Further Possibilities of Optimization Technology with the Advent of AI

Optimization technology has expanded its scope of use with the advancement of theory and IT in the past few decades. Systems can now automatically create viewpoints of judgment using AI technology, in particular, deep learning*, which have attracted attention in recent years, to calculate more appropriate and advanced solutions than ever before. We are required to meet the need for technology that is capable of automatically providing solutions closer to human judgment and sensibility. To meet such needs, we have already started working on combining optimization technology and deep learning. This article outlines the current trends and describes the future of optimization.

Author profile:
Shinji Yamane
Principal Consultant
Business Analytics Consulting Group
Fujitsu Research Institute

Yamane joined Fujitsu Limited in 1990 and was temporarily assigned to Fujitsu Research Institute, where he has been engaged in modeling, system development and consulting business related to optimization, including delivery strategies using genetic algorithms and transportation route improvement using mathematical models. He has also been engaged in policy support research projects related to promoting efficient distribution systems and environments.

Optimization Technology and Problem Solution Approach

Optimization technology is a technology for maximizing or minimizing objective functions under certain constraints. For example, in transportation planning in logistics, optimization technology is used to create the shortest transportation distance and time for a limited number of trucks within a limited number of days. People make decisions in the same way of thinking as systems; however, systems can calculate solutions by using vast amounts of data that cannot be handled by humans and computation beyond humans' ability.

In recent years, the necessity to solve large-scale optimization problems has increased in the real world. Optimization technology is used for improving production and logistics as well as for formulating disaster prevention plans to prepare against earthquakes, typhoons and other natural disasters. With the development of technologies, such as mathematical programming, genetic algorithms and metaheuristics, methods combining multiple solutions have been devised and put to practical use.

What is noteworthy recently is a change from a method-centered approach to a data-based problem-solving approach, which first analyzes real data and finds problems and bottlenecks and then designs solutions. Under this approach, Fujitsu consultants have comprehensively ascertained conditions, understood bottlenecks and then extracted conditions to be incorporated into optimization technology to derive solutions while prioritizing them.

Using AI to Address Challenges in Applying Optimization Technology

In 2016, since AlphaGo, which used deep learning technology, defeated a professional Go player, people have become more aware of the potential of using AI technology. AI is a technology that performs human intellectual processing with computers. Optimization technology is, as described above, an alternative to human decision-making and an element of AI. However, currently, machines are required to acquire deep learning capabilities to autonomously learn algorithms of knowledge from data.

Conventional optimization technology requires humans to decide conditions and objective items so that the computer can calculate a solution, and make adjustments to the solution afterwards. As it is based on this PDCA cycle, applying conventional technology requires substantial knowledge and experience as well as time. In addition, because practical operations are too complicated to provide perfect conditions to the computer, the derived solution needs to be adjusted before putting it into practice. However, given the increasing speed of business and complex factors in companies' decision making, innovation is now required to respond to changes more quickly. We cannot meet such needs any more with conventional technology.

(Figure 1) Applying Optimization Technology Using AI

To address the above issues, we have started reviewing the utilization process of conventional optimization technology to automatically incorporate decisions made by people when making adjustments and calculate more advanced solutions. The first step is to use AI for the data-based problem-solving approach. It is an attempt to automatically extract various tendencies and features from data and automatically detect bottlenecks to incorporate them into optimization conditions for problem solving. The second step is to then take advantage of the knowledge data gained through evaluation and corrections of past solutions calculated by optimization technology as inputs to derive the next solution. If this cycle works successfully, decision criteria that so far have not been incorporated in data can be utilized to create more accurate solutions for practical operations.

To realize this cycle, we are working to develop a faster and more accurate solution for practical use by combining optimization technology with conventional statistics/prediction technology and other state-of-the-art technologies such as deep learning. Problems in practical operations are complicated, and AI varies depending on the area of utilization. Since there are numerous things to be considered depending on the problem and purpose, it is necessary to specify a target area, examine individual implementation methods and create a model for calculating practical solutions.

Analysis and Formulating a Vision--AI Providing Advanced Strategic Support

The current AI technology cannot completely replace human decisions, and steady efforts are required to realize the above-mentioned cycle. However, with the advent of movements of big data and IoT after 2010 and technologies that support and even exceed human decision-making, there is no doubt that intellectual processing will become more and more sophisticated going forward. In addition to the sophistication of conventional IT utilization methods, roles, which consultants and engineers have played in analyzing and constructing systems, will greatly change into those of supporting customers' vision development and business transformation from a more strategic point of view.

*:Deep learning: A neural network-based learning technology. A neural network is a model simulating the mechanism of the neural circuits of the human brain. A major difference from conventional technology is that the computer automatically extracts features from data as the basis of learning.