Master On-Device ML & AI for Mobile Apps — 2025 Edition

Explore how machine learning uses data and algorithms to find patterns and make predictions, training models that recognize dog breeds or predict salaries.

Explore the three main types of machine learning: supervised, unsupervised, and reinforcement learning, along with dataset, training data, testing data, label, and features.

Explore supervised learning and distinguish regression and classification by predicting a continuous salary from experience and certifications, or a categorical gender from height and weight.

Explore unsupervised machine learning with unlabeled data to uncover patterns through clustering, and understand reinforcement learning as trial-and-error learning guided by rewards and penalties.

Explore how deep learning uses artificial neural networks to recognize patterns in data, inspired by the brain. Learn how input, hidden, and output layers enable classification and regression tasks.

Understand how a neural network predicts house prices by training on a dataset of area and rooms, using input, hidden, and output layers, weights, and epoch-based learning.

Learn how neural networks classify images by converting images to 1D input arrays, training with backpropagation, and using an output layer sized to the number of classes for final predictions.

Explore the basics of machine learning, including supervised and unsupervised learning, then dive into deep learning with neural networks, feedforward, backpropagation, learning rate, loss function, and dropout to prevent overfitting.

Discover Google Colab to run Python code in your browser with zero installation, use notebooks with code cells and text cells, and access free GPUs for training machine learning models.

Learn Python basics for training machine learning models in mobile apps. Explore variables and data types like numbers, string, list, tuple, and dictionary in Google Colab notebooks.

Learn how to create and delete Python number variables, and explore integers, long, floats, and complex numbers with real and imaginary parts, including type inspection.

Declare Python strings with single quotes, print them, and use indexing and slice notation with non-inclusive end indices to extract substrings; repeat, and concatenate with plus.

Explore Python lists as a versatile, compound data type that holds diverse items in square brackets, supports indexing, slicing, appending, counting, inserting, removing, sorting, and reversing.

Explore python tuples as read-only lists in parentheses, contrast with lists in square brackets, and learn dictionaries with unique keys in curly braces for key-value pairs and value access.

Explore conditional statements and loops in Python, including if, elif, else, while and for loops, list and string iteration, and taking user input with the input function.

Open, write, and read files in Python to save machine learning models in formats like tflite, manage file modes, and safely close and inspect file content.

Learn how numpy provides large multi-dimensional arrays and matrices, along with high-level mathematical functions for model training, including creating arrays, using the np alias, and inspecting shape, size, and dtype.

Create and manipulate numpy arrays using zeros, ones, random, and full to generate arrays with specific values or random values. Reshape and flatten arrays for training models and dummy data.

Explore unary and binary NumPy operators, compute min, max, and sum on arrays, perform element-wise operations, and calculate dot products.

Learn to multiply two n-dimensional arrays using numpy dot or matmul, then sort a 2d array with numpy sort along rows or columns.

Explore how pandas enables data manipulation and analysis in Python, demonstrate creating data frames from lists and dictionaries, and print tabular data for machine learning workflows.

Learn to load a csv into a pandas data frame with read_csv, uploading nba.csv in colab, and inspect 458 rows and 9 columns of nba players for machine learning.

Handle missing values in datasets with pandas by identifying nulls using isnull, filling values, dropping rows, and checking the data frame shape.

Explore matplotlib, a Python visualization library for 2D plots of arrays, using plt to plot, bar, and scatter charts; learn via Google Colab notebooks and practical plotting examples.

Load images in python with matplotlib, read with imread, display with imshow, and inspect the resulting numpy array (ndarray) and its shape for data visualization.

Learn TensorFlow and TensorFlow Lite basics, declare and manipulate variables and constants with tf.variable and tf.constant, inspect shapes and dtypes, and use assign to update values in Colab.

Learn how to inspect and reshape tensors in TensorFlow Lite, determine rank and size, access specific elements, cast to numpy, and perform element-wise operations.

Demonstrate matrix multiplication and transposition in TensorFlow using tf.matmul and transpose to align shapes. Show typecasting with tf.cast and constructing ragged range tensors with varying row lengths.

