Pretorius, ArnuKroon, SteveKamper, Herman2019-03-072019-03-072018-07Pretorius, A., Kroon, S. and Kamper, H. 2018. Learning dynamics of linear denoising autoencoders. ICML 2018: The 35th International Conference on Machine Learning, Stockholm, Sweden, 10-15 Jul 2018, pp. 4141-4150https://arxiv.org/abs/1806.05413http://hdl.handle.net/10204/10746Copyright 2018 The author(s).Denoising autoencoders (DAEs) have proven useful for unsupervised representation learning, but a thorough theoretical understanding is still lacking of how the input noise influences learning. Here we develop theory for how noise influences learning in DAEs. By focusing on linear DAEs, we are able to derive analytic expressions that exactly describe their learning dynamics. We verify our theoretical predictions with simulations as well as experiments on MNIST and CIFAR-10. The theory illustrates how, when tuned correctly, noise allows DAEs to ignore low variance directions in the inputs while learning to reconstruct them. Furthermore, in a comparison of the learning dynamics of DAEs to standard regularised autoencoders, we show that noise has a similar regularisation effect to weight decay, but with faster training dynamics. We also show that our theoretical predictions approximate learning dynamics on real-world data and qualitatively match observed dynamics in nonlinear DAEs.enDenoising autoencodersDAEsLearning dynamicsLinear DAEsConference PresentationPretorius, A., Kroon, S., & Kamper, H. (2018). Learning dynamics of linear denoising autoencoders. http://hdl.handle.net/10204/10746Pretorius, Arnu, Steve Kroon, and Herman Kamper. "Learning dynamics of linear denoising autoencoders." (2018): http://hdl.handle.net/10204/10746Pretorius A, Kroon S, Kamper H, Learning dynamics of linear denoising autoencoders; 2018. http://hdl.handle.net/10204/10746 .TY - Conference Presentation AU - Pretorius, Arnu AU - Kroon, Steve AU - Kamper, Herman AB - Denoising autoencoders (DAEs) have proven useful for unsupervised representation learning, but a thorough theoretical understanding is still lacking of how the input noise influences learning. Here we develop theory for how noise influences learning in DAEs. By focusing on linear DAEs, we are able to derive analytic expressions that exactly describe their learning dynamics. We verify our theoretical predictions with simulations as well as experiments on MNIST and CIFAR-10. The theory illustrates how, when tuned correctly, noise allows DAEs to ignore low variance directions in the inputs while learning to reconstruct them. Furthermore, in a comparison of the learning dynamics of DAEs to standard regularised autoencoders, we show that noise has a similar regularisation effect to weight decay, but with faster training dynamics. We also show that our theoretical predictions approximate learning dynamics on real-world data and qualitatively match observed dynamics in nonlinear DAEs. DA - 2018-07 DB - ResearchSpace DP - CSIR KW - Denoising autoencoders KW - DAEs KW - Learning dynamics KW - Linear DAEs LK - https://researchspace.csir.co.za PY - 2018 T1 - Learning dynamics of linear denoising autoencoders TI - Learning dynamics of linear denoising autoencoders UR - http://hdl.handle.net/10204/10746 ER -