Speculative Symbolic Graph Execution of Imperative Deep Learning Programs

Abstract

The rapid evolution of deep neural networks is demanding deep learning (DL) frameworks not only to satisfy the requirement of quickly executing large computations, but also to support straightforward programming models for quickly implementing and experimenting with complex network structures. However, existing frameworks fail to excel in both departments simultaneously, leading to diverged efforts for optimizing performance and improving usability. This paper presents JANUS, a system that combines the advantages from both sides by transparently converting an imperative DL program written in Python, a de-facto scripting language for DL, into an efficiently executable symbolic dataflow graph. JANUS can convert various dynamic features of Python, including dynamic control flow, dynamic types, and impure functions, into elements of a symbolic dataflow graph. Our experiments show that JANUS can achieve fast DL training by exploiting the techniques imposed by symbolic graph-based DL frameworks, while maintaining the simple and flexible programmability of imperative DL frameworks at the same time.

A Reverse Modification Method for Binary Code and Data

“…Symbolic executions [ 5 , 6 ] are technologies that analyze control flow or extract path constraints, which are used to find all feasible program paths and detect information leakage [ 7 ]. Jeong et al used fast and flexible deep learning via the symbolic graph execution of imperative programs [ 8 ]; program paths can be predicted with progressive symbolic execution [ 9 ]. However, symbolic execution technologies are not scalable for large programs with unbounded loop iterations [ 10 , 11 ].…”