2020/07/11 by Hedda Cohen Indelman, Indelman, Hedda Cohen, Tamir Hazan +1 · 1 citation
Computer Science · #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Data Classification #Music and Audio Processing
paper · pdf · doi:10.48550/arxiv.2007.05724
openalex publication_date 2020/07/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Direct loss minimization is a popular approach for learning predictors over\nstructured label spaces. This approach is computationally appealing as it\nreplaces integration with optimization and allows to propagate gradients in a\ndeep net using loss-perturbed prediction. Recently, this technique was extended\nto generative models, while introducing a randomized predictor that samples a\nstructure from a randomly perturbed score function. In this work, we learn the\nvariance of these randomized structured predictors and show that it balances\nbetter between the learned score function and the randomized noise in\nstructured prediction. We demonstrate empirically the effectiveness of learning\nthe balance between the signal and the random noise in structured discrete\nspaces.\n