Video Series: Hands-On AI—Part 5: Conclusion
The combination of these videos provides you with knowledge on how image data preprocessing can be used for cleaning and augmenting transformations on a dataset and why it matters.
Hey. Karl again, and this is the final episode in the Hands-On AI series. In this series of videos, we have gone over many common functions and techniques used in data preprocessing and augmentation–and more importantly, why we take these steps to ensure an accurate model. Take our practical use case of showing specific images based on a general mood. Even with a clean and orderly data set, there are still steps to complete before that information is ready to be ingested by a machine-learning algorithm.
In the second episode, we covered data cleaning. Remember that data cleaning is important so that our machine-learning model can process the data. In our example, we used images which programs process as arrays of numbers. We need to make sure there aren't any outliers in our data set to improve accuracy and prevent exploding gradients. In the third episode, we went over augmentation. Data augmentation is important, so that our model can take our training data set and apply it to real-world scenarios.
Finally, we showed how these different augmentation techniques are implemented in Karas, our framework of choice for this Hands-On AI series. Feel free to download the Jupyter* Notebook, so that you can use them in your own projects. Thanks for watching. Hopefully, you've enjoyed this series. Be sure to check out the links to read the article associated and to follow along on your own. Also, be sure to visit the Intel® AI Academy to learn more about developing your own AI models and applications.
Product and Performance Information
Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.