AutoML(Automated Machine Learning) is a technology that automates various processes in machine learning tasks. Many AutoML tools can automatically build machine learning models (AI models) by specifying datasets and its objective variable to be predicted by machine learning. AutoML is attracting attention as a technology to democratize AI.
Most of the existing AutoML tools have the following issues.
- It takes time/needs huge computer resources to build an AI model.
- Because during the automation process, it takes an exploratory approach to perform and evaluate all combinations of processes in building an AI model.
- It's unclear how the AI model was build.
- Because it provides just a built AI model and does not provide the program that was used to build it.
Fujitsu AutoML Automated AI Model Development
(Automated Machine Learning) - YouTube
We have developed Fujitsu AutoML, Fujitsu’s unique AutoML technology that quickly generates highly accurate AI models with machine learning programs to enable users to conduct trial-and-error approach easily. There are two main features:
- Technology to learn existing programs that can build highly accurate AI models and automatically generate programs by predicting the most plausible preprocesses and AI models.
- Technology to generate a program to build an AI model with the explanation of processes.
The features bring users the following values:
- High speed: It can generate an AI model quickly by evaluating not all the combinations of the machine learning pipelines but only the most plausible machine learning pipelines.
- Transparency: It is easy to understand how the AI model is build by checking the generated machine learning program containing the explanation.
- High Accuracy: It can generate highly accurate AI models by the prediction based on the past knowledge of the programs that were used to build highly accurate AI models.
SapientML, a technology used as a basis of Fujitsu AutoML, was accepted by the International Conference on Software Engineering (ICSE 2022), one of the premier conference in Software Engineering.
- Title: SapientML: Synthesizing Machine Learning Pipelines by Learning from Human-Written Solutions
- Conference: The 44th International Conference on Software Engineering (ICSE 2022)
The specifiation of Fujitsu AutoML (Supported data type/machine learning task)
Supported data types are:
- Table data (Structured data)
- Number, String, DataTime
Supported machine learning tasks are:
- Classification (Predict a category of each item)
- e.g. Sales category prediction, Product quality prediction, Good customer prediction
- Regression (Predict price, how many people, etc.)
- e.g. Demand prediction, Sales price prediction
Do you have some trouble in situations like:
- You are adopting AIs in your business but could not involve data scientists sufficiently.
- For you have some data and insight to build an AI model using the data, and you want to check the feasibility quickly by yourself before hiring data scientists
- You are intending to ask data scientists about AI model development more smoothly.
Fujitsu AutoML helps a lot in:
- Building an AI model in a short time from just datasets and machine learning task specifications provided
- Quickly evaluating the accuracy of the AI model by varying datasets
- Conducting a trial-and-error approach in a machine learning task always being aware of the rationale behind the built AI model
Comparison with other AutoML tools
- Other AutoML tools take longer time/need huge computer resources and it’s costly.
- It’s hard to understand how the output AI model was build.
- The AI model build by other AutoML tools can be used on its platform and tied to it. (When you use it once, you can not escape from it…)
How to use
We have prepared a free-of-charge app where you can try out Fujitsu AutoML. We hope that this will help you to further expand your image of utilizing Fujitsu AutoML.
In addition, we are also waiting for inquiries regarding business use of Fujitsu AutoML through this app. Please note that we have limited some core functions.