Explore core tensor operations in TensorFlow, using tf.reduce_mean to compute means of TensorFlow Lite variables, including 1D and 2D tensors with axis-based averaging.

Generate tensors with random values in TensorFlow using random.normal and random.uniform, control shape, mean, stddev, and data type, set a seed, then add and concatenate tensors and apply argmax/argmin.

Learn how TensorFlow checkpoints save and restore thousands of variables to preserve model training progress, enabling recovery and rollback during on-device ml workflows.

TensorFlow Lite enables on-device machine learning for mobile apps by converting models to the TF Lite format and running them with an interpreter, delivering low latency, privacy, and offline capability.

Train a simple regression model using TensorFlow and Keras, prepare data with NumPy, and convert the model to TensorFlow Lite for mobile apps.

Test the trained model to verify predictions near 2x-1, then convert it to TensorFlow Lite and save as linear.tf lite for download.

Review the end-to-end training process. Import libraries, build a one-neuron neural network, set optimizer and loss, train over epochs, test with a simple dataset, and export to tflite.

Train a regression model to predict vehicle fuel efficiency using data collection, normalization and one-hot encoding, train with neural networks, evaluate, and convert to tflite for on-device mobile use.

Train a real-world regression model to predict fuel efficiency (mpg) using the auto mpg dataset from Kaggle, featuring cylinders, displacement, horsepower, weight, and more.

Load and prepare a csv dataset in pandas, train a fuel efficiency prediction model with TensorFlow and Keras, and convert it to tflite in Google Colab.

Identify and count missing values using is.na and sum, noting six missing in displacement and horsepower. Then replace with zero or drop rows using dropna before training.

Apply one hot encoding to the origin column, creating USA, Europe, and Japan features and removing the original column to improve model training accuracy.

Split the dataset into 80% training and 20% testing with dataset dot sample and drop. Describe the train data and separate miles per gallon label with train dataset dot pop.

Normalization scales numeric features to a common range before model training, preventing area from dominating; use max-based scaling, as shown with area and rooms, noting there are different normalization types.

Define a Python function to normalize each data column by subtracting the mean and dividing by the standard deviation, preparing training and test data on a common scale.

Build a TensorFlow Keras sequential neural network to predict fuel efficiency, using two 64-neuron hidden layers and a final single neuron, with data normalization and epoch-based training.

Test and evaluate your trained model using a 20% test split, compare predictions with actual values, and measure loss, MAE, and MSE, then convert to tflite for mobile apps.

Train a real-world regression model on a Kaggle dataset using Colab, apply one-hot encoding and normalization, split 80/20, then evaluate and convert to tflite for on-device use.

Collect data for house price prediction, apply normalization and one-hot encoding, train a neural network, evaluate on testing data, and convert the model to tflite for on-device use.

Acquire a Kaggle dataset to train a house price prediction model, using the california dataset with features like longitude, latitude, house age, total rooms, total bedrooms, and ocean proximity.

Load the housing.csv in Colab with pandas and TensorFlow, handle missing values, one-hot encode ocean proximity with get_dummies. Split data 80/20 for training and testing.

Normalize the dataset using mean subtraction and standard deviation, then train a neural network with three hidden layers for house price prediction and evaluate with mean squared error.

Retrain a house price prediction model after converting to tflite, add layers and neurons, and evaluate training versus testing performance to identify overfitting.

Explore image classification by training a model to recognize dog breeds and other objects, then convert it to TensorFlow Lite for use in mobile apps.

Explore data collection methods for training a computer vision model, including datasets, web scraping, and augmentation. Ensure varied object sizes and 50 to 500 images per class.

Learn how to find ready-to-use image datasets for training fruit and vegetable classifiers, using Kaggle to locate datasets and inspect their train, test, and validation folders.

Unzip the downloaded dataset, review train, validation, and test folders, select six classes (apple, banana, watermelon, onion, lemon, potato), create a dataset folder, and zip it for training a classifier.