[세미나/포럼] [2021.12.23(목)] AI Convergence Network & Artificial Intelligence Colloquium
<AI Convergence Network & Artificial Intelligence Colloquium> * Title : Deep learning for end-to-end communications * When : 2021.12.23.(THU) A.M.11:00~ * Where : Zoom - Zoom 회의 참가 https://zoom.us/j/98276658445?pwd=blNFNDFoTGhYcDNXbGtCU1h4SnVHZz09 회의 ID: 982 7665 8445, 암호: 3647 * Speaker : Prof. Mojtaba Vaezi (Villanova University) * Abstract : Deep learning has successfully been applied to many communication problems in the physical layer. Using deep learning for end-to-end communication is, however, a novel concept with significant potential. Particularly, deep autoencoder (DAE)-aided communication has shown to be very competitive to the traditional block-based communication. In this presentation, we introduce the applications of DAEs in communication systems and discuss state-of-the-art applications of DAEs for point-to-point multiple-input multiple-output (MIMO) channels. Specifically, we show that embedding left- and right-singular vectors of the channel matrix into the DAE encoder and decoder further improves the performance compared to the state-of-the-art in terms of bit error rate (BER). A proper DAE design can largely outperform Shannon-theoretic-based linear precoding in terms of BER. We also discuss future directions in this road including using DAEs for interference management.