Abstract
Today’s mobile processors generally have multiple cores and sufficient hardware resources to support AI-enabled software operation. However, very few AI applications make full use of the computing performance of mobile multiprocessors. This is because the typical software development is sequential, and the degree of parallelism of the program is very low. In the increasingly complex AI-driven and software development projects with natural human–computer interaction, this will undoubtedly cause a waste of mobile computing resources that are originally limited. This paper proposes an intelligent system software framework, CellS, to improve smart software development on multicore mobile processor systems. This software framework mimics the cell system. In this framework, each cell can autonomously aware changes in the environment (input) and reaction (output) and may change the behavior of other cells. Smart software can be regarded as a large number of cells interacting with each other. Software developed based on the CellS framework has a high degree of scalability and flexibility and can more fully use multicore computing resources to achieve higher computing efficiency.
Highlights
It is not easy for engineers to write parallel programming directly, because, when engineers face low-level parallel processing libraries directly, they cannot focus on the design of business logic
When engineers want to modify the current program into a parallel program, because human are accustomed to linear thinking, engineers need to spend a lot of effort to find the parallel parts of the program and refactor the logic of the program on a large scale
We developed a framework called CellS, inspired by the cell theory
Summary
It is not easy for engineers to write parallel programming directly, because, when engineers face low-level parallel processing libraries directly, they cannot focus on the design of business logic. On the other hand, according to Amdahl’s law, if there is no part in the program that can be processed in parallel, with more cores, the application itself will not get any efficiency gains. Taking AlphaGo as an example, its three characteristics are (1) autonomously playing chess without human interference; (2) using value network and policy network, two deep learning techniques; and (3) using Monte Carlo tree search to judge the difference of the chessboard situation and infer the current move Based on these three characteristics, this study establishes the CellS software architecture to integrate artificial intelligence applications. The BDI agent is a software model developed for building an intelligent agent that provides a mechanism for separating the selection of a plan from the execution of a plan.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call