3. Undergraduate Technical AI Track

Course Positioning

A full undergraduate-level technical course that can serve as an AI minor foundation, elective, or intensive bridge into applied AI research.

Learning outcomes

Explain supervised, unsupervised, self-supervised, generative, and reinforcement learning paradigms.
Derive and implement core ML algorithms at a level sufficient for debugging and research adaptation.
Train, evaluate, and compare deep learning models, transformers, embeddings, and generative models.
Design reliable experiments using baselines, ablations, uncertainty estimates, and statistical reporting.
Build AI systems that include data pipelines, models, evaluation harnesses, deployment constraints, and monitoring.
Read modern AI papers and translate them into testable implementation plans.

Course Design Snapshot

Positioning: A full undergraduate-level technical course that can serve as an AI minor foundation, elective, or intensive bridge into applied AI research.
Audience: Undergraduates in CS, engineering, math, statistics, biology, economics, design technology, or other quantitative disciplines.
Duration: 14 weeks, 3-4 hours of lecture/lab per week plus independent project time.
Prerequisites: Python, data structures, linear algebra basics, probability basics, and comfort reading technical documentation.
Format: Lecture, derivation-lite theory, implementation labs, paper discussions, reproducibility assignments, and final project.

Expanded Topic-by-Topic Coverage

Module 1. Mathematical foundations

Module focus: Mathematical foundations: vectors, matrices, probability, expectations, distributions, optimization geometry. Primary live activity or lab: Diagnostic notebook: vector operations, probability simulation, and gradient visualization.

Topics and coverage

vectors

What it means: define vectors clearly and connect it to the module focus: Mathematical foundations: vectors, matrices, probability, expectations, distributions, optimization geometry.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

matrices

What it means: define matrices clearly and connect it to the module focus: Mathematical foundations: vectors, matrices, probability, expectations, distributions, optimization geometry.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

probability

What it means: connect probability to the data lifecycle from source and structure through analysis, interpretation, and decision-making.
What to cover: source reliability, missing or biased data, leakage, assumptions, calculations, and the difference between correlation and decision-ready evidence.
Demonstration: walk through a small dataset or example table and mark the checks required before trusting the result.
Evidence of learning: learners produce a short analysis note that includes assumptions, limitations, and verification steps.

expectations

What it means: define expectations clearly and connect it to the module focus: Mathematical foundations: vectors, matrices, probability, expectations, distributions, optimization geometry.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

distributions

What it means: define distributions clearly and connect it to the module focus: Mathematical foundations: vectors, matrices, probability, expectations, distributions, optimization geometry.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

optimization geometry

What it means: place optimization geometry inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

Practice and evidence of learning

Learners complete or discuss: Diagnostic notebook: vector operations, probability simulation, and gradient visualization.
Learners produce: Diagnostic notebook: vector operations, probability simulation, and gradient visualization.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 2. Supervised learning

Module focus: Supervised learning: regression, classification, regularization, bias-variance, and cross-validation. Primary live activity or lab: Implement and compare linear/logistic models, tree models, and boosted baselines.

Topics and coverage

regression

What it means: place regression inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

classification

What it means: place classification inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

regularization

What it means: define regularization clearly and connect it to the module focus: Supervised learning: regression, classification, regularization, bias-variance, and cross-validation.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

bias-variance

What it means in this course: define bias-variance in operational terms, not as an abstract principle.
What to cover: sensitive data boundaries, affected stakeholders, approval paths, documentation, and what Undergraduates in CS, engineering, math, statistics, biology, economics, design technology, or other quantitative disciplines must never delegate blindly to AI.
Use case: present one acceptable use, one borderline use, and one prohibited use, then ask learners to justify the classification.
Evidence of learning: learners add a risk control, review step, or escalation rule to their course project.

cross-validation

What it means: define cross-validation clearly and connect it to the module focus: Supervised learning: regression, classification, regularization, bias-variance, and cross-validation.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

Practice and evidence of learning

Learners complete or discuss: Implement and compare linear/logistic models, tree models, and boosted baselines.
Learners produce: Implement and compare linear/logistic models, tree models, and boosted baselines.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 3. Optimization

Module focus: Optimization: gradient descent variants, stochasticity, normalization, initialization, and training dynamics. Primary live activity or lab: Train the same model under different optimizers and analyze convergence.

