AI venture SOINN from Tokyo Technology Started Internet sales of image recognition / exterior test AI "Switch Vision (R) Light" and "Energy Demand forecast AI"

SOINN Inc. (head office: Machida City, Tokyo, CEO: Osamu Hasegawa, hereinafter referred to as our company), an AI venture from Tokyo Tech,

・Energy demand prediction AI "E-1F"

・ Ultra-lightweight, high-precision image recognition and appearance inspection AI "Switch Vision (R) Light"

Online sales of

started on January 4, 2022. In the future, we plan to expand the types and functions of the AI ​​we sell and develop it into the second pillar of our business. Taking advantage of the strengths of e-commerce, we are also considering overseas expansion.

Many of our AIs can achieve high accuracy comparable to deep learning with ultra-lightweight calculations from a small amount of data, and are also capable of automatic learning and automatic inference. This online sales was realized by utilizing the strengths of our AI.

About AI for online sales

(1) Energy demand prediction AI "E-1F"

"E-1F" is an ultra-lightweight AI that automatically learns from both the actual data on the energy demand of the forecast target and the weather data of the region where the forecast target is located, and exhibits the following functions. The energy demand prediction function of "E-1", which has a commercial track record as an AI for large-scale energy management systems, has been compactly summarized.

1. Predicting future energy demand based on weather forecasts

2. Detection of anomalies and signs of anomalies based on discrepancies between predicted and actual values

"E-1F" can be operated automatically from initial learning on a commercially available PC connected to the Internet (CPU is used, GPU is not used). Weather data used for learning and forecasting is also automatically downloaded from the Internet.

E-1F product web page: https://soinn.com/service/E-1F/

E-1F demo video: https://youtu.be/l-xKc7MDLmI

E-1F Operation Manual (excerpt): https://soinn.com/wp-content/uploads/2022/01/SOINN_EMS_manual.pdf

Energy demand forecast execution screen example

Evaluation results of follow-up performance to fluctuations in power performance data

(2) Switch Vision (R) Light "SV-L"

Switch Vision(R) Light is an image recognition/appearance inspection AI that can be used on-site without requiring huge amounts of learning data or calculations.

・Super lightweight supervised learning

・Good product learning

・Semi-supervised active learning

We can use the customer's data to perform the initial learning and adjustment, which requires the most man-hours, free of charge. If it is difficult to disclose the data to our company, it is also possible for the customer to make adjustments from the very beginning.

After confirming the effectiveness of Switch Vision(R) Light for your data, functions, performance, and usability free of charge, you can decide whether to transfer to a paid license. . In some cases, a free trial operation of Switch Vision(R) Light is also possible.

You don't need AI expertise to use Switch Vision(R) Light. For this reason, it is possible for a part-time inspector, for example, to directly adjust the Switch Vision(R) Light on a commercially available PC installed on-site, let it learn, and operate it.

Compared to Deep Learning, if you get the same level of accuracy with ultra-lightweight supervised learning, you can expect 1/50 to 1/100 reduction in the number of training images, hardware cost, and training time. .

Deep Learning: about 10 hours with GPU

Switch Vision(R) Light: A few minutes on the CPU of a commercially available PC

(2-1) Overwhelming ease of use with AI visualization/editing interface

The following video is an example of the operation of the "Switch Vision(R) Light" interface during ultralight supervised learning. As shown in the video, Switch Vision(R) Light enables (1) initial learning, (2) label adjustment and re-learning, and (3) additional learning while checking what the AI ​​has learned and how.

https://youtu.be/YOyNMLu53Fk

Generally, the data near the OK/NG boundary is the most important for OK/NG discrimination. With Switch Vision(R) Light, you can see through the "contents of the AI's head", where the boundary between OK / NG is, and what kind of image is coming in the vicinity. Therefore, by intensively adding images near the boundary, it is possible to improve the accuracy very efficiently.

