Normalizing flows leverage the Change of Variables Formula (CVF) to define flexible density models. Yet, the requirement of smooth transformations (diffeomorphisms) in the CVF poses a significant challenge in the construction of these models. To enlarge the design space of flows, we introduce $\mathcal{L}$-diffeomorphisms as generalized transformations which may violate these requirements on zero Lebesgue-measure sets. This relaxation allows e.g. the use of non-smooth activation functions such as ReLU. Finally, we apply the obtained results to planar, radial, and contractive residual flows.
Towards Generalization-Oriented Models for Vehicle Routing Problems with Mixture-of-Experts
State Key Laboratory of Autonomous Intelligent Unmanned Systems, Beijing Institute of Technology(自主智能无人系统国家重点实验室,北京理工大学);School of AI, Beijing Institute of Technology(北京理工大学人工智能学院)
Predicting Collision Cross Sections with GRACE: Geometric Residual Adduct Conditioning via Early-fusion
IBM Research(IBM研究院);Center for Computational Life Sciences, Cleveland Clinic(克利夫兰诊所计算生命科学中心);Department of Chemistry, Michigan State University(密歇根州立大学化学系)
Vanderbilt University(范德堡大学);University of Notre Dame(圣母大学);Department of Psychology and Human Development(心理与人类发展系);Department of Applied and Computational Mathematics & Statistics(应用与计算数学及统计学系);Nashville, TN(纳什维尔,田纳西州);Notre Dame, IN(圣母大学,印第安纳州)
A Geometry-Aware Residual Correction of Hagan's SABR Implied Volatility Formula
Abu Dhabi Investment Authority (ADIA)(阿布扎比投资局);SAFIR, Sorbonne University Abu Dhabi (SUAD)(萨菲尔,阿布扎比索邦大学);Sorbonne Université, LPSM, Paris(巴黎索邦大学);Institut Universitaire de France(法国高等研究院);ADIA Lab(阿布扎比投资局实验室)