Explainable AI — My Journey

Big data to Artificial Intelligence. Image courtesy of ‘readwrite.com’

The breathtaking pace of progress in recent years in Neural-networks and deep learning is transforming the world around us in profound ways.
Continuing on my path to reach the promised land of this field(Explainable AI XAI), I decided to learn it and advance my career.

A little bit about me. I currently work as a Senior/Lead Data Scientist in an IoT connectivity company located in San Francisco Bay area. My work currently revolves around building data science models for our IoT data and maintaining the big data pipelines. The insights and predictions, pave the way for data-driven decisions.

I was always fascinated by the way data scientists looked at any given sample of data and wanted to learn how differently they perceive data. So, I decided to gather as much knowledge as I can about data science and AI and apply it to the problems at work, building new prediction and classification models.

After carefully going through the resources available, I found the one that was perfect for my learning appetite. From deep learning fundamentals to AI practical implementations, I spent approximately 12–15 hours every week outside of 40 hours at work to complete a boot camp and it has transpired into great results already.

I started with the Fundamentals of Deep Learning of course! It was a six-month-long boot camp. It had an equal emphasis on understanding the theoretical foundations, as well as hands-on experience with real-world data analytics on the Google cloud platform.

With my prior experience as a data scientist, I was already clear on basic concepts in data science like EDA, Sampling, Bootstrapping, Cross-Validation, and the Geometrical background and intuition behind the major ML Regression and Classification and Clustering algorithms(Ensemble Methods, Random Forests, Support Vector Machines, Regularization, and Dimensionality Reduction Methods). Concepts like Bias-Variance Trade-off, model vs data complexity. I had to brush up my knowledge on hyperparameter tuning through grid-search, it really took its sweet time to register into me but it was worth it.

Working with the top data science libraries in Python and R at my workplace, I already had good momentum.

Practical Methods in Deep Learning; 6-month Bootcamp. I learned the concepts and practical application of Transfer Learning, Auto ML, Activation Functions, Loss functions, Back-propagation, and Optimizers. A couple of months into this boot camp, I was learning about top Neural Architectures like Convolutional Neural Networks, Recurrent Neural Networks ( LSTM & GRU), Attention models, and Transformers: BERT family, GPT-3, ELMO, etc. After working my way with the labs in Computer Vision: classification, segmentation, Neural Style Transfer, time-series data, Anomalies, and Fraud Detection, I was able to build the projects described below from scratch and deploy them in the cloud.

Eloquent Transformers -A cloud-based NLP application that harnesses the power of artificial intelligence and delivers content analysis and annotation. Given any arbitrary textual content: EloquentTransformer does an array of tasks including Parts Of Speech Tagging (Tagging the words in the first sentence with their parts), Dependency Graph, Lemmatization, NER, and more.


The project is live at https://eloquenttransformer.herokuapp.com/

Tree Pilgrims — I applied AI and computer vision techniques to identify the trees around us and bring out their beauty for the world to appreciate. The model distinguishes between two specific types of trees found in our neighborhood: the Weeping Willow, and the Peruvian Pepper tree. The pepper tree has a characteristic bark that distinguishes it, and of course the red pepper fruits. On the other hand, the weeping willow has its own unique look with the gracefully arched branches that dangle down towards the ground, and create a cozy arbor in its shade.


I wanted to go that extra mile and unmask these black-box AI models. I did this by using Explainable AI (XAI) using the LIME package. Here is an example of how the model decides among a set of images that it is indeed a Weeping willow.

A significant amount of time and effort in learning and implementing the various AI models, I have mastered much of Data Science and Deep Learning in considerable depth and thoroughness that goes way beyond most practitioners’ commonplace abilities in this field.

Creativity with AI, Data driven, Soccer player, Huge Manchester United fan.