We are delighted to share the latest instalment of our monthly CEO interview series. This time, Neurocle’s CEO Hongsuk Lee gives us a rundown of his company and explains what sets Neurocle apart from the competition.
He outlines the company’s main focuses, reveals which solutions have brought them the most success, and predicts how deep learning software might continue to develop in the future. This interview was featured in Connect, the latest issue of MVPro Magazine.
1.) What was your journey to becoming CEO of Neurocle?
I have over a decade of experience in the machine vision and display industries, focusing on deep learning and strategic market development. In 2019, I founded NEUROCLE Inc., where I currently serve as CEO.
I’ve led the development of our deep learning vision software, successfully releasing multiple versions and growing the company to over 35 employees. Under my leadership, the company reached its break-even point within three years, establishing itself as a key player in the vision inspection space.
Before founding NEUROCLE, I worked at LG Display in corporate strategy and marketing, where I provided strategic insights into the ICT and automotive display markets. This included securing contracts with major electric vehicle manufacturers and driving the creation of new development teams to innovate high-end display technologies.
2.) What are Neurocle’s main focuses?
In every production process, products must undergo quality inspection. Our company provides deep learning technology that enhances the accuracy of vision inspections, ensuring products of perfect quality.
Vision inspection has evolved from manual checks to automated processes. For automated inspections, deep learning, which combines human-like flexibility with machine speed and consistency, is becoming essential. We develop software that facilitates the creation and deployment of deep learning models in real production lines. Our software offers solutions with up to 99.9% accuracy.
3.) How does Neurocle differ from other developers of deep learning software?
While many companies develop deep learning software for vision inspection, few can match the model performance we achieve. This is due to our proprietary Auto Deep Learning Algorithm. It automates the essential tasks of hyperparameter tuning and model architecture optimisation.
Integrated into our no-code model creation software, this algorithm handles everything, so users only need to click a button. Even without any prior deep learning knowledge, users can generate high-performance models.
Anyone with experience in applying deep learning models knows that even a 0.1% margin of error can make a significant difference. We guarantee that level of precision, thanks to our core Auto Deep Learning technology. Users consistently report two things: our software delivers high performance, and it’s incredibly easy to use.
4.) Which products are your company most known for, and where are they used?
Our deep learning model training software, Neuro-T, which features the Auto Deep Learning Algorithm, has been widely acclaimed by our customers. Its success is driven by several key features. Firstly, it offers nine types of DL models, based on the Auto Deep Learning Algorithm. We offer nine models, allowing users to choose the one best suited to their application and data. Unique models like our patch classification model, synthetic defect image generator model, and two types of unsupervised anomaly models are exclusive to us.
Next, AI-based labelling features allow users to click on objects, drag areas, or enter a simple keyword, and labelling is done automatically. Users can manually label a subset of images, and then use the auto-labeling function to apply the same criteria to the entire dataset, drastically reducing the time and resources needed for manual labeling.
Also, our Auto Deep Learning Algorithm optimises models for faster inference, and we offer speed optimisation options tailored to the user’s environment. Whether running on lightweight embedded devices or high-performance machines, our models guarantee fast inspection speeds. As a software development company based in Korea, we’ve established a solid reputation, particularly with major corporations like Samsung, LG, and Hyundai, who use our software for inspection purposes. Building on this recognition, we’ve expanded into international markets, gaining notable references in Asia and Europe with companies such as P&G, Applied Materials, and Faber-Castell.
5.) In what way(s) do your solutions boost efficiency for its users?
The primary reason companies adopt our products is to automate inspections. Automation not only speeds up the process but also reduces errors compared to manual inspections, improving production efficiency and maximising yield. We simplify the deep learning-based inspection automation process for companies by streamlining several key steps, which are data acquisition, data labelling, hyperparameter tuning and model structure optimisation, model training, model validation (repeating training and validation until the desired performance is achieved), and deployment on the production line.
By using our solutions, steps 2, 4, and 5—labelling, training, and validation—are significantly expedited, allowing companies to achieve optimal model performance in the shortest time, thereby shortening project timelines. One of our large corporate clients was using an existing solution but faced excessive false positives during inspections. To address this, they had to manually filter the false positives after the deep learning model’s initial inspection, which was time-consuming. They attempted to develop an in-house deep learning algorithm to improve the results, but achieving the desired performance proved challenging.
As a result, they explored other solutions and conducted a blind test with three different software platforms. Ultimately, Neuro-T outperformed the competition with an accuracy rate of 99.8%. After implementing Neurocle’s Auto Deep Learning model, the need for a secondary manual inspection process was eliminated, and the deep learning model now handles the entire inspection process.
6.) Lastly, which trend(s) do you see being a factor for the rest of 2024 and beyond?
In 2024, there will be a growing demand for lightweight yet high-performance models. As industries increasingly rely on smart cameras and more resource-constrained environments, deep learning models need to offer high accuracy while minimising computational load. These lightweight models are essential for real-time processing, where fast, efficient decisions are required without the burden of large-scale infrastructure.
However, while the demand for such models is high, implementing them, especially in edge environments, remains a challenge. Maintaining both performance and efficiency is difficult when computational power and memory are limited. As a result, while lightweight models will continue to gain traction, the complexity of delivering high-performance solutions in constrained environments will pose significant hurdles for developers.
Hongsuk Lee is the Founder and CEO of NEUROCLE Inc., a leader in deep learning visionsolutions for manufacturing. Neurocle grew to 40 employees, launched multiple softwareversions, and achieved profitability within three years. Previously, at LG Display,HongsukLee developed strategies for the automotive display sector and secured key contracts withelectric vehicle manufacturers
