If you read the first installment of this series introducing our Intelligent Process Automation product modules, you’ll know that we’re all about helping you develop machine learning solutions specific to your business’ unstructured content challenges—no generic models here.
One of the greatest difficulties with developing deep learning models, however, is that they’re often a black box. Which, depending on your use case, just doesn’t fly. How can you trust your model correctly identified material changes when you don’t understand why or how it reached that decision?
We get that. So we’re leading the charge to incorporate AI explainability into our customizable machine learning system. Once you’ve finished labelling your data and training a model with Teach, you can seamlessly move into our interface designed to help you better understand what your model is thinking—Review. (P.S. We were named a Gartner Cool Vendor in Data Science and Machine Learning for this!)
How it works
Review is built for your SME and data science teams, allowing them to peek into why your model is making certain decisions through various modules on the dashboard, including (for a classification task):
- F1 scores
- ROC curve
- Confusion matrix
- Precision vs. recall curve
- A summary of all the above target metrics
For each of these metrics, Review will display specific examples that led the model to believe that was the correct decision. You can then quickly scroll through and make corrections to finetune your model by removing bad examples.
You can also test the model with cases that were not in your training or testing dataset using the “Try it Out” feature. Click to see which words were most important for your model’s decision making process, and then determine whether you need to include other examples to help it generalize better, or if it’s ready to ship out to your company’s main workflow.
We believe that AI explainability is imperative for not only developing robust machine learning solutions, but also to assist your SMEs and data science teams to successfully integrate and support their efforts through clear metrics. Review aims to meet these goals, and seamlessly feeds in from our Teach module (introduced in Part I of our product spotlight series). Stay tuned next month to learn about more features in Indico’s Intelligent Process Automation product modules!
Want to see our Review module for yourself? Click here to set up your account.
