The majority of data scientists say over 80% of their models fail to be deployed. What can ML practitioners do to get ahead of the biggest pitfalls — before they derail your projects, cause expensive delays, or erode trust in your product?
Duncan Curtis (SVP, AI Product and Technology at Sama) reviews three of the most impactful pitfalls when building ML models—and what you can do to avoid them.
Not Annotating Your Data Strategically.
It’s not cost effective to annotate all of your data. How do you know if you’re choosing the right training and edge-case data?
Poor Quality or Inefficient Annotations.
Too many incorrect examples can prevent your model from learning the signal.
Lack of Visibility When Validating Models.
You need to know when and why your models are failing in order to build a more resilient model ahead of an unpredictable future.
Visit Sama.com for notifications about our upcoming live discussions and webinars.
Duncan Curtis is the Chief Product Officer at Sama where he ensures the ML/AI models powering AI technology products are of the utmost quality. Previously the head of product at Zoox, VP of Product at Aptiv and Product Manager at Google, Duncan leads the teams powering ML/AI technologies for enterprises such as Walmart, Google, and NVIDIA.
Lisa Avvocato is a veteran product marketer/moderator specializing in AI and ML technologies. She’s passionate about the interaction of machine learning and digital transformation strategies to reduce inefficiencies and drive sustainability. With over 15 years of experience in Enterprise SaaS technology, she has worked across a diverse set of industries including retail, education, manufacturing, and healthcare.
Sama provides ML Professionals and AI team Leads with an indispensable solution for Computer Vision data labeling.
