Why A Solutions Approach And Partnership Orientation Makes AI Sense

December 08, 2020

Experience suggests AI companies are always more successful when they work with a labeling services provider that collaborates with other AI enablement companies. Partnerships tend to enhance speed to market and increase operational efficiency in AI. But, is it vital for your labeling service provider to work with many specialized partners? I say, absolutely! In our companion piece on taking a solutions approach to AI, we talked about the importance of taking a solution-oriented approach to data annotation in the AI and Machine Learning ecosystem. You always want to keep your eye on the prize of achieving a successful AI solution. You don’t want to be diverted by obstacles that other companies have already solved for, or challenges that other companies have already created specialized technology-based capabilities to address. That’s why, as a practical matter, people implementing a solution-oriented approach typically benefit from the right combinations of specialized tools and customized capabilities. The trick is to choose a labeling services provider that works closely with a rich set of specialized partners, and understands the value proposition associated with each of these third parties. It is also important that labeling companies be able to recommend partners and tools quickly, in order to enable AI solutions efficiently and effectively based on capabilities. These capabilities typically fall into two categories…

  • Capabilities that focus on the labeling/metadata creation.  
    • For example, the ability to draw a complex polygon around an object by bounding it with just four points is productivity-enhancing for expert people doing the labeling work. A similar thing can be said about the ability to automatically track an identified object across multiple frames of a video.
  • Capabilities that bring flexibility, power, and productivity to the workflows established for the labeling/metadata creation.
    • For example, the ability to setup a different type of workflow can enhance productivity in certain situations. For example, perhaps I want to first identify all objects of a certain category by drawing a box around them – and then later append additional metadata to the boxed objects.

To at least partially illustrate the broad range of specialized partner capabilities that iMerit is fluent in, we’ll next describe the specialized capabilities of some iMerit partners. Notice that some partners have incorporated iMerit into the one cloud-based service that they operate in, while others can operate either on a variety of cloud-based services and/or on-prem.  

1) Amazon Web Services (AWS) – SageMaker Ground Truth

Amazon introduced SageMaker Ground Truth at re:Invent 2018 as a service to help AI/ML practitioners build accurate data sets for training. A Ground Truth labeling job can be sent to iMerit from within the AWS Marketplace – with standard Marketplace protocols. Charges for iMerit services appear on the customer’s AWS bill.

Ground Truth provides labeling workflows for humans to work on image and text classification, object detection and semantic segmentation. Users can also build custom workflows for their data labeling jobs. As the AWS Machine Learning blog says, “Good ML models are built with large volumes of high quality training data to help a model learn how to make the right decisions. You typically need a human to label the training data”. Ground Truth takes the time and tedium out of contracting a labeling services vendor for AWS customers.By the way, AWS customers benefit from iMerit’s full time U.S. and offshore based experts being able to work in both English and Spanish.

2) LabelBox

LabelBox is an AI/ML training data platform that supports data labeling.  The training platform handles images, video, text and audio. It lucidly enables a fully managed and dedicated workforce with an integrated training data solution. LabelBox has created best practices and requirements for configuring hybrid cloud and on premise deployment models. The platform includes multiple tools to outline complex shapes at the pixel level. Additional platform capabilities include…

  • Customization based on ontology requirements
  • A streamlined user interface for performance across a wide variety of devices
  • Seamless data connection via API or Python SDK
  • Team collaboration and performance monitoring
  • Real-time humans in the loop
  • Labeling quality
  • Labeling automation
  • Training Data Management

3) SuperAnnotate

SuperAnnotate provides a broad set of solutions for image and video labeling services. It has integrated tooling and on-demand narrow expertise in various fields. It uses a custom neural network, automation and training models powered by AI. SuperAnnotate enables users to…

  • Create high quality training data sets and labeling for Computer Vision tasks
  • Setup projects and distribute the tasks automatically
  • Scale and manage teams for large projects
  • Use active learning to more rapidly label images for accurate recognition
  • Automate labeling for pre-defined classes
  • Use transfer learning to increase prediction accuracy of new classes and datasets
  • Detect incorrect labels with quality assurance automation
  • View advanced analytics to track labeling speed and quality – and measure progress

4) Datasaur

Datasaur provides an intuitive interface for all Natural Language Processing (NLP) related tasks. Its powerful intelligence “under the hood” is designed to increase productivity and accuracy of labeling, workflow management, and AI-based Quality Assurance (QA). The platform includes…

  • AI driven models that proactively suggest labels
  • The highlighting for verification of labels that do not align with prior tagging results or seem contextually out of place
  • Enablement of the Natural Language Processing (NLP) use case called “Named Entity Extraction” in order to understand the core meaning of a sentence or text corpus
  • Enablement of the “semantic analysis” use case to understand the tone of a sentence, a result of advanced classifiers that go beyond the binary (e.g., positive vs. negative tone)
  • Enablement of advanced NLP tasks like conference resolution, dependency pairing and syntax trees

5) Deepen AI

Deepen is a data life cycle tools and services company focused on enabling autonomous vehicles and robots with ML and AI. It provides multi-sensor data labeling tools to accelerate creation of data training sets for AI. The platform includes…

  • A safety-first full AI data tool suite for labeling, calibrating, and validating (the full data life cycle)
  • Flexible delivery either in cloud or on premise
  • Advanced 2-D and 3-D capabilities
  • A quality checker that identifies common labeling issues

6) DataLoop

DataLoop weaves human and Machine Learning intelligence to accelerate Computer Vision use cases from development to production. It provides a cloud-based platform with embedded data labeling and data operations capabilities. The labeling capabilities include…

  • Spam detections 
  • Segmentation
  • Complex ontologies 
  • Object attributes
  • Parent-child hierarchies
  • Use of customized plug-ins
  • AI Annotation Assistants

With the above illustrations of specialized capabilities in the iMerit partner ecosystem, we can see that, depending on a particular customer scenario, there is big potential to accelerate AI/ML projects. The key is for the labeling service provider to quickly identify and securely integrate the right combination of specialized capabilities for each customer engagement. This requires a very nuanced understanding of the specialized capabilities. The nuanced understanding is earned by experience – and by retaining the experience within people. 

Since an organization can have many AI use cases and each use case may need different specialized capabilities over time, the ability to quickly navigate a big ecosystem for specialized capabilities is a critical success factor when it comes to providing your AI efforts with required high-quality data, regardless of the detailed challenges that pop up. Very importantly, the labeling services provider has to do all this with workflow/project management processes and experienced people that “bake in” quality. From a customer perspective, “the icing on the cake” is that, with the right labeling service provider, the customer’s focus can stay on achieving successful AI/ML solutions — and getting the high value answers that the AI efforts are focused on. This is important because the customer’s ability to maintain a solution approach is critical to being successful with AI. — Anthony Palella

(Anthony Palella is an iMerit contributor. He created the CDO Offering at Accenture and led the deployment of the offering across commercial and government clients. Prior to his work at Accenture, Anthony was the CDO at Angie’s List, Head of Big Data Analytics at Kimberly-Clark, and Head of Analytics at Fox Interactive Media.)