State of real-time machine learning survey - 2022
This survey aims to create a holistic picture of where the industry is at with respect to real-time machine learning: online inference and continual learning.

The aggregated results and insights will be shared publicly. We won't share your responses with any third party.

The survey should take ~5 minutes.

We'd also love to send you some cute little swags when we're out of stealth as a small token of gratitude!

For explanations of online prediction, batch prediction, streaming features, stateful retraining, etc., please see the post https://huyenchip.com/2022/01/02/real-time-machine-learning-challenges-and-solutions.html.

Thank you!
-- Chip Huyen and Zhenzhong Xu

Sign in to Google to save your progress. Learn more
Company size *
How many data scientists does your company employ? *
What best describes your role? *
How many ML platform engineers does your company have? *
What are the major ML use cases at your company?
How are your ML models doing prediction? *
Required
Approximately, what's the amount of data your largest ML model handles daily? *
If you're more familiar with the number of predictions your largest model makes a day, you can use enter that number in Other...
Are you experiencing challenges with online prediction? *
If you're facing challenges with online prediction, what are they?
How are your ML models being updated? *
Required
How often do you update your models with new data? *
If you have multiple models, answer this question for the model that is updated most frequently. Choose the answer that is closest to the actual update frequency.
How do you obtain ground-truth labels for new data? *
"Natural label" refers to tasks that come with natural labels such as recommendations, price prediction. For example, in recommendations, labels for recommendations are inferred through users' feedback. If a user clicks on a recommendation, we assume that it's a good recommendation. For stock price prediction, the actual stock prices are provided.
Required
If your tasks have natural labels, how delayed are the labels?
For example, for ad recommendations, if users don't click on an ad after 5 minutes, we assume that users will never click on it. The labels are delayed by at most 5 minutes. For stock pricing prediction, if you predict the stock price in the next 2 minutes, then it takes 2 minutes for you to know whether your model's prediction is correct or not. We say the labels are delayed by 2 minutes.
Clear selection
Next
Clear form
Never submit passwords through Google Forms.
This form was created inside of Claypot AI. Report Abuse