Andrew Ng and Landing AI seek to democratize AI for all company sizes, drive wider industry adoption
Example of computer vision project created using LandingLens.
Image Credit: Landing AI
Check out all the on-demand sessions from the Intelligent Security Summit here.
Andrew Ng’s cloud-based platform for computer vision, Landing AI, is taking on the advent of artificial intelligence (AI) development among companies of all sizes with its latest offering, LandingLens. The solution promises to facilitate swift creation and testing of computer vision AI projects, without the need for intricate programming skills or prior AI experience.
“We started by exploring the manufacturing sector, one of the hardest industries in which to deploy computer vision. Then we found the tools we had built for manufacturing, with relatively few modifications, can also be useful for many other computer vision applications,” said Ng, noted AI academic, and founder and CEO of Landing AI.
The company announced today that its flagship computer vision product, LandingLens, is now available for a free trial, coupled with a new pricing scheme that enables pay-as-you-go usage beyond the initial trial period.
“With the new platform, we aim to expand our tool’s use cases across several other industries,” Ng told VentureBeat. “To me, it’s about achieving our goal of democratizing the creation of AI.”
Intelligent Security Summit On-Demand
Learn the critical role of AI & ML in cybersecurity and industry specific case studies. Watch on-demand sessions today.
“We want everyone to start for free and try it out to understand its use cases. We’re eager to make it available for more people,” he said.
A data-centric approach to AI and computer vision
According to Ng, the platform’s data-centric AI system focuses on data instead of code, and as various industries increasingly embrace AI solutions, a fundamental shift is necessary to unlock the complete potential of this technology.
LandingLens prioritizes enhancing data quality for AI models, thereby enabling its functionality, even in cases where companies have limited data available for training the AI models, a common challenge encountered by most firms. The “data-centric” strategy involves training AI models to function proficiently with modest amounts of quality data rather than relying solely on the vast datasets that typically underpin AI applications in large-scale internet companies.
“Over the last few years, we did much work with customers that often had small datasets. During these experiences, we discovered multiple technology steps and optimizations that now enable our algorithm to work well on smaller datasets,” said Ng.
He explained that the model was trained on a ResNet dataset for image recognition, and in the backend, LandingLens’s pretrained algorithm utilizes AI-based automatic hyperparameter tuning, enabling it to work well with datasets of every size. When data is passed through the model, it’s optimized through numerous steps to deliver well-analyzed, high-quality output and detailed insights.
Recently, therapeutic antibody discovery firm OmniAb used LandingLens to successfully automate its visual inspection process, significantly increasing efficiency and throughput. In addition, the platform aided OmniAb in increasing AI access within its organization for use cases that involve people who are not high-level scientists.
How does it work?
To maintain data consistency within LandingLens, the platform uses an advanced labeling technology that automatically detects and corrects mislabeled images, enhancing overall data quality.
This collaborative labeling approach allows multiple users to label images and facilitates the process of reaching a consensus through data cloud and edge device deployment capabilities. As a result, deploying and testing your model can be achieved with just a few clicks of the mouse. Users can select the deployment option that best suits their requirements, ranging from a windows application to a programmatic API.
Additionally, LandingLens employs a continuous-learning mechanism that ensures that the created model remains up to date by integrating new data from the deployment environment to retrain the model.
“We want to make the model development workflow easy for users. The traditional approach to developing AI models has always been labeling, training to deployment. We want to ease this development workflow by having users not write much code, but focus more on data entry,” added Ng.
Landing AI’s future focus on computer vision
Ng said the company would continue to focus on developing the LandingLens platform as a single tool that serves multiple computer vision applications.
“Use cases in computer vision are currently keeping us very busy. Many customers across industries are requesting us to add more features for cases such as streamlining heterogeneous data. So our current roadmap involves a lot more work to do in computer vision,” said Ng.
Through the LandingLens platform, Ng aims to solve issues found today with customization or longtail AI model development, which he sees as the most significant barrier to widespread AI adoption.
“The only way for organizations to unlock maximum value from their AI projects is when they have the liberty to customize their AI system as they need. They can do this by engineering the data rather than the code. This way, companies can adjust to the shifting market requirements and develop better models using lesser human resources,” explained Ng. “So, I’m excited about facilitating the goal of further democratizing access to AI creation.”
The company is pursuing applications in automotive, electronics and medical device manufacturing sectors. Ng said embracing a data-centric AI methodology and implementing AI and deep learning-based solutions for computer vision scenarios will benefit this diverse range of industries.
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.