Badges
Certifications
Work Experience
Machine Learning Engineer,
Wipro• January 2018 - Present
- Projects ● East West Bank •Build machine learning fraud detection system (Python, Logistic Regression, Decision Tree, Random Forest, SVM, Autoencoder), a complete end – to – end pipeline plan, design, and implementation. •Build semi and full automated Python jobs that daily scheduled to extract, transfer, load (ETL) and monitor data between network sharing files system and Azure SQL Database, and between Azure SQL Databases and tables. •Build SQL syntax streaming jobs utilize Azure Stream Analytics, and deploy them using Azure DevOps, to extract, transfer, load (ETL) and monitor data between Azure Cosmos DB and Azure SQL Database, and between Azure SQL Databases and tables. ● Apple •Build a complete NLP anomaly detection pipeline system (Python, PySpark, Scala, Spark, Hadoop, and Hive) based on statistical and deep learning model utilizing Airflow (schedule and monitor workflows) and Tableau, Incorta (dashboards visualization) •Data cleansing and transformation utilizing Python, Pandas and unsupervised clustering algorithms, such as PCA •Extract, transform, load (ETL) on big data utilizing Python, PySpark, Spark, Hadoop, and Hive •Installation and configurations Tableau server
Machine Learning Research Assistant
California State University, Fullerton• August 2016 - August 2017
Residual Squeeze VGG16 ○ Added the new Fire module for the first time to the VGG16 architecture ○ Added the Residual connections with shortcuts for the first time to the VGG16 architecture ○ Improved the training speed by 23.86% and reduced the size by 88.4% compared to the original VGG16, while maintained the accuracy, testing the model on the MIT Places365-Standard Dataset for image classification ● Residual Squeeze CNDS ○ Created a new Fire module inspired by the original paper SqueezeNet ○ Adapted the Residual connections with the new Fire module ○ Improved the training speed by 13.33% and reduced the size by 87.64% compared to the original Residual-CNDS, while maintained the accuracy, testing the model on the MIT Places365-Standard Dataset for image classification ● Novel CNN Architecture with Residual Learning and Deep Supervision ○ Deployed and tested the accuracy of the Residual-CNDS with two depths 8 and 10 convolutional layers on the MIT Places 205 and MIT Places365-Standard Datasets for image classification ○ Implemented the best method to add the residual learning connections with the shortcuts techniques ○ The Residual-CNDS neural network achieved the 3rd place in the MIT Places Leaderboard
Education
California State University, Fullerton
Computer Science, MS• August 2015 - June 2018
University of Basrah
Computer Science, BS• October 2001 - May 2005
Skills
hualkassam has not updated skills details yet.