3분기 세미나: 개인 연구주제 발표 (3주차)

---

2019학년도 여름 세미나 3주차 내용입니다.

장소 및 시간

• 장소: 고려대학교 라이시움 321B호
• 시간: 2019년 07월 30일 오후 2시

---

최우성: Attention Mechanism 기반의 음원 분리 기법

발표 내용

• Attention 개념 설명

• Attention-based source separation 구현 및 Demonstration

참고자료

• Reccurent Neural Network: Stanford Lecture - Stanford 231n Lecture Note: Recurrent Neural Network - Stanford 231n Lecture Youtube: Recurrent Neural Network - Stanfrod 231n Lecture Youtube (by Kyoseok Song): Recurrent Neural Network

• RNN, LSTM, Bidirectional RNN

  	- [Vanilla RNN 실습](http://intelligence.korea.ac.kr/members/wschoi/nlp/deeplearning/Vanilla-RNN-%EC%8B%A4%EC%8A%B5/)
  	- [Vannila LSTM 실습](http://intelligence.korea.ac.kr/members/wschoi/nlp/deeplearning/Long-Short-Term-Memory-Network/)
  	- [Bidirectional RNN과 Bidirectional LSTM (이론편)](http://intelligence.korea.ac.kr/members/wschoi/nlp/deeplearning/Bidirectional-RNN-and-LSTM/)
  	- [Bidirectional RNN과 Bidirectional LSTM (실습편)](http://intelligence.korea.ac.kr/members/wschoi/nlp/deeplearning/Pos-Tagging-with-Bidirectional-LSTM/)
  

• Attention Machanism - [ICLR 2015] Bahdanau, Dzmitry, Kyunghyun Cho, and Yoshua Bengio. “Neural machine translation by jointly learning to align and translate.” arXiv preprint arXiv:1409.0473 (2014). - [Paper Review: reniew’s blog]seq2seq with attention: neural machine translation by jointly learning to align and translate