Free online consultation available. Please feel free to contact us.

Fine Tuning

  • Glossary

"Fine Tuning"

This glossary explains various keywords that will help you understand the mindset necessary for data utilization and successful DX.
This time, let's think about how learning is done in machine learning, which is currently attracting a lot of attention in the use of IT.

What is fine-tuning?

Fine-tuning is a machine learning technique that involves additional training on an existing trained model to create a trained model with adjusted capabilities.
It can be used as a method to reduce the cost of training compared to training from scratch, and is sometimes used as a means of reusing existing trained models to adjust them to suit specific applications.

I want to use machine learning, but preparing a trained model is difficult.

"AI" is a hot topic in society these days, but in most cases, "AI" refers to "machine learning." Machine learning is a means of generating value from data, and how to utilize machine learning is an important theme in data utilization efforts.

I want to use machine learning, but preparing a trained model is difficult.

To utilize machine learning in business, you need a "trained model" that can perform the tasks required for your application.

For example (though this is a simplified example), if you want to automatically determine whether an object in an image is an "apple" or an "orange" from image data,

Image data (input) ⇒ Trained model ⇒ "It's an apple" or "It's an orange" (output)

It is actually difficult to determine where to obtain the input data and how to utilize the judgment results, but in order to make such use possible, it is necessary to prepare a ``trained model'' with sufficient performance.

However, preparing a trained model can be time-consuming and costly, and it can be difficult to fully commit to it, as it can require a lot of data preparation, highly skilled engineers, and high-spec hardware.

To prepare a "trained model":

  • Prepare "input data + teacher data" as training data
    (Image data is prepared and people label it as "this is an orange" or "this is an apple," so generally a lot of data is required.)
  • The data is used to train a machine learning engine, resulting in a "trained model."
    (Highly skilled personnel are required to make the computer do a lot of calculations and learn.)

Training a model from scratch to suit your needs is likely to result in a trained model with ideal performance, but in reality this can be difficult.

Reusing existing trained models

In addition, there are cases where it is "practically impossible" to prepare data and conduct learning on your own.

At the time of writing, generative AI is a hot topic, but the "LLM (Large Scale Language Model)," the trained model that powers ChatGPT and other AI models, is created using a tremendous amount of training data and undergoing a tremendous amount of calculation. In many cases, it would be unrealistic for each company in Japan to prepare a "tremendous amount of training data," perform massive calculations, and individually prepare its own dedicated LLM.

Therefore, "reusing existing trained models in some way" is an important point in utilizing machine learning.

Additional learning after the fact

Finetuning is one way to reuse such existing trained models.

The meaning of the term "fine-tuning" is somewhat unclear, but for the purposes of this explanation, we will generally refer to "additional training on an existing trained model" to adjust performance.

There are two types of machine learning: one that uses the entirety of a large amount of training data to "learn all at once," and one that uses "gradual learning" by providing data little by little.

If it is a type of machine learning that learns incrementally, it is not an algorithm that ends learning. In that case, you can provide "additional training data" to the trained model to "perform additional learning." For example, neural networks used in deep learning are basically of the latter type.

Fine-tuning example

Let's take the previous example, "automatically distinguishing between apples and oranges," as an example. Let's assume that there is already an existing trained model, and that model is "sufficiently capable of distinguishing between apples and oranges." We will reuse this model so that it can also "distinguish between green apples."

  • A pre-trained model that can automatically distinguish between apples and oranges is available
  • Prepare additional training data by labeling the "image data of green apples" with "This is a green apple."
  • Perform additional training on the trained model using the above training data.

As a result, the capabilities of existing trained models can be fine-tuned, potentially enabling you to:

Image data (input) ⇒ Fine-tuned model ⇒ "It's an apple," "It's a mandarin orange," "It's a green apple" (output)

In reality, rather than simply providing additional data and training, different technical techniques are used than when training from scratch (such as not retraining the entire neural network of an already trained model).

However, in terms of usage, fine-tuning is the process of "producing a 'trained model' with adjusted performance" by providing "additional training data" to an "existing model."

Fine tuning is not a panacea

Fine tuning can be used to give existing trained models new capabilities (e.g., adding the ability to identify green apples), repurpose them for different purposes (e.g., repurposing the ability to identify apples to identify vegetables), or specialize general capabilities for specific purposes (e.g., improving accuracy by specializing in apples from a specific region).

The idea of additional training may seem intuitively good and should always be used, but unfortunately, fine-tuning sometimes doesn't "work as expected."

Sometimes it's better to start from scratch

We would like to think that using something that is already "smart" will have a positive effect rather than learning it from scratch, but in reality, there are times when learning it from scratch is better.

This is a metaphor, but what would you think if someone told you to go from Tokyo to Niigata and then to Kofu in Yamanashi Prefecture? You would probably think it would be quicker to go directly from Tokyo to Kofu without stopping in Niigata (going all the way to Niigata would be a complete waste of time). That is possible.

Existing "learning" may deteriorate

In the example above, having the Pokémon learn "green apples" may result in a decrease in its ability to judge "oranges." You may not be able to acquire the abilities you want, and the abilities you already have may be weakened through learning, making it difficult to conveniently change the abilities you want to adjust.

You may not get the performance you want

Even if you try additional training, you may not be able to successfully acquire the desired capabilities. For example, at the time of writing, it is said that it is not uncommon for companies to try to create their own generative AI by fine-tuning it by additionally training their own data into an LLM (large-scale language model), but still not achieve the desired performance.

