AI has Evolved. Accelerate Machine Vision Innovation with Data-Centric AI.
Understanding Data-Centric AI
What Is Data-Centric AI?
Think of a Data-Centric AI system as programming with focus on data instead of code. Industries of all types continue to adopt AI solutions, and while AI models have improved over the years, a fundamental shift is needed to truly unleash AI’s full potential.
“Instead of focusing on the code, companies should focus on developing systematic engineering practices for improving data in ways that are reliable, efficient, and systematic. In other words, companies need to move from a model-centric approach to a data-centric approach.” — Andrew Ng, CEO and Founder of LandingAI
Why Does Data-Centric AI Matter?
By adopting a data-centric AI approach, companies from diverse industries such as automotive, electronics, and medical device production have seen improvements in deploying AI and deep learning–based solutions in manufacturing scenarios compared to traditional, rules-based implementations. Some improvements we’ve seen from adoption of a data-centric approach include:
build computer vision applications 10x faster
reduced time to deploy application by 65%
improved yield and accuracy by up to 40%
Challenges: Traditional, Rule-Based Machine Vision
Rules-based machine vision algorithms may struggle with inconsistencies that are the result of working with parts with complex features or ambiguous part defects.
Inconsistencies and High-Rejection Rates
Producing high-quality products remains a top priority for manufacturing companies. Defective parts present:
- safety issues
- damaged customer relationships
- lost revenue
Identifying defective parts early in the process helps mitigate these risks. Machine vision technologies offer an effective method for doing so but may come up short in certain instances.
Introducing new or custom parts or working in an environment that constantly changes, such as lighting conditions in a plant with large windows, can also be problematic. In these scenarios, rules-based algorithms produce a high rejection rate, as the technology cannot distinguish between actual defective parts and acceptable variation, which necessitates a high rate of human follow-up inspection, increasing costs and slowing down production lines.
Challenges: Deep Learning and AI Solutions
If human experts disagree on a label, how can the AI system be expected to make such a determination?
For applications with small datasets and rare defects, this becomes even more problematic.
“Is this really a scratch?”
In scenarios where rules-based algorithms will not suffice, many companies turn to AI and deep learning solutions yet still encounter challenges. Without a consistent data management approach, for instance, an AI system cannot accurately inspect products. While the system can analyze images of products on an assembly line and identify defects such as scratches and dents, this can present a challenge as well because human experts won’t always agree on appropriate labels when it comes to describing the damage.
Reliance on Developer Results in Delays
Teams rely on the developer to optimize performance and support AI models. For example, developers must work with subject matter experts on clearly defining defects. Maintaining models and adapting to new circumstances such as new parts or environmental changes presents challenges and delays in deployment. In many cases, it may take several months to create and deploy an AI model.
A Huge Workflow Headache
Standardizing a workflow and scaling AI solutions also presents challenges. Different teams use different methods to develop AI solutions, work with quality teams, and manage data, making it difficult for teams to learn from other groups and standardize workflow. Given such circumstances, scaling even one team to develop a solution can be onerous.
Improve Data, Improve Accuracy
A data-centric AI approach provides a systematic method for improving data, reaching a consensus on the data, and cleaning up inconsistent data.
Data-Centric Impacts Performance
A data-centric AI approach involves building AI systems with quality data — with a focus on ensuring that the data clearly conveys what the AI must learn. Doing so helps teams reach the performance level required and removes unnecessary trial-and-error time spent on improving the model without changing inconsistent data.
Data-Centric Promotes Collaboration
Quality managers, subject matter experts, and developers can work together during the development process to:
- reach a consensus on defects and labels
- build a model
- analyze results
- make further optimizations.
Data-Centric Reduces Development Time
With such an approach, teams can work in parallel and directly influence the data used for the AI system. By removing unnecessary back and forth among groups and looping in human input at the point where it’s needed most, the result is reduced development time.
Let’s compare the data-centric approach to the model centric approach in terms of which one is more approachable.
Even More Benefits of Data-Centric
Additional benefits of data-centric AI include the ability for teams to develop consistent methods for collecting and labeling images and for training, optimizing, and updating the models. Teams easily learn from a past project’s success and can apply that knowledge to quickly scale new projects.
LandingLens Is Data-Centric in Action
LandingLens, an industry-first data-centric AI visual inspection platform, helps ensure product quality by improving inspection accuracy and reducing false positives. The platform standardizes developing deep learning solutions that reduce development time and scale projects quickly to multiple facilities. Ultimately, LandingLens makes deep learning deployment easier, faster, and more effective, helping to increase throughput, maintain product quality, and drive revenue.