Class Philosophy
Three Interconnected Pillars
ML Concepts
Build intuition for each algorithm: what it does, when to use it, and its strengths and limitations.
Real-Life Bioengineering Data
Work with real-life biomedical datasets and experience real challenges such as missing values, noise, imbalance, and high dimensionality.
AI-Augmented ML Practice
Use AI tools and agents to implement, debug, compare, and interpret ML solutions—while keeping scientific judgment in human hands.
Our Recurring Workflow
Problem
Analysis
Interpretation
What You Will Learn
- Recognize supervised, unsupervised, and neural-network learning problems.
- Match bioengineering questions and dataset types to appropriate ML algorithms.
- Explain the intuition, advantages, limitations, and assumptions of common methods.
- Prepare real-world data, including missing, noisy, and imbalanced datasets.
- Choose meaningful evaluation metrics and recognize overfitting.
- Use AI coding agents to construct reproducible ML workflows.
- Inspect, test, and critically evaluate AI-generated code and analyses.
- Interpret ML outputs in the context of biological or biomedical questions.
Course Topics & Assignments
Each written homework is paired with a project demo. The pair is released together and due together: the written component checks individual conceptual understanding, while the demo applies those concepts to the same semester-long project. HW 2 + Demo 2 and HW 3 + Demo 3 are released one week before their due dates to provide additional preparation time. The three demos progressively build toward the final report and presentation.
| Module | Topics | Bioengineering & Implementation Focus | Assignments & Project Milestones |
|---|---|---|---|
| 1 · Foundations | Introduction to MLData processingMissing dataImputationNoise & jittering | Features, labels, dataset structure, train/validation/test splits, preprocessing, leakage, and biomedical-data challenges. | Project Start Choose/understand a real-life dataset and define the bioengineering question. Begin AI-assisted exploration and cleaning. |
| 2 · Regression | Linear Regressioncontinuous outcomes | Predict a continuous biological measurement. Interpret coefficients, residuals, error, assumptions, and limitations. | Written HW 1 Demo 1 Released together · Due together HW: data preparation, missing data, preprocessing, train/test concepts, and leakage. Demo: problem → data → AI-assisted cleaning, preprocessing, visualization, and verification. |
| 3 · Classification I | Logistic Regression · k-NN | Binary classification, probabilities, decision boundaries, scaling, interpretability, and neighborhood-based reasoning. | HW 1 + Demo 1 Due Submit the written concepts and first project demonstration together. |
| 4 · Classification II | Decision Trees · Random Forests | Nonlinear relationships, interpretability, variance, feature importance, and ensemble methods. | Written HW 2 Demo 2 Released · Due Week 6 Begin the written concepts and second project demo. One week is provided for preparation before submission. |
| 5 · Classification III | Support Vector Machines | Margins, kernels, feature scaling, and applications to smaller or high-dimensional datasets. | HW 2 + Demo 2 · Preparation Continue model selection, implementation, comparison, and evaluation before the Week 6 submission. |
| 6 · Evaluation | Model Evaluation & Reliabilitymetricsoverfittingimbalanced data | Accuracy, sensitivity/recall, specificity, precision, F1, ROC/AUC, cross-validation, generalization, and class imbalance. | HW 2 + Demo 2 Due Submit the written concepts and second project demonstration together. |
| 7 · Clustering | k-Means · Hierarchical Clustering · DBSCAN | Explore unlabeled biological data, discover groups, compare assumptions, and examine sensitivity to scale, density, and outliers. | Written HW 3 Demo 3 Released · Due Week 9 Begin the written concepts and third project demo. One week is provided for preparation before submission. |
| 8 · Dimension Reduction | PCA · UMAP | Visualize and summarize high-dimensional biomedical data; compare linear and nonlinear representations. | HW 3 + Demo 3 · Preparation Continue unsupervised/dimensionality-reduction analysis and refine the project before the Week 9 submission. |
| 9 · Neural Networks | Perceptrons · Multilayer Networks · Backpropagation | Understand layers, activations, loss, optimization, and learning conceptually; use AI tools to build and inspect a small neural network. | HW 3 + Demo 3 Due Submit the written concepts and third project demonstration together. |
| 10 · Modern Deep Learning | CNNs · RNNs · Transformers | Connect architectures to images, physiological signals/time series, sequences, and multimodal biomedical problems. | Final Project Report Integrate the three demos into a final report: problem → data → ML/AI workflow → evaluation → biological interpretation → limitations. Final Written Exam Mon, Dec 14 · 4:30–6:20 PM · OUG 141 |
Real-Life Bioengineering Data
Real datasets are a core part of the course—not an afterthought. Activities may use open-source biomedical datasets and datasets contributed by participating research labs.