First, we need to teach them

In either case, additional training is required. Data and hardware must be prepared, and highly skilled engineers who understand machine learning are also required. The fact that "training" is necessary means that it is not an easy or ideal method for everyone.

Fine Tuning and Other Options

Fine tuning is a useful technique, but unfortunately it is not a trump card, and you should be aware that other options are available and use them accordingly.

  • Creating a trained model from scratch
    (This may end up being less time-consuming and more efficient.)
  • Transfer Learning
  • In-Context Learning
    • RAG（Retrieval-Augmented Generation）

In addition, performance can sometimes be improved through trial and error, such as by improving the data provided or by devising ways to utilize pre-trained models (see the machine learning article).

"Connecting" technology that is useful in situations where trial and error is required

When using machine learning, trial and error is often necessary, not only in collecting the necessary data, but also in determining what data to use, how to train it, and even how to use the trained model in business.

If every trial and error like this, for example, collecting data and trying to combine it with business operations, takes time and effort, it will not be possible to work efficiently.

In fact, it is not uncommon for attempts to utilize machine learning to take a long time to prepare, and even if a PoC is carried out and some results are achieved, it seems that the technology is not yet deployed in actual business operations.

To solve this problem, it is necessary to be able to smoothly and efficiently carry out the various types of collaborative processing that will be required through trial and error.

Please utilize "connecting" technology

There are ways to efficiently develop these various "integration processes" using only a GUI. These are "connecting" technologies known as "EAI," "ETL," and "iPaaS," such as "DataSpider" and "HULFT Square." By utilizing these, new and old systems can be integrated smoothly and efficiently.

Can be used with GUI only

Unlike regular programming, there is no need to write code. By placing and configuring icons on the GUI, you can achieve integration with a wide variety of systems, data, and cloud services.

Being able to develop using a GUI is also an advantage

​No-code development using only a GUI may seem like a simple compromise compared to full-scale programming. However, being able to develop using only a GUI allows on-site personnel to proactively work on cloud integration themselves. On-site personnel are the ones who know the business best.

Full-scale processing can be implemented

There are many products that claim to allow development using only a GUI, but some people may have a negative impression of such products as being too simple.

It is true that things like "it's easy to make, but it can only do simple things," "when I tried to execute a full-scale process it couldn't process and crashed," or "it didn't have the high reliability or stable operating capacity to support business operations, which caused problems" tend to occur.

"DataSpider" and "HULFT Square" are easy to use, but also allow you to create processes at the same level as full-scale programming. They have the same high processing power as full-scale programming, as they are internally converted to Java and executed, and have a long history of supporting corporate IT. They combine the benefits of "GUI only" with full-scale capabilities.

No need to operate in-house as it is iPaaS

DataSpider can be operated securely on a system under your own management. With HULFT Square, a cloud service (iPaaS), this "connecting" technology itself can be used as a cloud service without the need for in-house operation, eliminating the hassle of in-house implementation and system operation.

Related keywords (for further understanding)

Machine learning related keywords

Keywords related to Generative AI/ChatGPT

Keywords related to data integration and system integration

  • EAI
    • It is a concept of "connecting" systems by data integration, and is a means of freely connecting various data and systems. It is a concept that has been used since long before the cloud era as a way to effectively utilize IT.
  • ETL
    • In the recent trend of actively working on data utilization, the majority of the work is not the data analysis itself, but rather the collection and preprocessing of data scattered in various places, from on-premise to cloud.
  • iPaaS
    • A cloud service that "connects" various clouds with external systems and data simply by operating on a GUI is called iPaaS.

Are you interested in "iPaaS" and "connecting" technologies?

Try out our products that allow you to freely connect various data and systems, from on-premise IT systems to cloud services, and make successful use of IT.

The ultimate "connecting" tool: data integration software "DataSpider" and data integration platform "HULFT Square"

"DataSpider," data integration tool developed and sold by our company, is a "connecting" tool with a long history of success. "HULFT Square," a data integration platform, is a "connecting" cloud service developed using DataSpider technology.

Another feature is that development can be done using only the GUI (no code) without writing code like in regular programming, so business staff who have a good understanding of their company's business can take the initiative to use it.

Try out DataSpider/ HULFT Square 's "connecting" technology:

There are many simple collaboration tools on the market, but this tool can be used with just a GUI, is easy enough for even non-programmers to use, and has "high development productivity" and "full-fledged performance that can serve as the foundation for business (professional use)."

It can smoothly solve the problem of "connecting disparate systems and data" that hinders successful IT utilization. We regularly hold free trial versions and hands-on sessions where you can try it out for free, so we hope you will give it a try.


Why not try a PoC to see if "HULFT Square" can transform your business?

Why not try verifying how "connecting" can be utilized in your business, the feasibility of solving problems using data integration, and the benefits that can be obtained?

  • I want to automate data integration with SaaS, but I want to confirm the feasibility of doing so.
  • We want to move forward with data utilization, but we have issues with system integration
  • I want to consider data integration platform to achieve DX.

Glossary Column List

Alphanumeric characters and symbols

A row

Ka row

Sa row

Ta row

Na row

Ha row

Ma row

Ya row

Ra row

Wa row

»Data Utilization Column List

Recommended Content

Related Content

Return to column list