The author proposes learning-based schemes for lossy compression in the absence of decoder-only side information, mimicking the achievability part of the Heegard–Berger theorem and operating close to information-theoretic bounds.
学習された圧縮器はHeegard-Berger定理の達成部分を模倣し、情報理論的な限界に近い操作可能な結果をもたらします。