Enterprises building AI applications recognize that mastering anomaly detection and test case generation brings incredible value to improving machine learning model performance.
In this session, Eric Chu, Senior Manager, Data Science at Zoox and Anshuman Patnaik, Deep Learning Lead at Embark Trucks, unveil the challenges and opportunities of identifying data anomalies and generating test cases to optimize machine learning.
Here are 3 key takeaways from the session:
- To deal with anomalies, autonomous vehicle companies are setting up robust systems and metrics that capture and characterize different anomalous situations. Retrofitted vehicles with safety drivers drive around the city to gather various anomaly situations and once the data is ready, structured testing is carried out wherein the situations are replicated as much as possible both in simulation as well as the real-world environment.
- It’s very important to categorize anomalies. There are various categories and patterns of anomalies not only from a detection and perception standpoint but also from a planning standpoint.
- A challenge with machine learning is that we tend to generalize what we observe in the training data. The more you can discover in the tail (anomalies/edge cases), the more robust your models will be.
