↓Skip to main content

MMDT Applied Deep Learning Projects

Table of Contents

Myanmar Data Tech | Foundations of Deep Learning | A+ Grade | 2025
GitHub Repository → · A2C Capstone Project →

Overview

Completed a rigorous series of applied deep-learning assignments and benchmarks across diverse modalities: tabular data regression, low-resource NLP classification, computer vision model evaluation, and transformer-based geocoding.

Explore GitHub Repository

1. Singapore HDB Resale Price Prediction (Tabular MLP)

  • Trained multi-layer perceptron (MLP) architectures on Singapore public housing open data.
  • Engineered features including remaining lease duration, flat model categorization, and geographical proximity metrics.
  • Achieved $R^2 = 0.92$ and an RMSE of SGD 59,051, setting strong tabular deep-learning regression baselines.

2. Myanmar Location Name Classification (NLP)

  • Implemented 24 distinct NLP classification experiments across 6 neural architectures with 4 regularization and dropout schedules.
  • Systematically diagnosed representation bottlenecks and error distributions caused by extreme class imbalance across Myanmar administrative townships.

3. CNN Vision Benchmarking on Primary Dataset

  • Benchmarked five modern computer vision architectures: ResNet50, VGGNet16, InceptionV3, ConvNeXt, and EfficientNet.
  • Analyzed trade-offs between Top-1 accuracy, FP32 parameter volume, memory footprint, and forward-pass inference latency on target hardware.

4. Geospatial NLP Capstone

  • Designed and built the A2C Myanmar Address-to-Coordinates system, featured separately in detail on the A2C Project Page.

Tech Stack: PyTorch · Hugging Face Transformers · Scikit-learn · Streamlit · Kaggle GPU · Pandas · NumPy