2026 ASEE Annual Conference & Exposition

In this complete evidence-based practice paper, we detail the full workings of our class, Introduction to Machine Learning. Through our ten-week course, learners create several types of Neural Networks, including their own customized architecture for an application of their choice, all without requiring any enrollment prerequisites.

Since the explosion of Neural Networks and Large Language Models (LLMs), Machine Learning (ML) skills such as understanding neural network architecture, creating and deploying models, data processing and analysis, Python for ML, and evaluating ML models have been highly sought-after by employers, researchers, and students alike. An IMF survey of millions of jobs in the US found postings requiring these skills jumped from less than 0.01% of all jobs to over 2%, and that these skills were correlated with wage increase four times more than learning any other skill.

Despite the societal demand for ML professionals, resources to learn ML are often difficult to understand or inaccessible to many interested learners, especially first-year students. We find the commonly-accepted approach to teaching ML problematic: first, learners are exposed to the mathematics underlying ML, then are introduced to technical understanding, and finally, are taught the tools to build the models. Only after this bottom-up process do learners intuit the big picture. We argue that this order not only holds students back from learning ML until after complex math prerequisites are met, but also makes every step of the process more difficult due to the absence of a big-picture understanding from the beginning.

We addressed these issues with our class, Introduction to Machine Learning, which provides learners a holistic and applied understanding of ML through an experiential and collaborative learning environment. Our pedagogical approach relies on strategic skill- and community-building activities, for example, an activity teaching the “train, test, deploy” pipeline activity immerses students to act as an ML model, giving learners a big-picture understanding of how a machine “thinks.” Then, we develop learners’ technical capabilities through high-structure instructional design and assignments that prepare them for mathematical reinforcement in the future.

By the end of the ten-week class, learners are able to design, implement, and test their own customized architecture for a convolutional or recurrent neural network on a dataset of their choosing. Learners demonstrate their ability to process datasets, create custom models, effectively use remote server computation, solve ML problems, and apply statistical and mathematical understanding by predicting outcomes of changes to the model structure - all skills required of an ML professional and reported to be highly in demand. Students achieve these skills despite only 18.8% reporting Python proficiency, and only 4.1% having any PyTorch experience at the start of the quarter. Overall, these outcomes indicate that the holistic and experiential approach in Introduction to Machine Learning successfully teaches advanced and high-demand topics to first-year students without requiring prerequisites, giving them a head start on later ML experience in education and industry.