Josiah Samuel, advisory software engineer, IBM
I recently attended the OpenPOWER and AI Workshop at the Indian Institute of Technology Delhi. This workshop gathered 30 students to learn about IBM and its work with Artificial Intelligence.
I was able to offer these students hands-on sessions discussing PowerAI, SnapMl and Machine Learning. One portion of the workshop focused on walking through a problem statement. This workshop’s statement was: “How to make a quick prediction whether a credit amount can be sanctioned or not.” After explaining the assignment, the students were taught how to do explanatory data analysis using the Matplotlib library. These charts showed the co-relation between various attributes. Students were taught how to convert raw data into a format machine learning algorithms can understand. All along, the students were allowed to try on their own based on the provided Power8 setup.
Other tools that were used to solve this problem included the Scikit-learn’s logistic Regression API to train the model, using a small dataset which shows low accuracy. This allowed the students to view the metrics. Students learned that the more the dataset was increased, the more accurate the data became.
Following this, we contrasted the Scikit-learn’s to SnapML. SnapML can perform ML training on large datasets with at least a 10x decrease in training time compared to Scikit-learn’s training time with no compromise on the model’s Accuracy.
It was an incredible experience to share my work with the IIT Delhi students and walk them through a real-life scenario.