2 minute read

Basic Machine Learning & Deep Learning, Word Embedding, CNNs, RNNs, LSTM and Transformer

Throughout my E2E Deep Learning course journey, I realized that my foundational knowledge could have been more robust. Facing this challenge pushed me to dive deeper into the core concepts of machine learning and NLP, ensuring I didn’t just scrape the surface but truly understood the intricacies. This visual map became my go-to guide, helping me grasp complex ideas and connect theoretical principles with real-world applications. Studying these concepts took me three weeks as my semester started, so I’ve tried to spend at least 3 hours daily.

Each section represents a stepping stone, helping me overcome obstacles and bridge knowledge gaps I hadn’t anticipated. From fundamental machine learning concepts to the most advanced architectures, this process solidified my understanding and prepared me to tackle more sophisticated NLP challenges.

I studied these concepts in Korean 🇰🇷 as part of building my foundational knowledge, using the fantastic resource from Wikidocs. It was a challenging yet gratifying process that reminded me that facing obstacles is part of growth. I’m excited to continue deepening my knowledge in this field and applying these skills to real-world NLP challenges! 🔥 I will be studying further in English and posting those concepts, so please stay tuned!

Here’s a breakdown of the topics I tackled during this deep dive:

1. Basic Machine Learning Concepts:

  • Logistic Regression
  • Linear Regression
  • Gradient Descent (Batch, Stochastic, Mini-Batch)
  • Cross-Entropy Loss and Cost Functions
  • Overfitting, Regularization (L1, L2), and Generalization Techniques
  • Bias-Variance Tradeoff
  • Decision Trees and Random Forests

2. Deep Learning Architectures:

  • Fully Connected Neural Networks
  • Activation Functions (Sigmoid, ReLU, Leaky ReLU, Tanh)
  • Backpropagation and Optimization Techniques (Adam, RMSprop, SGD)
  • Weight Initialization and Batch Normalization
  • Convolutional Neural Networks (CNNs):
    • Convolution Layers, Filters, and Pooling Layers
    • Applications of CNNs in Image Recognition and NLP tasks

3. Recurrent Neural Networks (RNNs) and Variants:

  • Architectures of RNN, LSTM, and GRU models
  • Sequence Processing and Time-Series Data
  • Exploding and Vanishing Gradient Problems in RNNs

4. Word Embeddings:

  • Architectures of Word2Vec, GloVe, ELMo, and FastText
  • Embedding Layers for Dense Vector Representation
  • Applications of Embeddings in NLP

5. Transformers:

  • Self-Attention and Multi-Head Attention
  • Sequence-to-Sequence Models with Attention
  • Transformers and their Role in NLP
  • Positional Encoding, Multi-Head Attention
  • Position-wise FFNN, Residual Connections, and Layer Normalization



The Full Image file can be downloaded here on my GitHub! I have also added the link for each image below.

Deep Learning Architecture

📚 Download Here: Deep Learning Architecture Tree



Word Embedding, RNN, and CNN

Word 2 BVec. Keras Embedding Layer, ELMo, Glove. FastText, and Keras application

📚 Download Here: RNNs & CNNs



The architecture of RNN, Transformer, and Attention Mechanism

Seq2sec, Attention, Dot-Product Attention, Transformer, Multi-head Attention, and Position-wise FFNN

📚 Download Here: Transformers



The Basic Deep Learning

Activation Function- Sigmoid, ReLU, Tanh, and Softmax, Perceptron, Loss Function, and Optimizers

📚 Download Here: Deep Learning



The Basic Machine Learning

Linear Regression, Logistic Regression, Vector and Matrix, and Softmax Regression

📚 Download Here: Machine Learning



Leave a comment