Exploring Intelligence: Neural Networks, Machine Learning, and Deep Learning

The following activities reflect my strong and consistent engagement in the fields of Neural Networks, Machine Learning, and Deep Learning. The work spans practical implementation, experimentation, and application of AI methods to real-world problems, contributing to both foundational understanding and advanced innovation.


Neural Networks

Neural Network Layers

Re-generated image by: M A Hoque



  1. Implemented shallow neural networks from scratch to understand forward and backward propagation.
  2. Utilized sigmoid, tanh, and ReLU activation functions to study non-linear transformation in layers.
  3. Applied neural networks to solve classification problems on datasets like MNIST.
  4. Tuned hyperparameters to evaluate learning rates, number of epochs, and layer sizes.
  5. Used dropout regularization to prevent overfitting during training.
  6. Compared optimizers like SGD, Adam, and RMSprop for training stability.
  7. Created custom neural architectures using TensorFlow and Keras.
  8. Visualized training curves using TensorBoard to interpret model performance.
  9. Applied early stopping techniques to halt training once validation performance declined.
  10. Experimented with batch normalization to stabilize and speed up learning.


Machine Learning

Machine Learning contour graph

Graph generated by: M A Hoque



  1. Applied linear regression for housing price prediction.
  2. Used logistic regression for binary classification tasks.
  3. Built decision tree models for data classification.
  4. Implemented SVMs to distinguish between overlapping class boundaries.
  5. Performed data preprocessing including normalization and handling missing values.
  6. Conducted feature selection to improve model generalization.
  7. Used PCA for dimensionality reduction in high-dimensional datasets.
  8. Evaluated models using accuracy, precision, recall, and F1-score.
  9. Performed k-fold cross-validation for robust model assessment.
  10. Employed grid search and randomized search for hyperparameter optimization.
  11. Built ensemble models using random forests and gradient boosting.
  12. Addressed overfitting using regularization techniques like L1 and L2.
  13. Implemented spam detection models using Naive Bayes.
  14. Analyzed ROC and AUC to measure classifier performance.
  15. Designed and tested pipelines using scikit-learn for reproducibility.


Deep Learning

Object detection

Image identification by: M A Hoque



  1. Developed convolutional neural networks (CNNs) for image classification.
  2. Applied transfer learning with pre-trained models such as VGG16 and InceptionV3.
  3. Built a face recognition system using deep learning embeddings.
  4. Designed a self-driving car simulator using CNNs to predict steering angles.
  5. Implemented data augmentation to improve generalization in image tasks.
  6. Used ResNet architecture to explore residual learning benefits.
  7. Built RNN and LSTM models for time series prediction.
  8. Trained NLP models for sentiment analysis using word embeddings.
  9. Applied attention mechanisms in sequence modeling.
  10. Used GRU units to simplify RNN architectures while preserving sequence memory.
  11. Fine-tuned MobileNet for lightweight image classification.
  12. Integrated model tracking using MLflow during training experiments.
  13. Applied batch and layer normalization for network stabilization.
  14. Monitored training and validation loss to avoid underfitting and overfitting.
  15. Deployed models using TensorFlow Serving.
  16. Developed object detection pipelines using YOLO architecture.
  17. Implemented image segmentation using U-Net architecture.
  18. Built text generation models using character-level RNNs.
  19. Used autoencoders for dimensionality reduction and anomaly detection.
  20. Applied GANs to generate synthetic images.
  21. Explored SHAP and LIME for model explainability.
  22. Participated in DeepLearning.ai and Coursera specializations.
  23. Followed best practices in model reproducibility and version control.
  24. Evaluated deep learning models using confusion matrices and heatmaps.
  25. Contributed to open-source deep learning repositories and forums.


YOLO Ultralytics & NVIDIA CNN-based Self-Driving Car


YOLO Forest fire detection

Forest fire detection using YOLOV12 by: M A Hoque



  1. Explored YOLO (You Only Look Once) for real-time object detection, implementing models to detect multiple objects in images and videos with high accuracy and speed, including applications like forest fire and brain tumor detection.
  2. Trained custom YOLOv12 models using annotated datasets tailored for specific object detection tasks.

Brain tumor detection

Brain tumor detection using YOLOV12 by M A Hoque



  1. Optimized YOLO models for real-time performance on low-resource devices by experimenting with lightweight architectures and pruning for efficient inference on mobile and embedded systems.
  2. Evaluated YOLO model performance using standard object detection metrics like mAP and IoU to assess precision and recall.
  3. Developed a YOLO-based pipeline integrated with OpenCV for real-time surveillance systems, enabling video streaming and object detection such as car registration number recognition.

Car license number detection

Car registration number detection




  1. Designed a self-driving car simulation using CNNs trained on driving behavior data to model autonomous navigation in a virtual environment.
  2. Implemented lane detection and steering angle prediction using CNNs trained on simulated road images for accurate autonomous navigation.

YOLO Forest fire detection

Self-Driving Autonomous Car





  1. Integrated real-time inference in a driving simulator by deploying trained CNN models to evaluate performance during live driving scenarios.
  2. Experimented with end-to-end deep learning for autonomous driving by implementing neural networks that map raw input images directly to steering commands.
  3. Conducted performance benchmarking and fine-tuned the self-driving CNN model by adjusting hyperparameters and architecture to enhance generalization and robustness across diverse driving conditions.
Road lanes detection. Steering Angle prediction

Self-Driving car simulation. Lane detection and steering angle prediction






Leave Your Feedback

sap.hoque@gmail.com

+1 473 607 1949

Website Icon www.HoqueAi.com

© 2025 HoqueAI. All rights reserved. | Privacy Policy