An Introduction to AI for Chemical Engineers

Explain ai and machine learning by comparing rule-based and data-driven approaches; illustrate with a cat vs dog example, discuss data needs, processing power, and the potential to combine methods.

Explore how a fixed-volume gas experiment links temperature settings to observed pressure, introducing samples and the terms feature and target in a data science context.

Extend a simple temperature-based pressure model to multiple predictors by including gas type, and distinguish real numbers from categorical data such as oxygen, nitrogen, and helium.

Explore data-driven, mechanistic, and hybrid modeling to predict system behavior from experiments, balancing neural networks and physics insights, with non-ideal gas examples like van der Waals.

Explore how chemical engineering data include continuous numerical features and discrete categorical variables, along with images and texts, to analyze process patterns.

Visualize the lab data to relate temperature and gas pressure, then fit a linear model by estimating parameters a0 and a1 for the best predictive power.

Learn how to train a model by choosing and minimizing a loss function. Compare absolute error and squared error, and understand convex loss surfaces and gradient descent.

Compare data-driven and mechanistic models by showing how training data and physical insight shape form in chemical engineering. Consider trust, extrapolation, and how multiple predictors influence model choice.

Understand how input features such as molecular structure and physicochemical properties enable classification of compounds into toxicity risk groups, and distinguish classification from regression.

Explore artificial neural networks as a powerful data-driven modeling approach. See how simple neurons in networks approximate complex input–output mappings, using gas pressure and temperature as an example.

Artificial neural networks map inputs to outputs by stacking mathematical functions, making them data driven, with training data used to set parameters; hybrid models add mechanistic insights from physical sciences.

Explore the core idea behind training neural networks for chemical engineers: set a fixed architecture, initialize parameters, compute loss, and adjust parameters using gradient descent and derivatives to minimize error.

Explore how natural language and image data drive the boom of deep learning, fueled by large datasets like Wikipedia and ImageNet that enable training massive neural networks.

Learn how image inputs pass through a series of mathematical transformations to output a four-element vector with x, y coordinates for the bounding box, enabling object detection, segmentation, and classification.

Explore how language models process text by converting word sequences into numeric vectors, train transformations to classify inputs, and apply learned parameters to new documents.

Explore how decoder large language models like GPT and Lama function as generative language models, predicting next words from prompts and using temperature to control creativity.

Examine neural networks and deep learning for vision and language, from input to numerical representations and learned transformations, highlighting convolutions, self-attention, and transformer architectures.

Explore how cloud computing enables data storage, model training, and sharing. See how AWS, Google Cloud, and Azure enable saving and sharing trained models via the cloud.

Master AI and ML basics without coding, then explore low-code tools. Develop problem understanding and communication through Cs229, Kaggle, deep learning, and cloud computing, while exploring PyTorch or Keras.