As a general rule, the series of processes such as confirmation of the above learning process, pinpoint data addition, OK/NG label adjustment, and re-learning are all performed on the customer's own commercial PC (no GPU). can be run with You don't have to place an order with us every time you make adjustments, which is costly and time-consuming. The AI ​​model adjusted and learned by the customer can be immediately put into recognition and inspection work as it is.

(2-2)Switch Vision(R) Light can handle tasks that Deep Learning cannot handle

In addition, Switch Vision (R) Light may be able to handle tasks that cannot be handled by Deep Learning due to technical and cost issues, starting with a free trial.

Please feel free to contact us for tasks that you have given up on AI until now.

(2-3) Application example of Switch Vision(R) Light

・Defective product visual inspection at food packaging factory

・Previously, field workers labeled tens of thousands of learning images

・Creating a model over 10 hours with Deep Learning using a high-performance machine

・Consult us if you don't want to spend the above time and money on each product

・Currently, hundreds of OK/NG images are adjusted and learned directly on the on-site notebook PC to reach the same performance

・Defective visual inspection of PET bottle products

・Tasks that deep learning companies said they couldn't handle

・Confirm OK/NG identification and defect heat map function with free trial

・Collective learning of all directions even for cylindrical products

・It is possible to change the shape, color, and label design of the bottle simply by giving a good product.

・Learning is about tens of seconds on the CPU of a commercially available notebook PC

・Can be put into inspection work immediately after learning

・Defective visual inspection of cut food on belt conveyor

・Introduce the image mixed with OK/NG to Switch Vision(R) Light

・Provide OK/NG candidates on the interface by automatic analysis

・Perform supervised learning by simply labeling about 20 candidates presented, quickly reaching high accuracy

・It is possible to detect and notify unknown targets and encourage the addition of new labels

・Learning is about tens of seconds on the CPU of a commercially available notebook PC

SV-L product web page: https://soinn.com/service/v-1l-switch-vision-light/

SV-L demo video: https://youtu.be/YOyNMLu53Fk

SV-L execution screen example

How to use

Web page dedicated to AI online sales will be released soon. At this time, please email us the following information: We will get back to you with a link to the web page.

Send to: biz@soinn.com

Information: Your name, company name, affiliation, email address, phone number,

Desired AI (E-1F, SV-L), request for free AI adjustment,

Others (questions, requests, etc.)

If we can keep your data, we can also learn and adjust the AI, which requires specialized knowledge, free of charge. (If the amount of learning data is huge, the task is extremely difficult, etc., we may not be able to accept the free AI adjustment.)

After downloading and setting up AI, both (E-1F, SV-L) can be used immediately. 14 days from the start of operation is a free trial period.

If you continue to use it after the trial period ends, it will be transferred to a paid license. If you cancel during the trial period, you will not be charged anything.

■Usage fee

(1) Energy demand prediction AI (E-1F)

It is a flat rate setting of 4.980 yen per month.

*Please feel free to contact us regarding payment methods other than monthly payment (annual payment, etc.) and operation format adjustments (multi-site cloud collective operation, etc.).

(2) Switch Vision(R) Light(SV-L)

After hearing the difficulty of the task to which SV-L is applied and the type of usage, we will set the fee individually based on cost-effectiveness. We also flexibly respond to cases where the SV-L is incorporated into a customer's product and sold. Please feel free to contact us.

■Other consultations and inquiries

Please feel free to contact us via the CONTACT form on our website, email, or phone.

We also accept questions such as whether our AI can be applied to your data and tasks.

CONTACT form: https://soinn.com/contact/

Phone number: 050-3196-2118

Email: biz@soinn.com

■Company Profile

Trade name: SOINN Co., Ltd.

Representative: Osamu Hasegawa

Location: 8-4-30 Tsuruma, Machida City, Tokyo

Established: 2014

Business content: Utilizing advanced AI technology including the patented technology “artificial brain SOINN (R)”

　Intelligence of various devices, devices, and information systems

URL: https://soinn.com/