A recent paradigm views deep neural networks as discretizations of certain controlled ordinary differential equations, sometimes called neural ordinary differential equations. We make use of this perspective to link expressiveness of deep networks to the notion of controllability of dynamical systems. Using this connection, we study an expressiveness property that we call universal interpolation and show that it is generic in a certain sense. The universal interpolation property is slightly weaker than universal approximation and disentangles supervised learning on finite training sets from generalization properties. We also show that universal interpolation holds for certain deep neural networks even if large numbers of parameters are left untrained and are instead chosen randomly. This lends theoretical support to the observation that training with random initialization can be successful even when most parameters are largely unchanged through the training. Our results also explore what a minimal amount of trainable parameters in neural ordinary differential equations could be without giving up on expressiveness.
We develop a comprehensive mathematical framework for polynomial jump diffusions in a semimartingale context, which nest affine jump diffusions and have broad applications in finance. We show that the polynomial property is preserved under polynomial transformations and Lévy time change. We present a generic method for option pricing based on moment expansions. As an application, we introduce a large class of novel financial asset pricing models with excess log returns that are conditional Lévy based on polynomial jump diffusions.
1 January 2020 in SIAM Journal on Financial Mathematics
We propose a multifactor polynomial framework to model and hedge long-term electricity contracts with delivery period. This framework has several advantages: the computation of forwards, risk premi...
We characterize the minimal time horizon over which any equity market with stocks and sufficient intrinsic volatility admits relative arbitrage with respect to the market portfolio. If , the minimal time horizon can be computed explicitly, its value being zero if and if . If , the minimal time horizon can be characterized via the arrival time function of a geometric flow of the unit simplex in that we call the minimum curvature flow.