Differentiable Invariants

Abstract. Invariants that incrementally maintain the value of expresder assignments to their variables are a natural abstraction to build high-level local search algorithms. But their functionalities are not sufficient to allow arbitrary expressions as constraints or objective functions as in constraint programming. Differentiable invariants bridge this expressiveness gap. A differentiable invariant maintains the value of an expression and its variable gradients, it supports differentiation to evaluate the effect of local moves. The benefits of differentiable invariants are illustrated on a number of applications which feature complex, possibly reified, expressions and whose models are essentially similar to their CP counterparts. Experimental results demonstrate their practicability.

• Type: Conference
• Year: 2006
• Where: CP
• Authors: Pascal Van Hentenryck, Laurent Michel
• Keywords: Arbitrary Expressions, Artificial Intelligence, CP 2006, Expresder Assignments, Expressiveness Gap