Topics and coverage

gradient descent variants

What it means: define gradient descent variants clearly and connect it to the module focus: Optimization: gradient descent variants, stochasticity, normalization, initialization, and training dynamics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

stochasticity

What it means: define stochasticity clearly and connect it to the module focus: Optimization: gradient descent variants, stochasticity, normalization, initialization, and training dynamics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

normalization

What it means: define normalization clearly and connect it to the module focus: Optimization: gradient descent variants, stochasticity, normalization, initialization, and training dynamics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

initialization

What it means: define initialization clearly and connect it to the module focus: Optimization: gradient descent variants, stochasticity, normalization, initialization, and training dynamics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

training dynamics

What it means: place training dynamics inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

Practice and evidence of learning

Learners complete or discuss: Train the same model under different optimizers and analyze convergence.
Learners produce: Train the same model under different optimizers and analyze convergence.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 4. Representation learning

Module focus: Representation learning: embeddings, metric learning, dimensionality reduction, clustering, and retrieval. Primary live activity or lab: Build an embedding search system and evaluate nearest-neighbor quality.

Topics and coverage

embeddings

What it means: explain how embeddings changes the interaction between human intent, model behavior, external information, and final output.
What to cover: inputs, constraints, examples, output format, grounding, iteration, failure modes, and when a human must intervene.
Demonstration: show a weak attempt, a stronger structured attempt, and a reviewed final version with explicit checks.
Evidence of learning: learners create a reusable prompt, schema, retrieval note, or workflow pattern and test it on at least two examples.

metric learning

What it means: define metric learning clearly and connect it to the module focus: Representation learning: embeddings, metric learning, dimensionality reduction, clustering, and retrieval.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

dimensionality reduction

What it means: define dimensionality reduction clearly and connect it to the module focus: Representation learning: embeddings, metric learning, dimensionality reduction, clustering, and retrieval.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

clustering

What it means: define clustering clearly and connect it to the module focus: Representation learning: embeddings, metric learning, dimensionality reduction, clustering, and retrieval.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

retrieval

What it means: explain how retrieval changes the interaction between human intent, model behavior, external information, and final output.
What to cover: inputs, constraints, examples, output format, grounding, iteration, failure modes, and when a human must intervene.
Demonstration: show a weak attempt, a stronger structured attempt, and a reviewed final version with explicit checks.
Evidence of learning: learners create a reusable prompt, schema, retrieval note, or workflow pattern and test it on at least two examples.

Practice and evidence of learning

Learners complete or discuss: Build an embedding search system and evaluate nearest-neighbor quality.
Learners produce: Build an embedding search system and evaluate nearest-neighbor quality.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 5. Deep learning

Module focus: Deep learning: MLPs, CNNs, normalization, dropout, residual connections, and GPU training basics. Primary live activity or lab: Implement a PyTorch training loop with logging and checkpointing.

Topics and coverage

MLPs

What it means: define MLPs clearly and connect it to the module focus: Deep learning: MLPs, CNNs, normalization, dropout, residual connections, and GPU training basics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

CNNs

What it means: define CNNs clearly and connect it to the module focus: Deep learning: MLPs, CNNs, normalization, dropout, residual connections, and GPU training basics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

normalization

What it means: define normalization clearly and connect it to the module focus: Deep learning: MLPs, CNNs, normalization, dropout, residual connections, and GPU training basics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

dropout

What it means: define dropout clearly and connect it to the module focus: Deep learning: MLPs, CNNs, normalization, dropout, residual connections, and GPU training basics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

residual connections

What it means: define residual connections clearly and connect it to the module focus: Deep learning: MLPs, CNNs, normalization, dropout, residual connections, and GPU training basics.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

GPU training basics

What it means: place GPU training basics inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

Practice and evidence of learning

Learners complete or discuss: Implement a PyTorch training loop with logging and checkpointing.
Learners produce: Implement a PyTorch training loop with logging and checkpointing.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 6. Sequence models and transformers

Module focus: Sequence models and transformers: tokenization, attention, positional encodings, pretraining objectives. Primary live activity or lab: Code a small attention block and train a toy character-level model.

