Statistician's Deeplearning
    • 블로그
    • Github
    • 논문/Git 모음
    • About Me

    Statistician's Deeplearning

    급변하는 분석 패러다임의 변화의 물결을 온몸으로 받아내는 통계학자입니다.
    통계와 머신러닝의 차이는 무엇일까요?

    Featured Posts
    • Dynamic Few-Shot Visual Learning without Forgetting

      시작하며 데이터 분석에 있어서 비지니스 환경 하에서는 언제나 imbalance 문제에 시달리게 됩니다. 새로운 상품은 계속 나오고 그 상품이 충분히 팔리기 전에 뭔가를 알고 싶어하는 마음이 큰 거죠. 그런 정보가 있으면 보다 효과적으로 새로운 상품을 고객에게 appeal할 수 있을 거니깐요. Few shot learning은 이러한 문제를 해결하기에 아주 적합한 모형입니다. 되기만 하면 말이죠. 그러한 가능성을 tapping해보는 관점에서 Dynamic Few-shot Visual Learning without Forgetting을 읽는 중에 정리도 할겸 posting 하고자 합니다. Few-shot learning은 약 2~3년 전쯤에 한참 화제가...

      deeplearning Few-shot learning attention cosine similarity

      kion.kim's profile image

      kion.kim

      2018-08-09 17:00

    • Sentiment Analysis - Self attention based on Relation Network

      Introduction There are many methods for sentence representation. We have discussed 5 different ways of sentence representation based on token representation. Let’s briefly summarize what is dealt with in the previous posts. What we have discussed so far… Just averaging token embeddings in sentence works pretty well on text classification problem. Text classification problem, which is relatively easy and simple task, does not need to understand the meaning of the sentence in semantic way but...

      deeplearning self-attention nlp sentence representation

      kion.kim's profile image

      kion.kim

      2018-08-02 17:00

    • WGAN and WGAN-GP

      Introduction It has been a while since I posted articles about GAN and WGAN. I want to close this series of posts on GAN with this post presenting gluon code for GAN using MNIST. GAN is notorious for its instability when train the model. There are two main streams of research to address this issue: one is to figure out an optimal architecture for stable learning and the other is to fix loss function, which...

      deeplearning GAN Image generation MNIST Wasserstein distance

      kion.kim's profile image

      kion.kim

      2018-07-26 17:00

    • Sentiment Analysis - Convolutional Neural Network

      Introduction Let’s think about the way human understand sentence. We read the sentence from left to right (it is not the case in the ancient asisan culture though) word by word memorizing the meaning of words first. Words themselves may have very different meaning depending where they are placed or how they were used. To understand real meaning of words, we break the sentence down into smaller phrases, groups of words, to get the right...

      deeplearning nlp sentence representation CNN sentiment analysis text classification

      kion.kim's profile image

      kion.kim

      2018-07-18 17:00

    • Sentiment Analysis- Self Attention

      시작하며 LSTM을 이용해서 문장의 여러 특성들을 뽑을 수 있습니다. 지난 블로그들에서는 주로 hidden state의 정보를 이용해서 문장을 표현하는 코드들을 짜보았는데, 사실 hidden state의 정보 이외에도 각 time step의 ㅡ로을 이용해서 문장을 요약할 수도 있을 것 같습니다. 하지만 각 time step의 output은 seq2seq 문제에서 실제 그 진가를 발휘합니다. 이전 단계의 output이 그 다음 time step의 input으로 들어감으로써, 순차적으로 문장을 생성할 때 유용하게 사용됩니다. 나중에 nmt 쪽에서 살펴 보려고 합니다. Self Attention 지금까지 여러가지 모형을 만들어 보면서...

      deeplearning LSTM nlp sentence representation

      kion.kim's profile image

      kion.kim

      2018-07-12 17:00

    • No Previous Page
    • 1
    • 2
    • 3
    • Next Page
    • github
    • facebook
    • twitter
    • rss

    Copyright © Kion Kim All rights reserved.