Skip to content
Global
Japanese
Site Map
Global
Contact Us
Japanese
Site Map
Global
Contact Us

Toshiba Develops Quantum-Inspired Real-Time Black-Box Optimization Technology for Dynamically Changing Environments in Communications, Advertising and Finance

-Realizes real-time optimization by tracking environmental changes, even when the states of users and markets cannot be directly observed-

September 4, 2026
Toshiba Corporation

Overview

Kawasaki, Japan – Toshiba Corporation has developed quantum-inspired real-time black-box optimization technology that realizes real-time control in dynamically changing environments, even when the states of users, markets, and other targets cannot be directly observed, in fields such as communications, digital advertising and finance.
Black-box optimization treats a target system as an unknown, where the internal state of the system cannot be directly observed. A desirable response from the system is obtained through iterative trial-and-error cycles in which actions, such as control inputs, are repeatedly updated and the resulting responses are observed.
Conventional black-box optimization assumes that the state of the target remains unchanged. However, Toshiba’s technology extends black-box optimization to targets that change dynamically, such as user locations and market conditions *1. This approach realizes more efficient control in applications that include allocating users to base stations in wireless communications, tailoring advertising to audience preferences, and dynamic pricing based on supply and demand. The technology leverages Toshiba’s high-speed quantum-inspired combinatorial optimization computer, the Simulated Bifurcation Machine (SBM).
In fields such as wireless communications, digital advertising and finance, there is a need for control systems that achieve desirable outcomes without directly observing the states of users and markets. For example, in wireless communications, the goal is to maximize communication throughput, the volume of data that can be transmitted and received per unit time, by controlling the direction and power of radio signals transmitted from base stations without directly observing the locations of large numbers of users. Black-box optimization is one promising approach to addressing such challenges.
Toshiba has now realized control of targets that were previously difficult to handle by combining an enhanced black-box optimization algorithm with the innovative low-latency and high-performance optimization capability of its proprietary quantum-inspired optimization computer, the SBM, as shown in Figure 1. Specifically, the technology realizes control of practical yet previously challenging systems characterized by: a. states such as user locations and preferences that cannot be directly observed; b. user, market, or other states that change continuously; and c. the need to select the optimal action from a vast number of combinations of control patterns. In an evaluation assuming real-time control of wireless base stations serving large numbers of moving users, Toshiba confirmed that the technology improved the average total communication throughput by 21% compared with conventional methods (Figure 2 and video). The technology also updates control conditions with a sampling cycle that averages 38.5 milliseconds, enabling real-time control.
These results were published online in the international scientific journal Nature Communications on September 3, 2026 in the UK*1.

Figure 1. Toshiba’s Quantum-Inspired Real-Time Black-Box Optimization Technology

Development background

Black-box optimization is a technique that treats a target system as an unknown quantity, without requiring information about its internal state, and that achieves a desirable response based on information about actions applied to the system and the resulting observed responses.
In industrial systems that provide services to users, the service provider often cannot directly observe information about users’ states, actions, or preferences. Black-box optimization is therefore a useful approach for such systems. However, when a control system must select a combinatorial action consisting of multiple elementary actions from among the enormous number of possible actions (combinatorial explosion), while considering interactions among the elementary actions, such as cooperative effects or trade-offs, selecting the optimal combinatorial action becomes a difficult problem. On top of this, conventional black-box optimization technologies capable of handling combinatorial actions assume that the internal state of the target does not change. As a result, they cannot handle targets whose internal states change dynamically over time.
These factors have made it a significant challenge to extend black-box optimization technology to handle combinatorial actions and track dynamic changes.

Features of the technology

Toshiba addressed the challenge by developing quantum-inspired real-time black-box optimization technology that extends conventional black-box optimization and realizes control using combinatorial actions even when the target’s internal state cannot be directly observed and changes dynamically over time. The technology combines a new black-box optimization algorithm and an accelerated trial-and-error cycle using the SBM, enabling black-box optimization to track dynamic changes and handle combinatorial actions.

1. The Black-Box Optimization Algorithm

In the black-box optimization, a surrogate model that approximates the target system is constructed from historical learning data comprising previously executed actions and the responses observed at the time.
To enable the technology to track dynamic changes, Toshiba introduced three mechanisms: a. a sliding-window mechanism that progressively discards older learning data; b. a forgetting mechanism that slightly attenuates the influence of the previous surrogate model in each trial-and-error cycle, allowing the model to better adapt to the latest learning data; and c. an exploration incentive that modifies the surrogate model to encourage exploration (Figure 1). Together, these mechanisms achieve and maintain stable, desirable responses while continuously adapting to dynamically changing internal states.

2. Acceleration of the Trial-and-Error Cycle Using SBM

In a system employing the technology, the next action is determined as a solution to an optimization problem defined by the surrogate model. Finding the best action using a surrogate model is challenging because it requires searching through a vast number of possible combinations while accounting for interactions among elementary actions. Toshiba’s SBM technologies rapidly find solutions to large-scale, complex combinatorial optimization problems*2*3*4. Toshiba has also developed various industrial systems built on the SBM, including applications in automotive systems*5, communications*6 and finance*7. The black-box optimization technology uses an embedded SBM characterized by short computation times and low communication latency with other system modules. This reduces the execution time of the entire trial-and-error cycle and delivers real-time responsiveness.
Toshiba applied the technology to the real-time control of wireless base stations serving large numbers of moving users (Figure 2 and video). In the evaluation problem, each of 37 base stations selected one of nine predefined radio transmission patterns (beam patterns) with different directions and power levels. This results in an enormous search space of 9^37 possible combinations (approximately 2.03 × 10^35). Without directly knowing the locations of the moving users, the control system optimizes the directions and power levels of radio waves transmitted from multiple base stations while accounting for radio interference, with the objective of maximizing the total communication throughput. The evaluation confirmed that the technology achieves both improved communication throughput and real-time control.

Figure 2. Comparison of the Proposed Technology and Conventional Technology in Wireless Base Station Control

Video: Demonstration of Wireless Base Station Control

Future developments

Toshiba will advance the application of the technology to wireless communication systems, advertising recommendation systems, dynamic pricing systems, and other fields. Toshiba also plans to provide application implementation examples incorporating the technology with the embedded SBM. Through these efforts, Toshiba will promote the implementation of quantum-inspired optimization technologies in society.