AI / ML

Machine Learning Foundations & Data Workflow

ML fundamentals, data acquisition and EDA, and feature engineering and preprocessing.

beginner

ML fundamentals

AI vs ML vs Deep Learning; supervised vs unsupervised vs reinforcement learning, batch/offline vs online learning, common ML challenges and the ML development life cycle.

beginner

Data acquisition & EDA

CSV, JSON, SQL, APIs and web-scraped data, feature types and data quality, univariate, bivariate and multivariate analysis.

beginner

Feature engineering & preprocessing

Scaling, encoding, ColumnTransformer and sklearn Pipelines, transforms, binning, mixed variables, missing values and outliers.

beginner

Supervised Learning - Regression

Predict a continuous target, learning not only how to fit a model but why its objective function and assumptions matter.

beginner

Simple Linear Regression

Line equation y = mx + b, geometric intuition of fitting a best line, residuals and the squared-error objective.

beginner

Multiple Linear Regression

Hyperplane intuition and matrix-form thinking, coefficient interpretation, and the effect of correlated predictors.

beginner

Gradient Descent for Regression

Cost surface and direction of steepest descent, learning rate, and batch, stochastic and mini-batch gradient descent.

intermediate

Polynomial Regression

Creating nonlinear decision curves using polynomial features while the model remains linear in parameters.

intermediate

Regression Metrics

MAE, MSE, RMSE, R² score and Adjusted R², choosing metrics based on business cost rather than convenience.

beginner

Bias-Variance & Regularized Regression

The bias-variance trade-off and L1/L2 regularized linear models.

intermediate

Bias-Variance Trade-off

Underfitting vs overfitting, high bias vs high variance, and how model complexity changes generalization.

intermediate

Ridge Regression (L2)

Penalty on squared coefficients, geometric intuition of shrinking coefficients, and the effect of lambda on bias and variance.

intermediate

Lasso Regression (L1)

Penalty on absolute coefficients, sparsity, and Lasso as embedded feature selection.

intermediate

Elastic Net

Combines L1 and L2 regularization using alpha and L1-ratio for sparsity with correlated predictors.

intermediate

Ensemble Learning

Voting, bagging, random forests, boosting and stacking.

intermediate

Voting Ensembles

Why combining diverse models can improve generalization; hard voting vs soft voting.

intermediate

Bagging

Bootstrap sampling and parallel learners for variance reduction.

intermediate

Random Forest

Bagged trees + random feature subsampling, out-of-bag score, GridSearchCV / RandomizedSearchCV and feature importance.

intermediate

AdaBoost

Sequential weak learners increasing focus on misclassified examples.

intermediate

Gradient Boosting

Stage-wise additive modeling with residual / negative-gradient correction.

intermediate

XGBoost

Boosted-tree objective and regularization intuition with second-order optimization.

intermediate

Stacking & Blending

Base learners and meta-learner using out-of-fold predictions to avoid leakage.

advanced

Clustering & Dimensionality Reduction

Centroid, hierarchical and density-based clustering plus the curse of dimensionality and PCA.

intermediate

K-Means Clustering

Centroid-based clustering objective, assignment and centroid-update steps, and choosing the number of clusters.

intermediate

Agglomerative Hierarchical Clustering

Bottom-up merging process, linkage concept and dendrogram interpretation.

intermediate

DBSCAN

Density-based clustering with core, border and noise points, eps and min_samples.

intermediate

Curse of Dimensionality

Distance concentration and sparse high-dimensional spaces, and motivation for feature selection and dimensionality reduction.

intermediate

Principal Component Analysis (PCA)

Rotate to directions of maximum variance using covariance matrix, eigenvectors and explained variance.

intermediate

Hyperparameter Tuning & Reliable Experiments

Search strategies and experiment hygiene for reproducible, leakage-free experiments.

intermediate

Search strategies

Train/validation/test split discipline, cross-validation, GridSearchCV, RandomizedSearchCV, Optuna and Bayesian optimization.

intermediate

Experiment hygiene

Put preprocessing inside pipelines, set random seeds, record dataset version and parameters, and use a simple baseline.

