skip to Main Content

Sparse Modeling Technology
and How We Make It Work

Hacarus specializes in AI (Artificial Intelligence). We use Sparse Modeling as the main method for extracting meaning from small data sets.

When fast and efficient results are expected, our technology makes a strong difference.

Sparse Modeling has several key benefits compared to Deep Learning – which is the most common approach in AI:

HACARUS’ Sparse Modeling Technology

Small data

Sparse Modeling focuses only on the very essential parts and works with small amounts of information.

Explainable AI

Sparse Modeling maintains a transparent model that can be both reviewed and verified. Hence the results are understandable and explainable.

Minimal computing power

Since Sparse Modeling only consumes minimal power, it can be easily embedded into low-cost equipment or FPGA.

Common Deep Learning AI Technology

Big data

Deep Learning algorithms need a very large amount of training data to learn and build up a model.

Black box AI

The models created in a Deep Learning setup work as a black box. They don’t explain why they return a certain result or decision.

Large computing power required

The special computer equipment required to process Big Data is expensive.

Key Challenges with Deep Learning

We apply newest scientific research insights to bring AI into areas where Deep Learning is too costly.

Learn how your company can benefit from our solutions.

Big Data Collection

Big data collection

In many scenarios, it is difficult or costly to collect Big Data.
One good example is cases that rely on using drone pictures: due to battery and memory limitations, drones can operate only for short durations. As a result, the amount of image information that can be collected is very limited.

Black Box Problem

Black Box problem

Deep Learning algorithms don’t expose how they came up with a result or recommendation. Hence the computed results and recommendations are not explainable. Particularly in the medical field, it is critical to have a transparent decision process.

Contact us for more information about our solutions for medical applications.

Computing Cost

Computing cost

Deep Learning algorithms require expensive computers on which to run, and such machines can easily cost 15k USD or more.
Our Sparse Modeling algorithms even work on low power FPGA chips and can be easily deployed as an extension to various types of equipment.
Schedule a demonstration with us to learn more about Hacarus’ embedded and FPGA technology.

Proven by Results – Our Sparse Modeling Technology works

Our technology is based on Sparse Modeling. That means it focuses on only the most essential elements of the underlying information. With this approach, the complexity of the data analysis is reduced. As a result, we can derive meaningful information from small amounts of input data.

Here is how it works

If you have only a few samples with a large variety of characteristics, it’s generally impossible to apply AI algorithms with reasonable performance. However, focusing on a few really important characteristics to outcomes can solve the issue. Sparse Modeling has the capability to identify such important characteristics automatically. with very reasonable accuracy in prediction or classification of the problem.

Another feature of Sparse Modeling technology is that it can clarify the causal relationships between different pieces of data. We can use it to derive “the correct data itself”, which is not possible with deep learning.

Limitations of Sparse Modeling

Depending on the use case, the benefits of our technology cannot always be directly applied. Some AI scenarios, such as autonomous driving or natural language processing, require Deep Learning.

In these types of scenarios, our technology can be combined with Deep Learning to combine the benefits of both approaches.

Our experts are here to help. Please contact us to find out how Hacarus’ technology can be applied for your specific set-up.

Contact us

Applied Science

Since 2016, we have successfully applied the newest scientific insights to develop and enhance our Sparse Modeling technology for different use cases. This makes Hacarus a pioneer in machine learning with Sparse Modeling.

Masayuki Ohzeki

Chief Science Advisor
Masayuki Ohzeki

Masayuki OHZEKI is an associate professor at the Graduate School of Information Science at Tohoku University. His research interests are broad, including machine learning and its potential, not only in terms of theoretical physics, but in a variety of other fields as well. He is actively promoting “Sparse Modeling” techniques. In 2008, he graduated with a Ph.D. in physics from Tokyo Institute of Technology and subsequently spent one and a half years as a postdoctoral fellow. He worked as an assistant professor in the Kyoto University and also as a researcher at Rome University. Since 2016, he has held his present position. He was awarded the 6th Young Scientists’ Award of the Physics Society of Japan, and the Young Scientists’ Prize by The Commendation for Science and Technology by the Minister of Education, Culture, Sports, Science and Technology in 2016.

Kaoru Kawamoto

Kaoru Kawamoto

Kaoru KAWAMOTO was appointed as a professor at Shiga University’s Department of Data Science in April 2018. Prior to his current role, he was the chief of Osaka Gas’ Business Analysis Center. Through his work and influence, Osaka Gas became a data-centric organization, that guides all strategic decisions of the company based on data.  Mr. Kawamoto received many awards for his achievements, including the Nikkei Information Strategy’s Chosen Data Science of the Year. One of his publications, “How to Change a Company Through the Power of Analysis” (Koudansha Modern Age Paperback) is very popular among data science experts and recently became the topic of many lectures and media appearances.