Kirankumar Ashokkumar Email & Phone Number

Fremont, California, United States

San Mateo County, California, United States

• Reduced network traffic by utilizing an additive regression model to predict anomalies in time seriesdata (0.88 RMSE)
• Built data collector and processor for event streams and worked on various data transformations
• Built an anomaly detector that identifies anomalies from the given 100+ features using a univariantand multivariant isolation forest

Merced, California, United States

• Collaborated with 4 other colleagues to develop a web application using the MERN stack that allows truckers to share relative information to the admin at the storage facility to increase efficiency
• Utilized a google maps API for truckers to send in real-time location to the admin's UI

Merced, California, United States

• Received funding to construct and analyze CNN models for detecting COVID-19.
• Compared AlexNet, ResNet, and a novel prediction-based model using Keras and Python
• Demonstrated that the proposed prediction-based model outperforms alternatives
• Presented findings at the annual UC Merced UROC symposium

Merced, California, United States

• Learnt the mathematical side of machine learning topics such as CNN and LSTM
• Created scripts to generate additional image data using image processing techniques (rotation, zooming, translating) to increase dataset size
• Tuned hyperparameters for deep learning models that classified items from Columbia Object Image Library, using Bayesian Optimization
• Explored various deep-learning techniques that can be used for stock market predictions
• Implemented and trained Lasso, Linear, Ridge, Net Elastic regressions, and LSTMs to predict the closing price of different companies using Python (0.90 RMSE)
• Compared R-squared and Mean Squared Error for each model to identify better performing model

Merced, California, United States

• Developed a machine learning pipeline to detect asteroids from raw observatory data to aid national defense with natural disaster detections
• Built using an architecture that combined CNN models (LesNet & Average Pooling) in PyTorch
• Demonstrated high performance of the architecture (97% precision and recall)