Statistics & Probability for Data Science:

Theory:

• Descriptive stats: mean, variance, skew, kurtosis
• Probability basics
• Distributions (Normal, Bernoulli, Binomial)
• Correlation vs causation
• Confidence intervals, hypothesis testing

Hands on:

• Distribution visualizations
• t-test, chi-square test on sample data
• Correlation heatmaps and interpretation

Data Analysis, Feature Engineering & Preparation:

• Data cleaning workflow
• Handling missing values & outliers
• Feature scaling & encoding
• Train-test splits
• Leakage and good ML practice
• Handling NaNs in Pandas
• Encoding categorical variables (OneHot, Label)
• Normalization & standardization
• Pipeline building in scikit-learn
• Mini-project: full cleaning workflow

Supervised Learning:

• ML building blocks
• Bias–variance tradeoff
• Cross-validation
• Regularization (Lasso, Ridge)
• Algorithms:
  o Linear regression
  o Logistic regression
  o KNN
  o Naïve Bayes
  o Decision trees
  o Random forest, Gradient Boosting (intro)
• Build regression + classification models in sklearn
• Model evaluation metrics: RMSE, MAE, Accuracy, Precision, Recall, F1
• Hyperparameter tuning with GridSearchCV

Mini-project: Prediction task (housing prices or churn)

Unsupervised Learning:

• Clustering concepts
• K-Means algorithm
• Hierarchical clustering
• Dimensionality reduction: PCA theory & intuition
• Elbow method, silhouette score
• Apply K-Means on a customer segmentation dataset
• Hierarchical clustering dendrogram
• PCA decomposition & visualization

ADVANCED DATA SCIENCE & DEEP LEARNING

Goal: Build modern applied ML skills with neural networks, deep learning, vision, transformers & BERT fine-tuning.

Neural Network Foundations:

• Neuron, weights, biases
• Activation functions
• Loss functions
• Gradient descent & backpropagation (conceptual)
• Overfitting & regularization in neural nets
• Build a small neural net from scratch in NumPy (forward pass only)

PyTorch Essentials

• Tensors
• Autograd
• Datasets/Dataloaders
• Optimizers
• Build a classifier for MNIST or a tabular dataset
• Custom training loop (forward → loss → backward → step)
• Saving/loading models
  Deep Learning Architectures
• CNNs — convolution, filters, pooling
• RNNs, LSTMs, GRUs (overview only)
• Regularization in deep nets: dropout, batch norm
• CNN for image classification (CIFAR-10 or fashion-MNIST)
• Compare performance with data augmentation

Computer Vision Deep Dive

• Modern vision models
• Transfer learning
• Feature extraction vs fine-tuning
• ResNet, EfficientNet (high-level)
• Fine-tune a pretrained ResNet on a small custom dataset
• Implement augmentations using torchvision

Transformers & Attention

• Why RNNs fail
• Self-attention mechanism
• Encoder–decoder design
• Transformer block anatomy
• Positional embeddings
• Overview of BERT, GPT, Vision Transformers
• Attention mechanism in PyTorch

NLP with BERT Fine-Tuning

• Tokenization, subword embeddings
• Pretraining vs fine-tuning
• Task types: classification, NER, QA
• Fine-tune BERT (HuggingFace) for text classification
• Evaluate & interpret embeddings
• Export model for inference

Bonus Content

• Intro to LLMs beyond transformers
• Vector databases & retrieval
• MLOps foundations
• Model interpretability (SHAP/LIME)
• Reinforcement learning (very high level)