dave-vedant / deep_learning_nanodegree-udacity

This repository contains different projects and deep learning concept notebooks. I mostly used PyTorch to develop ANN, RNN, CNN, GAN/DCGAN algorithms. I used AWS services such as Sagemaker, lambda, Restful API, EC2 and EMR during learning phase. 'Orca is deep diver dolphin, shows my honest approach to deep dive in the field of AI.

project.Orca

This is My Deep Learning Repository, where you can find different deeplearning approach such as ANN, CNN, RNN, GAN. I also include the relevant project and my understaning guides. I will share my personal notebook in future. My main motive is to share my approach, which I gain during my recarvation period. You can find web deployment project such as sentiment analysis and dog breed analysis for CNN.

Each adn every folder has thir relevant name so, Still you can search relevant thihg here from index.

1. Neural_Network Basic :

• Basic About Neural network fundamentals

2. Convolutional_Neural_Network:

• major topics : CNN form pytorch, style transfer, auto-encoder, projects(CIFAR -10), and weight initialization effects.

3. ML deployment- AWS.SageMaker

• major topics: Boston house project with batch transform, Boston House - Hyper parameter turing, Bostn House - model updation.

4. Sentiment Analysis :

• Major Topics : Neural Network approach for Sentiment analysis, Text feature extraction, detail architecture from scratch, noise reduction, Hypothesis testing, Model evaluation.

5. [Sentiment Web Application Deployment AWS - SageMaker-with Radom Forest:

• Major Topics : Sagemaker batch transform, Sagemaker Hyper parameter tuing, Sagemaker model updation, AWS lambda deployement

6. Project:

1. Bike Sharing :

• Predict Bike sharing requirement to solve business problem. Implement Shallow Neural network with pytorch. and get Training Accuracy : 97.3 % and validation accuracy : 93 %.

2. Dog_breed image classification :

• develope CNN algorithm from scratch to classify Image dataset. get 63 % accuracy on first try.
• Try to improve accuracy by using deep network, tuning parameters and also use different weight initialization appraoch.
• Due to gpu inefficiency, I can not get satisfied reply, then use pretrained models
• Try different pretrained networks for accuracy and get each result based on true positive and false positive testing.
• Find out best reply, and use transfer learning method to implement successful netowrk in application algorithm.
• Make visualization result as algorithm prediction output.

---

---

Repository Main Features.

---

---

---