Challenges in delivering AI models

JavaScript isn't enabled in your browser, so this file can't be opened. Enable and reload.

Hi, my name is Petar and I am currently investigating the challenges in building robust and performant AI models. My goal is to understand which tasks are most important to improve and what type of tool/app/platforms would help AI teams the most for those critical tasks.

The questions are split in 3 topics: data quality, model improvement, and safety assessment.

Thank you for taking your time to fill out this survey. This survey should only take 30 seconds of your time.

Please fill in your name

Data quality questions

Below you find a few examples of typical data quality topics that are covered by your job. To understand which of the topics is most important to you, can you please mark every topic with a 1 to 5 score? 1 being not relevant to 5 being a very relevant topic to improve.

Data quality topics *

1 = Not Relevant 5 = Very Relevant.

Visualize dataset

Find wrong/inconsistent labels

Define data quality rules (to catch label errors)

Define data slices (edge cases)

Automate the above tasks

Any additional feedback on Data Quality? *

Model improvement questions

Below you find a few examples of Model Improvement topics that are covered by your job. To understand which of the topics is most important to you can you please mark every topic with a 1 to 5 score? 1 being not relevant to 5 being a very relevant topic to improve.

Model improvement topics *

1 = Not Relevant 5 = Very Relevant.

Find model blind spots (i.e. data slices with high error rate)

Find spurious correlations

Define custom data augmentations

Collect data in a targeted way (for specific scenarios)

Generate synthetic data (for specific scenarios)

Any additional feedback on Model Improvement? *

Safety assessment questions

Below you find a few examples of safety assessment topics that are covered by your job. To understand which of the topics is most important to you can you please mark every topic with a 1 to 5 score? 1 being not relevant to 5 being a very relevant topic to improve.

Safety assessment topics *

1 = Not Relevant 5 = Very Relevant.

Stress test model (robustness testing)

Construct a safety argument

Avoid uncertain decisions

Improve model robustness (e.g. invariance to noise)

Email (optional)

If you are interested in receiving the results of this survey, you can leave your email below

Submit

Clear form

Never submit passwords through Google Forms.

This form was created inside of LatticeFlow. Report Abuse

Forms