Structured Data
Clinical variables, biomarkers, physiological measurements, experimental conditions, and tabular laboratory data.
Signals & Time Series
ECG, wearable sensors, movement, neural recordings, or other physiological signals.
Images & High-Dimensional Data
Biomedical images, microscopy, omics, or other datasets where dimension reduction and deep learning become useful.
Research bridge: When research-lab datasets are used, the activity can serve as a small “research trial”—helping students understand an active research question, meet research groups, and identify areas they may want to explore further.
AI-Augmented ML Practice
AI tools and agents are integrated throughout the course rather than treated as a separate topic. Students learn to use AI to accelerate implementation and exploration while developing the judgment needed to verify, challenge, and improve AI-generated solutions.
Five AI-Agent Competencies
1 · Problem Formulation
Translate a biological or biomedical question into a well-defined machine learning problem with appropriate inputs, outcomes, and constraints.
2 · Agent Instruction
Communicate the dataset, research objective, assumptions, constraints, and desired analysis clearly enough for an AI agent to contribute effectively.
3 · Critical Verification
Inspect AI-generated code and reasoning for data leakage, inappropriate preprocessing, incorrect metrics, methodological errors, and unsupported conclusions.
4 · Iterative Experimentation
Use agents to explore alternatives, compare preprocessing choices and algorithms, debug pipelines, and refine experiments rather than accepting the first answer.
5 · Scientific Interpretation
Connect model outputs to biological meaning, distinguish prediction from explanation, recognize uncertainty, and identify what the data cannot establish.
AI Across the ML Workflow
| ML Topic | Bioengineering Focus | AI-Agent Practice |
|---|---|---|
| Data Processing | How should messy physiological or biomedical data be prepared? | Ask an agent to inspect the data and propose preprocessing; then critique and revise its choices. |
| Regression | Can measurements or biomarkers predict a continuous outcome? | Generate a baseline pipeline, examine assumptions and residuals, and verify the implementation. |
| Classification | Can we distinguish clinically or biologically meaningful groups? | Build competing classifiers, justify preprocessing, and compare their behavior. |
| Model Evaluation | Is the model actually useful and trustworthy? | Use AI to help detect overfitting, leakage, imbalance, and inappropriate evaluation metrics. |
| Clustering | Are meaningful subgroups present in unlabeled biological data? | Explore multiple clustering approaches and challenge the agent's interpretation of clusters. |
| PCA / UMAP | What structure appears in high-dimensional biomedical data? | Generate visualizations and critically distinguish visual patterns from scientific conclusions. |
| Neural Networks | Does a neural model add value over classical ML? | Build a neural network with agent support and compare it against a simpler baseline. |
| CNN / RNN / Transformer | Which architecture matches images, signals, time series, or sequences? | Ask an agent to propose an architecture, then require a data- and problem-based justification. |
Students remain responsible for understanding, validating, and explaining all submitted analyses. AI-generated code and explanations are tools for scientific work—not substitutes for scientific judgment.
Application Studios & Final Project
Assignments can be organized as connected application studios. Students repeatedly practice the same scientific workflow on increasingly complex bioengineering datasets and culminate in a final project.
Studio 1 · Predict
Clean a real dataset, choose a regression or classification method, build a baseline model, and justify the choice.
Studio 2 · Compare
Apply multiple ML algorithms to one bioengineering question. Compare performance, interpretability, assumptions, and failure modes.
Studio 3 · Discover
Use clustering and dimensionality reduction to explore biological structure without predefined labels.
Studio 4 · Deep Learning
Use an appropriate neural architecture for images, signals, or sequences and compare it with a classical ML baseline.
Final Project · Research Bridge
Choose a real bioengineering dataset, formulate a meaningful question, select and implement appropriate ML methods, evaluate them, and interpret the findings.
Communicate
Present the problem, data, algorithm choice, AI-assisted workflow, evaluation, biological interpretation, limitations, and next steps.
Grading (Tentative)
- Written Homework: 30%
- Project Demos: 30%
- Final Project Report: 20%
- Final Written Exam: 20%
Course Emphasis
The goal is: “Given a real bioengineering problem and dataset, can you choose an appropriate method, use modern tools to implement it correctly, evaluate whether it works, recognize its limitations, and explain what the results mean?”