34 courses from zero to deployed: foundation for everyone, data tracks picked by your dataset, plus the deployment gap.
Python from zero inside Colab and Jupyter. Variables, loops, NumPy, and pandas, ready for machine learning.
Explore, clean, visualize, and prepare real data. Descriptive statistics, distributions, correlation, missing values, encoding, scaling, and train test splits done right.
Confusion matrix, precision, recall, F1, threshold tuning, cross validation, and hyperparameter search.
Neuron to network, backprop animated, epochs, batches, learning rate, loss curves, and overfitting.
Classic computer vision with OpenCV: thresholding, filtering, edges, morphology, and contours. Learn on MNIST and CIFAR-10, then real Malaysian images. Hands-on checkpoint on PlantVillage (CC0) crop leaf disease.
CNN internals, data augmentation, transfer learning with ResNet, MobileNet, and EfficientNet on Imagenette, and object detection with YOLO and the two-stage R-CNN family on the Malaysia road-sign dataset.
Cleaning Malay and mixed language text, tokenization, TF-IDF, embeddings, sentiment analysis, and fine-tuning multilingual BERT.
Frame extraction from your own phone clips, optical flow, action recognition, and LSTM or GRU on UCI HAR sensor data. Malaysia angle: Bahasa Isyarat Malaysia (BIM-SSD).
Feature engineering and selection, XGBoost and LightGBM, and SHAP explainability for real tabular problems.
Teach agents to learn from reward. Markov decision processes, Q-learning, and where reinforcement learning fits in the ML landscape.
Handle data at scale. Distributed storage, MapReduce, Spark, and parallel processing for datasets too big for one machine.
Save and load models, build a FastAPI service, connect to a web app, cloud free tiers, and Docker basics.
Quantization FP32 to INT8, pruning, ONNX, TensorFlow Lite, latency vs accuracy, batch inference, caching, and monitoring.
Bring everything together: take a Kaggle dataset from a raw CSV to a deployed model, and earn your overall EpochLab certificate.
What it means for a model to learn, plus loss and gradient descent. The intuition before any algorithm, capped by a hands-on checkpoint. Free.
Fit a straight line to predict a continuous number. The classic first ML algorithm.
Turn features into a probability for a yes or no decision. A fast, interpretable classifier.
Regularized regression that curbs overfitting, and Lasso can select features for you.
Classify a point by the labels of its nearest neighbors. Simple, no real training.
A fast probabilistic classifier that shines on text and spam detection.
Learn readable if-then rules from data, the building block of forests and boosting.
An ensemble of trees that vote together for a strong, robust tabular prediction.
Build trees that fix each other's errors for top tabular accuracy (XGBoost, LightGBM).
Find the widest margin between classes, with kernels for non-linear boundaries.
Group unlabeled data into k clusters by nearest centroid. The go-to first clustering method.
Build a tree of nested clusters (a dendrogram) and cut it wherever you like.
Density-based clustering that finds any-shape clusters and flags outliers as noise.
Compress many correlated features into a few directions of greatest variance.
How CNNs read images: convolution, filters, pooling, and the feature maps behind modern computer vision.
The feedforward neural network: neurons, layers, activations, and backpropagation, the base of all deep learning.
Networks with memory for sequences and time-series, and how LSTM gates beat the vanishing gradient.
Self-attention and the architecture behind modern NLP and large language models.
Unsupervised encoder-decoder networks for compression, denoising, and anomaly detection.
Soft, probabilistic clustering with the EM algorithm, for overlapping and elliptical groups.
Learn by dataset. Tell us what data you have: image, text, video, or tabular, and we map the exact course path for you.