Topics and coverage

tokenization

What it means: define tokenization clearly and connect it to the module focus: Sequence models and transformers: tokenization, attention, positional encodings, pretraining objectives.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

attention

What it means: define attention clearly and connect it to the module focus: Sequence models and transformers: tokenization, attention, positional encodings, pretraining objectives.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

positional encodings

What it means: define positional encodings clearly and connect it to the module focus: Sequence models and transformers: tokenization, attention, positional encodings, pretraining objectives.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

pretraining objectives

What it means: place pretraining objectives inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

Practice and evidence of learning

Learners complete or discuss: Code a small attention block and train a toy character-level model.
Learners produce: Code a small attention block and train a toy character-level model.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 7. Foundation models

Module focus: Foundation models: scaling laws intuition, instruction tuning, alignment, prompting, RAG, fine-tuning, adapters. Primary live activity or lab: Compare prompting, RAG, and lightweight fine-tuning for the same task.

Topics and coverage

scaling laws intuition

What it means: define scaling laws intuition clearly and connect it to the module focus: Foundation models: scaling laws intuition, instruction tuning, alignment, prompting, RAG, fine-tuning, adapters.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

instruction tuning

What it means: define instruction tuning clearly and connect it to the module focus: Foundation models: scaling laws intuition, instruction tuning, alignment, prompting, RAG, fine-tuning, adapters.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

alignment

What it means: define alignment clearly and connect it to the module focus: Foundation models: scaling laws intuition, instruction tuning, alignment, prompting, RAG, fine-tuning, adapters.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

prompting

What it means: explain how prompting changes the interaction between human intent, model behavior, external information, and final output.
What to cover: inputs, constraints, examples, output format, grounding, iteration, failure modes, and when a human must intervene.
Demonstration: show a weak attempt, a stronger structured attempt, and a reviewed final version with explicit checks.
Evidence of learning: learners create a reusable prompt, schema, retrieval note, or workflow pattern and test it on at least two examples.

RAG

What it means: explain how RAG changes the interaction between human intent, model behavior, external information, and final output.
What to cover: inputs, constraints, examples, output format, grounding, iteration, failure modes, and when a human must intervene.
Demonstration: show a weak attempt, a stronger structured attempt, and a reviewed final version with explicit checks.
Evidence of learning: learners create a reusable prompt, schema, retrieval note, or workflow pattern and test it on at least two examples.

fine-tuning

What it means: define fine-tuning clearly and connect it to the module focus: Foundation models: scaling laws intuition, instruction tuning, alignment, prompting, RAG, fine-tuning, adapters.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

adapters

What it means: define adapters clearly and connect it to the module focus: Foundation models: scaling laws intuition, instruction tuning, alignment, prompting, RAG, fine-tuning, adapters.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

Practice and evidence of learning

Learners complete or discuss: Compare prompting, RAG, and lightweight fine-tuning for the same task.
Learners produce: Compare prompting, RAG, and lightweight fine-tuning for the same task.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 8. Generative models

Module focus: Generative models: autoencoders, diffusion intuition, language generation, sampling, evaluation limits. Primary live activity or lab: Experiment with sampling settings and evaluate diversity, quality, and factuality.

Topics and coverage

autoencoders

What it means: define autoencoders clearly and connect it to the module focus: Generative models: autoencoders, diffusion intuition, language generation, sampling, evaluation limits.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

diffusion intuition

What it means: define diffusion intuition clearly and connect it to the module focus: Generative models: autoencoders, diffusion intuition, language generation, sampling, evaluation limits.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

language generation

What it means: define language generation clearly and connect it to the module focus: Generative models: autoencoders, diffusion intuition, language generation, sampling, evaluation limits.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

sampling

What it means: define sampling clearly and connect it to the module focus: Generative models: autoencoders, diffusion intuition, language generation, sampling, evaluation limits.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

evaluation limits

What it means: connect evaluation limits to the data lifecycle from source and structure through analysis, interpretation, and decision-making.
What to cover: source reliability, missing or biased data, leakage, assumptions, calculations, and the difference between correlation and decision-ready evidence.
Demonstration: walk through a small dataset or example table and mark the checks required before trusting the result.
Evidence of learning: learners produce a short analysis note that includes assumptions, limitations, and verification steps.

