Companies want to build new business models with the help of AI, but getting well-prepared data as a basis is a problem
Viele Unternehmen wollen KI in ihre Prozesse einbauen, aber nur wenige haben die Reichweite und finanzielle Mittel, um die dafür benötigten Daten gut aufzubereiten.
"Labelling as a service" makes it possible to outsource the process of cleaning, labelling and structuring the data
Eine Kombination aus KI und Experten für Datenlabeling macht es möglich, Daten in hoher Qualität schnell und kostengünstig zurückzusenden: So werden durch den „Labelling as a service” Daten strukturiert, bereinigt und gelabelt.
The process is more iterative due to experience and more work is done on the instructions with the customers, as there is a lot of (improvement) potential here.
What had the customer tried before?
Label/annotate yourself using Excel and other tools
Build your own labelling tool (requires 2-3 months work)
Outsourcing: Mechanical Turk, but here the quality is poor.
Additional challenges at the customer
Quality of the annotation
Data labeling is iterative. You are constantly learning and may have to start over again
Customization: Data Labelling needs tailor-made solutions because every company and every project is unique
80% time saving
Iterative: Customers can initiate changes in the middle of the process instead of starting from the beginning
The project schedule
Depending on the application, the first thing to discuss with the customer is the choice of the correct process, unless the customer already knows exactly what he needs.
Input: Data (=image, video, text, audio or satellite data) + instructions on what to do with the data: Output: What format should the output have.
Optional: Customer uploads 'ground truth' data that we need for quality measurement.
Provider annotates, sends data back.
Customer checks, changes ground truth, changes instructions, changes data etc. until everything is correct.
From one day to continuously.
Skills required by the customer
Data Science or KI know-how; alternative Product Management.
The project schedule
Data upload via API or directly, or via Curl, Python, Json. Output via Json files.
Where is the data stored?
The customer sends the data. This data is temporarily stored on AWS servers in Europe or the USA and then sent back to the customer.
Brad Cordova (CEO)
Brad Cordova is the CEO of Canotic. Previously founder of TrueMotion (150 person AI company), Forbes 30 under 30, String Theory @ Caltech/CERN.
The information may of course vary in individual cases. Please contact the provider for an assessment of your project.