Learn how you can streamline your AI workflow by labeling only the data that provide valuable information and by using an arbitrarily large hardware infrastructure to execute it – all of this comes in an easy-to-use environment powered by Samsung SDSA’s Brightics AI Accelerator and RedBrickAI.
Active Learning Performance in Labeling Radiology Images Is 90% Effective
Automate 80% - 90% of manual data labeling effort
autoLabel is a human-in-the-loop, active learning system which sorts data by uncertainty, and human labelers start by labeling a small, automatically and carefully selected portion of the data. The autoLabel system trains a model on the labeled data and uses it to sort the remaining unlabeled data in order of confidence. Human labelers label the most informative images and autoLabel improves its model in several iterations. After 5% - 16% of the data is labeled manually, the confidence is typically so high that no further manual labeling needs to happen, and the remaining dataset can be labeled automatically. These automatically labeled images can now be checked by human domain experts with over 80% less effort than creating the labels themselves.
As the complexity and scale of datasets increase, the processes surrounding labeling datasets also need to evolve. You need a comprehensive toolset to accelerate labeling projects, reduce cost, and maintain high quality output. Structure, automate and qualify your labeling workflows by using the RedBrick AI platform. By building a completely custom labeling workflow, your team can easily carry out autoLabel active learning iterations.
Gain and maintain visibility into the productivity of your workforce, and track how autoLabel automates data labeling over time. Your labeling tasks will get automatically routed in the Active Learning workflow, and assigned to the appropriate stakeholders -- simplify the project management and focus on your science.
The rapid pace of AI innovation makes designing and training accurate AI models challenging. With autoLabel, you can eliminate guesswork and get started faster by automating the initial tasks of labeling data. This will get you a fully, accurately labeled dataset in the most efficient way possible. After that, you can train and fine-tune your model using Brightics AI Accelerator and later run it at inference on unlabeled data.
Compared to manually labeling all of the dataset, autoLabel reduces the time it takes to start labeling 66% and the Total-Cost-of-Ownership (TCO) up to 54% depending on the use case.
Because Samsung SDS’s autoLabel solution pre-processes the dataset in order of confusion before the first manual labeling iteration, it is able to exploit the most informative data to reduce active learning iterations and cloud infrastructure costs 33% over competing cloud offerings.
Customers pay for the number of 512 x 512 pixel or lower resolution tiles labeled in addition to cloud training and inference instance costs.
autoLabel software License
Tier 1 Price per Tile (<50K tiles): $0.08
Tier 2 Price per Tile (50K < x < 1M): $0.04
Tier 3 Price per Tile (>1M): $0.02
Total autoLabel SaaS
Tier 1 Price per Tile (<50K tiles): $0.135
Tier 2 Price per Tile (50K < x < 1M): $0.095
Tier 3 Price per Tile (>1M): $0.075
Example Pricing: For applying object detection labels to a dataset of 100,000 tiles, we estimate 11 autoLabel iterations, each consisting of 1% of the dataset, which would require manual labeling of 11,000 tiles. The autoLabel system will apply labels to the entire dataset of 100,000 tiles, automatically. The autoLabel SaaS price estimate would be 50,000 * $0.135 + 50,000 * $0.095 = $11,500, and the end customer would also pay 11,000 tiles * 7 labels/tile * $0.036/label * 3 labelers = $8,316 in estimated manual labeling costs. The customer TCO would be $19,816 = $11,500 for autoLabel SaaS + $8,316 for 3rd party manual labeling assuming that each manually labeled tile is separately labeled by 3 human labelers to disambiguate labeling quality. Note: Manual labeling is not part of this offering, and the price for a human-generated label is stated here only for comparative purposes.
Ben Bongalon, Advanced Analytics & Development Manager, Intuitive Surgical
Nasim Eftekhari, Applied AI and Data Science Director, City of Hope
Zakia Rahman, Clinical Professor of Dermatology, Stanford University School of Medicine
Pranav Kumar, Senior Technical Lead, Innovecture
Pranav Kumar, Senior Technical Lead of Innovecture
Save time and deliver value faster with automated machine learning and automated deep learning in only a few lines of code.
Whether you’re looking for a specific business solution or just need some questions answered, we’re here to help.
"All supervised models that we use today have been trained on labeled data. It is always the most expensive part of any AI solution to label the data for training in every industry. It is even more difficult in healthcare because … these are very specialized images … that need expert labeling and annotation. It is very time-consuming and expensive because doctors and healthcare professionals have to annotate these images and each of them takes hours and hours of time. It is very difficult to feed a whole slide image into a deep learning method because each takes up multiple Gigabytes and so we have to break them into tiles. Usually, each slide is broken into thousands of tiles. Not every tile is cancerous. Thus, each tile needs to be examined and labeled. We wanted to explore the advantage of Samsung’s active learning in auto-labeling these images. There is no question about the value that deep learning brings to auto-labeling pathology images. We are really excited to explore the idea of a system that can learn little-by-little from pathologists."
"Dermatology represents two great frontiers for new business models powered by artificial intelligence. Patients can take mobile phone pictures of their skin and obtain instant reliable diagnoses of any condition, and people may use pictures of their healthy faces and bodies to compare them against society’s beauty standards for precision cosmetics to look better while not looking abnormal. The biggest obstacle to both is having an expertly annotated large dataset of images. We want to create such datasets and models in a partnership with Samsung SDS. The skin, in addition to being the largest organ, is also the most visible. This accessibility has resulted in an exponential increase in the number of images. The skin is, and will likely continue to be, the most imaged organ. While there is potential for democratizing diagnosis for the general public, the impact to mental health through image distortion cannot be overstated."