Practice and evidence of learning

Learners complete or discuss: Experiment with sampling settings and evaluate diversity, quality, and factuality.
Learners produce: Experiment with sampling settings and evaluate diversity, quality, and factuality.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 9. Reinforcement learning

Module focus: Reinforcement learning: MDPs, value functions, policy gradients, exploration, reward design. Primary live activity or lab: Train a small RL agent in a gridworld and study reward hacking.

Topics and coverage

MDPs

What it means: define MDPs clearly and connect it to the module focus: Reinforcement learning: MDPs, value functions, policy gradients, exploration, reward design.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

value functions

What it means: define value functions clearly and connect it to the module focus: Reinforcement learning: MDPs, value functions, policy gradients, exploration, reward design.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

policy gradients

What it means in this course: define policy gradients in operational terms, not as an abstract principle.
What to cover: sensitive data boundaries, affected stakeholders, approval paths, documentation, and what Undergraduates in CS, engineering, math, statistics, biology, economics, design technology, or other quantitative disciplines must never delegate blindly to AI.
Use case: present one acceptable use, one borderline use, and one prohibited use, then ask learners to justify the classification.
Evidence of learning: learners add a risk control, review step, or escalation rule to their course project.

exploration

What it means: define exploration clearly and connect it to the module focus: Reinforcement learning: MDPs, value functions, policy gradients, exploration, reward design.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

reward design

What it means: show where reward design appears in the learner's real workflow and which parts are judgment-heavy versus draftable.
What to cover: current workflow, pain points, AI-assisted steps, human review checkpoints, quality standard, and ownership of the final decision.
Demonstration: convert one messy real-world input into a structured brief, draft, analysis, checklist, or next action.
Evidence of learning: learners produce a reusable template or playbook entry that can be used after the course.

Practice and evidence of learning

Learners complete or discuss: Train a small RL agent in a gridworld and study reward hacking.
Learners produce: Train a small RL agent in a gridworld and study reward hacking.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 10. Evaluation science

Module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage. Primary live activity or lab: Reproduce a result from a small paper or benchmark and report deviations.

Topics and coverage

baselines

What it means: define baselines clearly and connect it to the module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

ablations

What it means: define ablations clearly and connect it to the module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

confidence intervals

What it means: define confidence intervals clearly and connect it to the module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

calibration

What it means: define calibration clearly and connect it to the module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

robustness

What it means: define robustness clearly and connect it to the module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

benchmark leakage

What it means: define benchmark leakage clearly and connect it to the module focus: Evaluation science: baselines, ablations, confidence intervals, calibration, robustness, and benchmark leakage.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

Practice and evidence of learning

Learners complete or discuss: Reproduce a result from a small paper or benchmark and report deviations.
Learners produce: Reproduce a result from a small paper or benchmark and report deviations.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 11. Interpretability, safety, and governance

Module focus: Interpretability, safety, and governance: saliency, probes, causal claims, privacy, security, and misuse analysis. Primary live activity or lab: Audit a model using error slices, interpretability probes, and a risk register.

Topics and coverage

saliency

What it means: define saliency clearly and connect it to the module focus: Interpretability, safety, and governance: saliency, probes, causal claims, privacy, security, and misuse analysis.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

probes

What it means: define probes clearly and connect it to the module focus: Interpretability, safety, and governance: saliency, probes, causal claims, privacy, security, and misuse analysis.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

causal claims

What it means: define causal claims clearly and connect it to the module focus: Interpretability, safety, and governance: saliency, probes, causal claims, privacy, security, and misuse analysis.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

privacy

What it means in this course: define privacy in operational terms, not as an abstract principle.
What to cover: sensitive data boundaries, affected stakeholders, approval paths, documentation, and what Undergraduates in CS, engineering, math, statistics, biology, economics, design technology, or other quantitative disciplines must never delegate blindly to AI.
Use case: present one acceptable use, one borderline use, and one prohibited use, then ask learners to justify the classification.
Evidence of learning: learners add a risk control, review step, or escalation rule to their course project.

security

What it means in this course: define security in operational terms, not as an abstract principle.
What to cover: sensitive data boundaries, affected stakeholders, approval paths, documentation, and what Undergraduates in CS, engineering, math, statistics, biology, economics, design technology, or other quantitative disciplines must never delegate blindly to AI.
Use case: present one acceptable use, one borderline use, and one prohibited use, then ask learners to justify the classification.
Evidence of learning: learners add a risk control, review step, or escalation rule to their course project.

misuse analysis

What it means in this course: define misuse analysis in operational terms, not as an abstract principle.
What to cover: sensitive data boundaries, affected stakeholders, approval paths, documentation, and what Undergraduates in CS, engineering, math, statistics, biology, economics, design technology, or other quantitative disciplines must never delegate blindly to AI.
Use case: present one acceptable use, one borderline use, and one prohibited use, then ask learners to justify the classification.
Evidence of learning: learners add a risk control, review step, or escalation rule to their course project.

