コンテンツにスキップ

Home

Data-Driven Decision Making
Rapidly and comprehensively calculates causal relationships under all conditions from massive datasets, discovers individually useful causal relationships typically overlooked, and recommends the most effective, non-detrimental measures based on extensive information. Creates entirely new service and business models, revitalizing the economy.

Challenges for System Engineers and Consultants

Extracting causal relationships under specific conditions is difficult, not just overall data relationships. Furthermore, conventional measure recommendation technologies inadequately consider adverse effects, making it hard to arrive at optimal proposals. Additionally, using a single dataset limits the scope of proposed measures.

With Fujitsu's Causal AI

Equipped with the high-speed causal exploration technology LayeredLiNGAM, achieving speeds over 1,000 times faster than competing technologies. This enables efficient discovery of causal relationships under specific conditions and unknown causal relationships, allowing for optimal policy proposals that consider adverse effects. It also boasts a strong track record of paper acceptance at top international AI conferences.

The benefits of Causal AI

  1. Efficiently identifies causal relationships that manifest under specific conditions and unknown causal relationships that lead to new insights
  2. Proposes optimal measures based on multiple causal relationships, considering feasibility and side effects
  3. High-speed causal exploration significantly reduces computational load when extracting causal relationships from massive datasets
  4. It continuously corrects the divergence between real-world data and causal models during operation, maintaining an optimal state at all times
  5. By combining diverse datasets, it enables comprehensive and detailed countermeasure proposals, providing deep insights that go beyond limited suggestions

Technology Summary

Target Industry/Users

  • Systems Engineers
  • Consultants

Challenges in Target Industry and Operations

  • We want to extract causal relationships that appear only under specific conditions, not just those within the entire dataset
  • We want to discover unknown causal relationships that lead to new insights from the large number of known causal relationships extracted
  • Conventional policy recommendation technologies* struggle to account for adverse effects and cannot consider multiple causal relationships, potentially leading to suboptimal policy proposals
  • If only a single dataset is used for causal relationship estimation, the volume and bias of the data limit the scope of actionable recommendations

Technical Challenges

  • 1.The sheer volume of data combinations makes computation difficult.
  • 2.Difficulty in proposing measures to achieve objectives (due to feasibility/side effects)
  • 3.Difficulty in correcting discrepancies between causal models and reality during implementation

Solutions

  • 1 and 3 are explained below.
    • technology 1
  • 2 is explained as follows.
    • technology 2

The benefits of Causal AI（Detailed version）

  • Efficiently identify causal relationships that manifest under specific conditions, as well as unknown causal relationships that lead to new insights.
  • Proposing optimal policies based on multiple causal relationships while considering feasibility and side effects
  • Significantly reduces computational load in extracting causal relationships from massive datasets through high-speed causal exploration
  • Continuously corrects discrepancies between real-world data and causal models during operation to maintain optimal performance
  • Enables comprehensive and detailed action proposals by combining diverse datasets, providing deep insights beyond limited suggestions

Fujitsu's Technological Advantage

  • High-speed causal exploration is achieved using the LayeredLiNGAM method, which is over 1,000 times faster than OSS or other companies' technologies.
  • We have not seen any competing technologies capable of proposing measures that do not cause adverse effects.
  • As noted throughout sections 1-3, this technology has been accepted as papers at top international AI conferences (PKDD 2024, ECML 2024, AAAI 2025, WSC 2024, ICML 2024, ICML 2025, NeurIPS 2025).

Use Cases

  • Solution
    • Unmatched Scalability
    • Optimal Action Recommendation Technology
  • Users
    • Medical Institutions
    • Manufacturers
    • Large Corporations

Case studies

Technical Trial

Documents

Document Title Description
Technical Document Kozuchi Causal Discovery
Technical Document Kozuchi Causal AI
White Paper Fujitsu Causal Knowledge Graph whitepaper