Lets_learn_new_skills_together

---

Motivation

This repository was have a personal archive of State-of-the-art projects / codes in mostly Deep learning field. The codes contained are ready to be used in any industry right-away if well tuned. Tweaked a few parts of posts from original author in order to expedite the learning process in a logical way.

---

Table of contents

• Python
• R
• Contact

---

Python

• Deep Learning

  Generate fluent English text using transformers GPT2:

  A simple toy example to look at different decoder method provided by Transformer library. We look at different methods to see which decoder gerenates the most human-like languages when given texts.

  End-to-End Speech Recognition in Tensorflow:

  Build a simpel End-to-End Speech-to-Text model using librosa library. The model takes recordings of 10 different classes or words (data from Kaggle Speech Recognition Challenge), trains algorithm that is in Convolutional 1D, and predicts the sound in text.

  Image captioning with visual attention using Tensorflow 2.0:

  Build a simpel attention-based Image captioning (annotation) model using InceptionV3, pretrained on ImageNet, in Tensorflow. For better accuracy, Utilize "Transformer" tokenization when tokenizing the captions

  Prototypical Networks for Few-Shot-Learning in Pytorch:

  Basic walkthrough of Few-shot learning, that's been widely used in image classficiation with a limited training images. Build a model to classify Omniglot dataset (language characters) to correct classes. Using Image2vector CNN model.

  Transfer Learning in Pytorch:

  Transfer Learning complete model building by building CIFAR-10 model (VGG16 Pretrained on ImageNet). Freeze the last few layer to fit into CIFAR-10 tasks.

  f-AnoGAN (Image Anomaly Detection) in Pytorch:

  Use Unsupervised leraning method to solve the image anomayl detectcion widely used in medical fields. Craete a Fake image to compare with the real image. Has it's robustness where only a good volumnes of traning images are needed at the beginning, and it takes care of anomaly detection. Pytorch Implementation acheives over 99.5% accuracy in FP/TP and precision/recall.

  Source: Github

  Basic Pytorch Concepts in practice by building MNIST CNN model:

  An all-in-one Pytorch starting codeset, which can be applied to many of real-world image classification tasks

  Source: Sebastian Raschka Github

  AlexNet CIFAR-10 Classifier:

  AlexNet algorithm in Pytorch to solve CIFAR-10. Shows a complete step of model building (The CIFAR-10 dataset contains 60,000 32x32 color images in 10 different classes.[3] The 10 different classes represent airplanes, cars, birds, cats, deer, dogs, frogs, horses, ships, and trucks. There are 6,000 images of each class)

  Source: Sebastian Raschka Github

---

R

---

Contact

Created by @hyunjoonbok - feel free to contact me!