< Overview > Web application which has the ability to detect the bird species with bounding box that uses Faster R-CNN deep learning model at backend. Created a Flask web application that communicates with a TensorFlow server hosted in Docker through gRPC communications protocol Has the ability to detect 3 bird species – Pica_pica, Periparus_ater and Erithacus_rubecula More than 2400 images were tagged for deep learning model to train using RenomTag Bootstrap and HTML/CSS was used to design the website Complete Project
Environmental Sound Classification
< Overview > An AI project that can help in the classification of sounds which can be used for crime scene investigations. The MLP (Multi Layer Perceptron) and Random Forest model has been used for the comparison to see which performs better UrbanSound8k dataset composed of 8732 tagged sound clips with a duration of less than or equal to 4 seconds were used to train the model Libra which helps in audio analysis with MFCC was being used for feature extraction MLP model performs best with an accuracy of 92.73 %, whereas RF model performs best with an accuracy of 61.11 %, indicating that MLP performs better than RF on audio or time series data. The model has been saved in .h5 (1.3 MB) and.joblib (150.6 MB) formats for MLP and RF respectively. Complete Project
Higgs Boson Detection using Accelerated Machine Learning
< Overview > The Random Forest and XGBoost model has been used in this project to train datasets in both CPU and GPU to classify Higgs boson signal from the background noise. Dataset contains almost 11 million simulated collison occurrences which was generated using Monte Carlo simulations. Both models were trained on the CPU and GPU (used RAPIDS on Nvidia RTX 3090) to determine how well each model performs in different environments In the RF and XGBoost models, the GPU is over 177 times and 300 times quicker than the CPU, respectively, allowing for significant time savings that can be leveraged to improve the model In terms of accuracy, the CPU performed slightly better than the GPU in both models when the same parameter was utilized Complete Project