intermediate

Neural Network Fundamentals

The perceptron, multi-layer perceptrons, loss functions and backpropagation.

intermediate

Perceptron

Artificial neuron vs perceptron, geometric decision-boundary intuition and the perceptron update rule.

intermediate

Multi-Layer Perceptron (MLP / ANN)

Layer notation, forward propagation, and TensorFlow/Keras workflow for classification and regression networks.

intermediate

Loss Functions

Regression losses vs classification losses and loss as the scalar objective optimized by gradient descent.

intermediate

Backpropagation

Chain-rule flow from loss to each parameter, forward cache, and why gradients vanish or explode.

intermediate

Optimization, Activations & Regularization

Training deep networks: gradient descent, activations, initialization, regularization and optimizers.

advanced

Gradient Descent in Neural Networks

Batch, stochastic and mini-batch training, learning rate, batch size and vanishing/exploding gradients.

intermediate

Activation Functions

Sigmoid and saturation, Tanh, ReLU and dead-neuron issue, and Leaky ReLU, PReLU, ELU and SELU.

intermediate

Initialization

Xavier / Glorot initialization, He initialization for ReLU-family activations, and gradient stability.

intermediate

Regularization & Generalization

Early stopping, dropout, L1 / L2 regularization and weight decay, and batch normalization.

advanced

Optimizers

Momentum, Nesterov Accelerated Gradient, AdaGrad, RMSProp and Adam trade-offs.

advanced

Neural Network Hyperparameter Tuning

Learning rate, units, layers, batch size, dropout and optimizer with the Keras Tuner workflow.

advanced

Convolutional Neural Networks (CNN)

CNN intuition, the convolution operation, pooling, CNN architecture and backpropagation, augmentation and transfer learning.

advanced

CNN Intuition

Why dense ANNs are inefficient for images: local connectivity, receptive fields and parameter sharing.

intermediate

Convolution Operation

Filters / kernels sliding over input, feature maps and channels, stride and padding, and output-shape calculation.

intermediate

Pooling

Max pooling and downsampling, effect on spatial resolution and robustness.

intermediate

CNN Architecture & Backpropagation

LeNet-5 as a canonical architecture and backpropagation through convolution, max-pooling and flatten layers.

advanced

Data Augmentation & Transfer Learning

Geometric / photometric augmentation, ImageNet / ILSVRC pretrained models, feature extraction vs fine-tuning.

advanced

RNN, LSTM & GRU

Recurrent networks for sequence data, including LSTM, GRU and bidirectional variants.

advanced

Recurrent Neural Networks (RNN)

Hidden-state recurrence and forward propagation, sequence-to-label patterns and Backpropagation Through Time (BPTT).

advanced

LSTM

Cell state and hidden state, forget, input and output gates, and next-word prediction.

advanced

GRU

Reset and update gates, simpler state structure than LSTM, and trade-offs between GRU and LSTM.

advanced

Deep & Bidirectional Recurrent Networks

Stacked / deep RNNs, LSTMs and GRUs, and bidirectional RNN, BiLSTM and BiGRU.

advanced

Seq2Seq, Attention & Transformers

Encoder-decoder sequence-to-sequence, the attention mechanism, and the full transformer architecture.

advanced

Encoder-Decoder / Sequence-to-Sequence

Encoder compresses the input sequence into a representation, decoder generates the output, and the bottleneck problem of fixed-size context vectors.

advanced

Attention Mechanism

Attention scores, weights and context vector; Bahdanau (additive) vs Luong (multiplicative) attention.

advanced

Transformer Foundations

Self-attention with Query, Key and Value, scaled dot-product attention, and multi-head attention.

advanced

Positional Encoding & Layer Normalization

Sinusoidal positional encodings and layer normalization versus batch normalization.

advanced

Transformer Encoder

Token embedding + positional information, multi-head self-attention, residual connections + layer normalization, feed-forward network, stacking encoder blocks.

advanced

Transformer Decoder & Inference

Masked self-attention / causal attention, cross-attention to encoder output, and autoregressive token-by-token inference.

advanced