– Neural Network implementation from scratch and advanced model implementation in Tensorflow, hyperparameter tunining, regularization, and optimization techniques - – Convolutional Neural Networks: LeNet-5, AlexNet, VGG-16, ResNets, Residual Neural Network, Inception Network, MobileNet, EfficientNet, Localization & Detection Techniques, YOLO Algorithm, U-NET, Siamese Network, Neural Style Transfer – Sequence Models: Recurrent Neural Network, GRU, LSTM, Bidirectional RNN, Word2Vec, GloVe, seq2seq, Beam Search, Attention Model, Transformer https://coursera.org/share/4ccec80846c7c94bd591f18bf0c35b6a

– Analyzed fake import declarations data provided by Korea Customs Service – Built a machine learning model for detecting future fake import products – Provided 91.74% accuracy in detecting fake import declarations – Technology Used: Python, Scikit-learn, Pandas

– Introduced Perceptron as a binary classifier that should also be able to detect spam texts – Visualized real-time how perception works in 2D dimensions – Compared the built model with other ML architectures – Technology Used : Python, Scikit-learn, Pandas, Matplotlib, IPyWidgets

– Explored the reasoning capabilities of ChatGPT(mainly GPT 3.5) in Bengali – Increased Accuracy of Bengali from 30% to 59% by proposing our back-translation method – Technology Used: Python, OpenAI API

I found a classified security vulnerability in the KAIST system. For this I was awarded in the Introduction to Information Security course taught at KAIST. https://bit.ly/3KsXBtS