Iteration-wise Adjoining
نویسندگان
چکیده
The reverse or adjoint mode of Automatic Diierentiation is a software engineering technique that permits eecient computation of gradients. However, this technique requires a lot of temporary memory storage. In this paper, we present iteration-wise adjoining, a reenement that reduces memory consumption in the case of parallel loops, and we give a proof of its correctness, based on properties of the data-dependence graph of adjoint programs and parallel loops. Iteration-wise adjoining is particularly suitable for assembly loops, that dominate in mesh-based computations. Application is done on the kernel of a real-size Navier-Stokes solver.
منابع مشابه
Adjoining Idempotents to Semigroups
Our paper is inspired by the paper [3] by M. Nathanson, in which he considers iteration of the process of adjoining identities or zeros to semigroups and characterizes the resulting semigroup in the case when one starts with the trivial semigroup. We consider also the third type of adjoining, namely, we adjoin idempotents “that behave like some already existing elements.” We investigate the pro...
متن کاملStochastic Tree-Adjoining Grammars
A B S T R A C T The notion of stochastic lexicalized tree-adjoining grammar (SLTAG) is defined and basic algorithms for SLTAG are designed. The parameters of a SLTAG correspond to the probability of combining two structures each one associated with a word. The characteristics of SLTAG are unique and novel since it is lexically sensitive (as N-gram models or Hidden Markov Models) and yet hierarc...
متن کاملCoordinate-wise Power Method
In this paper, we propose a coordinate-wise version of the power method from an optimization viewpoint. The vanilla power method simultaneously updates all the coordinates of the iterate, which is essential for its convergence analysis. However, different coordinates converge to the optimal value at different speeds. Our proposed algorithm, which we call coordinate-wise power method, is able to...
متن کاملBlock-wise ADMM with a Relaxation Factor for Multiple-block Convex Programming
It has been shown that the alternating direction method of multipliers (ADMM) is not necessarily convergent when it is directly extended to a multiple-block linearly constrained convex minimization model with an objective function in the sum of more than two functions without coupled variables. Most recently, we proposed the block-wise ADMM, which was obtained by regrouping the variables and fu...
متن کاملPower Iteration Clustering
We present a simple and scalable graph clustering method called power iteration clustering (PIC). PIC finds a very low-dimensional embedding of a dataset using truncated power iteration on a normalized pair-wise similarity matrix of the data. This embedding turns out to be an effective cluster indicator, consistently outperforming widely used spectral methods such as NCut on real datasets. PIC ...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2007