Wide Learning

(This site was translated by LLM)

About This Site

  • Wide Learning is an XAI (Explainable AI) developed by Fujitsu Research.
  • This site provides information on the latest activities related to Wide Learning and the Trial Tool (hands-on experience tool).
  • Related research: Fujitsu AutoML, Fujitsu Causal AI

Contents

Wide Learning NEWS

Four Features of Wide Learning

Feature 1. Able to explain the reasons for its judgments

Wide Learning derives its answers using hypotheses that are interpretable by humans, so the intermediate computation process and the final judgment result are logically and objectively easy to understand — making it explainable.

Because of this characteristic, Wide Learning is positioned as one of the XAI (Explainable AI) approaches, standing in contrast to black-box AI such as Deep Learning and its more advanced form, generative AI (e.g., ChatGPT).

*1: A "black box" refers to a device or tool whose internal working principles cannot be observed from the outside, so only its outputs can be used.

Feature 2. Able to discover all important hypotheses without omission

So why is Wide Learning explainable?

The reason is that, within the hypothesis space for the input data, it exhaustively verifies every hypothesis and discovers, without omission, the important hypotheses (which we call "knowledge chunks").

Furthermore, if there is not a single important hypothesis within the hypothesis space, Wide Learning can rigorously prove that none exists.

Here, the hypothesis space that Wide Learning handles consists of every possible combination of data items in the input data.

Verifying every combination of data items requires efficiently computing an astronomical number of combinations.

Through years of research in discovery science (*2), Fujitsu Laboratories has developed ultra-high-speed combinatorial computation technology.

*2: Discovery science is a research field proposed in the 1990s by Professor Setsuo Arikawa of Kyushu University (now Professor Emeritus). It develops both theoretical and practical research aimed at using computers to discover hypotheses and knowledge.

Feature 3. Capable of highly accurate judgments even with small amounts of data

Even without a large amount of training data, Wide Learning can build a sufficiently large hypothesis space from the small amount of data at hand and exhaustively discover important hypotheses.

For example, suppose you want to analyze the causes of defects on a factory production line. In general, defects rarely occur, so you run into the problem that "it is difficult to collect large amounts of data on defect occurrences."

However, with only a few dozen to a few hundred data points, Wide Learning can discover important hypotheses that apply specifically to defective items, and begin analyzing the causes of the defects.

Feature 4. Able to automatically present action plans

The last feature is that, based on Wide Learning's learning results, actual actions can be carried out.

Wide Learning exhaustively verifies all hypotheses within the hypothesis space.

Using this information, for example in digital marketing, Wide Learning can compute the difference between "hypotheses with low purchase rates" and "hypotheses with high purchase rates," and present the combination with the most pronounced difference (i.e., the one with the largest increase in purchase rate) as an "action plan."

In experiments conducted by Fujitsu's marketing department, it was confirmed that the action plans proposed by Wide Learning achieved both higher customer coverage and higher average expected purchase rates than the action plans devised by marketing experts.

Trial Tool

  • Fujitsu Laboratories provides "Trial Tool," an environment where you can try out Wide Learning hands-on (*).
  • (*)Please note that some core features are restricted.
  • Please see the link below for details.
  • Wide Learning Trial Tool

© 2020 - 2026 FUJITSU Limited