This repo contain all my Deep Learning work completed by me for academic, self learning, and hobby purposes. Presented in the form of iPython Notebooks markdown files. Out of many framework avaliable for Deep Learning I have build my specialization working on the following frameworks.
-
- Built a CNN and trained over 65k fruit images to predict the class of fruit from a set of 131 classes.
- Model is built on PyTorch along with the implementation of techniques like Data augmentation, Batch normalization, Learning rate scheduling, Weight Decay, Gradient clipping to achieve the best results.
- The tensors are trained and evaluated on GPU using PyTorch built-in CUDA library to build the model.
- Inference - Achieved model accuracy of 98% under 5 mins undermining the power of all techniques.
-
CFAR-10: The dataset contains over 60,000 images belonging to 10 classes. I have developed the following neural networks to evaluate their performance.
- Built Feed Forward neural network(ANN) and achievied an accurcy of 48%.
- Built Convolutional Neural Network(CNN) and improved the accuracy till 75%.
- Applied technique like Data normalization, Data augmentation, Batch normalization, Learning rate scheduling, Weight Decay, Gradient clipping...etc to further improve the model accuracy to 90%. You can access my notebook here
-
Wheat Seed Dataset: The datasets involve the prediction of species given measurements of seeds from different varieties of wheat. I build a logistic regression model and achieved an accuracy of 78% under 15 epochs.
- GTSRB Dataset: The dataset contains over 39k images and over 40 classes. I have to build a neural network with CNN architecture using Tensorflow and applying techniques like image augmentation to achieve accuracy of 85%.
- Telco Customer Churn: ...In progress
- CNN Model - Age, Gender, Ethnicity:...In progress
- Twitter Disaster Tweets: The dataset involves processing the text to predict if a tweet signifies disaster information or not. I build the model using the NLTK lib for text processing and TensorFlow for building up the neural network. I have applied regular expression, stopwords, tokenization, pad sequences to build the model and implemented it using TensorFlow.
- NLP — Detecting Fake News On Social Media: The program help in identifying news articles programmatically if a news article is Fake or Not. I have tried comparing the techniques - Bag of words and TF IDF while approching to solve the problem and compared their accuracies.
I like writing and love to explain my work in layman's language. Through these blogs, I reach out to the general audiences who are not familiar with these technologies and want to learn in this area
- 7 Best Techniques To Improve The Accuracy of CNN W/O Overfitting
- Training Convolutional Neural Network(ConvNet/CNN) on GPU From Scratch
- Training Feed Forward Neural Network(FFNN) on GPU — Beginners Guide
- Logistic Regression With PyTorch — A Beginner Guide
- PyTorch - Training Fruit 360 Classifier Under 5 mins
- Deep Learning for Beginners Using TensorFlow
- Fake or Not ? Twitter Disaster Tweets
- NLP — Detecting Fake News On Social Media
- Building Recommendations System? A Beginner Guide
In progress will update soon
I also dabble in all other technology. You can access by complete portfolio here
If you liked what you saw, want to have a chat with me about the portfolio, work opportunities, or collaboration, shoot an email at gurjeet333@gmail.com