Efficient Data Storage in Large Nanoarrays

We explore the storage of data in very large crossbars with dimensions measured in nanometers (nanoarrays) when h-hot addressing is used to bridge the nano/micro gap. In h-hot addressing h of b micro-level wires are used to address a single nanowire. Proposed nanotechnologies allow subarrays of 1s (stores) or 0s (restores) to be written. When stores and restores are used, we show exponential reductions in programming time for prototypical problems over stores alone. Under both operations, it is NP-hard to find optimal array programs. Under stores alone it is NP-hard to find good approximations to this problem, a question that is open when restores are allowed. Because of the difficulty of programming multiple rows at once, we explore the programming of single rows under h-hot addressing. We also identify conditions under which good approximations to these problems exist.     

Strong Stability Preserving Properties of Runge–Kutta Time Discretization Methods for Linear Constant Coefficient Operators

Strong stability preserving (SSP) high order Runge–Kutta time discretizations were developed for use with semi-discrete method of lines approximations of hyperbolic partial differential equations, and have proven useful in many other applications. These high order time discretization methods preserve the strong stability properties of first order explicit Euler time stepping. In this paper we analyze the SSP properties of Runge Kutta methods for the ordinary differential equation u _t =Lu where L is a linear operator. We present optimal SSP Runge–Kutta methods as well as a bound on the optimal timestep restriction. Furthermore, we extend the class of SSP Runge–Kutta methods for linear operators to include the case of time dependent boundary conditions, or a time dependent forcing term.     

Apportioned margin approach for cost sensitive large margin classifiers

We consider the problem of cost sensitive multiclass classification, where we would like to increase the sensitivity of an important class at the expense of a less important one. We adopt an apportioned margin framework to address this problem, which enables an efficient margin shift between classes that share the same boundary. The decision boundary between all pairs of classes divides the margin between them in accordance with a given prioritization vector, which yields a tighter error bound for the important classes while also reducing the overall out-of-sample error. In addition to demonstrating an efficient implementation of our framework, we derive generalization bounds, demonstrate Fisher consistency, adapt the framework to Mercer’s kernel and to neural networks, and report promising empirical results on all accounts.