despite having Amazon in it? Phone/Hangout Screening: Usually takes 30 - 60 minutes with Potential Peer/Manager On-site Interview: Meet Four Googlers for 35-40 minutes each Expectation on Candidates General Cognitive Ability: To explain your answer for open-ended questions in a smart manner, e.


Introduction Gradient descent optimisation algorithms, while increasingly popular, are often used as black-box optimizers, especially when it comes to the actual implementation using some DL libraries. Indeed, practical explanations of their strengths and weaknesses are hard to come by. This article aims to provide the reader with intuitions with regard to the behaviour of different algorithms that will allow us to put them to use. Original Papaer : https://arxiv.org/pdf/1609.04747.pdf


Profile Titile: MultiAgent Reinforcement Learning: Independent vs Cooperative Agents Author: Ming Tan Published Year: 1993 Link: http://web.media.mit.edu/~cynthiab/Readings/tan-MAS-reinfLearn.pdf Abstraction Since the author got inspired by the learning behaviour of human beings, he investigated the multi-agent in reinforcement learning by comparing two generic assumption. Agents who can cooperate with other agents by sharing information Agent who cannot cooperate with others And he found that basically in the cases described below, agents could efficiently learn through the cooperation each other.


今回はパーセプトロンについてまとめたいと思います。 イントロ 械学習を大別すると分類問題と数値問題に分けられます. 人間の脳細胞ニューロンをまね, コンピューター上で最初に表したものがパーセプトロン(perceptron)です. 1957年にローゼンブラット(Rosenblatt)によって,データを2つのクラスに線形で分離するために考案されたアルゴリズムです. それゆえ非線形問題は解決できず, そのために多層パーセプトロンという、パーセプトロンを多層に重ねたアルゴリズムが考案されました. ニューロンのイメージ 画像もと:http://tsushan.hatenadiary.jp/entry/2018/03/10/200305 数学的な背景 ここからちょっと数学的な背景の説明を行なっていきたいと思います。 まずはざっと数学的な記号を付した下記の図をご覧ください。 画像もと:https://towardsdatascience.com/what-the-hell-is-perceptron-626217814f53 そして、この図を元にパーセプトロンを数式で表すと次のように書けます. また、数式内部の変数については、下記の様な読み替えをして見てください。 $x$:入力 $h(x)h(x)$:出力 $w$:重み $b$:バイアス \begin{align} h_w (x) & = w_0+w_1x_1+\cdots +w_nx_n\\ &= \sum_{i=0}^n w_ix_i\\ &=\boldsymbol{w}^T\boldsymbol{x}\\ \end{align} ここまでの段階では、ただの線形回帰の式となんら変わりはございません。 しかし、パーセプトロンの違うところはここからになります。 このアルゴリズムを用いて線形分離をしていきたいので、この吐き出された $h_w (x) = \boldsymbol{w}^T\boldsymbol{x}$ に対して、閾値($\theta$)を設けます。 h_w (x) = \boldsymbol{w}^T\boldsymbol{x} \geq \theta \\ or\\ h_w (x) = \boldsymbol{w}^T\boldsymbol{x} < \theta \\ というにこの分け方ができます。 しかし、これでは見辛いので少々整形します。 h_w (x) = \left\{ \begin{array}{ll} 1 & (\boldsymbol{w}^T\boldsymbol{x} \geq \theta) \\ 0 & (\boldsymbol{w}^T\boldsymbol{x} \lt \theta) \end{array} \right.