Practice and evidence of learning

Learners complete or discuss: Audit a model using error slices, interpretability probes, and a risk register.
Learners produce: Audit a model using error slices, interpretability probes, and a risk register.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 12. AI systems

Module focus: AI systems: data versioning, model serving, latency, batching, caching, cost, observability, and incident response. Primary live activity or lab: Deploy a small model endpoint and simulate monitoring drift.

Topics and coverage

data versioning

What it means: connect data versioning to the data lifecycle from source and structure through analysis, interpretation, and decision-making.
What to cover: source reliability, missing or biased data, leakage, assumptions, calculations, and the difference between correlation and decision-ready evidence.
Demonstration: walk through a small dataset or example table and mark the checks required before trusting the result.
Evidence of learning: learners produce a short analysis note that includes assumptions, limitations, and verification steps.

model serving

What it means: place model serving inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

latency

What it means: place latency inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

batching

What it means: define batching clearly and connect it to the module focus: AI systems: data versioning, model serving, latency, batching, caching, cost, observability, and incident response.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

caching

What it means: define caching clearly and connect it to the module focus: AI systems: data versioning, model serving, latency, batching, caching, cost, observability, and incident response.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

cost

What it means: place cost inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

observability

What it means: place observability inside the AI system stack so learners know what problem it solves and what tradeoffs it introduces.
What to cover: inputs, outputs, system boundaries, evaluation criteria, cost or latency implications, and common failure cases.
Demonstration: use a diagram, small code sample, worksheet, or tool trace to make the mechanism visible.
Evidence of learning: learners compare two approaches and explain which one they would choose for a realistic constraint.

incident response

What it means: define incident response clearly and connect it to the module focus: AI systems: data versioning, model serving, latency, batching, caching, cost, observability, and incident response.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

Practice and evidence of learning

Learners complete or discuss: Deploy a small model endpoint and simulate monitoring drift.
Learners produce: Deploy a small model endpoint and simulate monitoring drift.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 13. Research workshop

Module focus: Research workshop: paper reading, method extraction, implementation planning, and experimental design. Primary live activity or lab: Prepare a one-page paper-to-project translation memo.

Topics and coverage

paper reading

What it means: define paper reading clearly and connect it to the module focus: Research workshop: paper reading, method extraction, implementation planning, and experimental design.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

method extraction

What it means: define method extraction clearly and connect it to the module focus: Research workshop: paper reading, method extraction, implementation planning, and experimental design.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

implementation planning

What it means: define implementation planning clearly and connect it to the module focus: Research workshop: paper reading, method extraction, implementation planning, and experimental design.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

experimental design

What it means: show where experimental design appears in the learner's real workflow and which parts are judgment-heavy versus draftable.
What to cover: current workflow, pain points, AI-assisted steps, human review checkpoints, quality standard, and ownership of the final decision.
Demonstration: convert one messy real-world input into a structured brief, draft, analysis, checklist, or next action.
Evidence of learning: learners produce a reusable template or playbook entry that can be used after the course.

Practice and evidence of learning

Learners complete or discuss: Prepare a one-page paper-to-project translation memo.
Learners produce: Prepare a one-page paper-to-project translation memo.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Module 14. Capstone presentations and peer review

Module focus: Capstone presentations and peer review: technical demo, experiment report, and future work. Primary live activity or lab: Submit code, report, and reproducibility checklist.

Topics and coverage

technical demo

What it means: define technical demo clearly and connect it to the module focus: Capstone presentations and peer review: technical demo, experiment report, and future work.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

experiment report

What it means: define experiment report clearly and connect it to the module focus: Capstone presentations and peer review: technical demo, experiment report, and future work.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

future work

What it means: define future work clearly and connect it to the module focus: Capstone presentations and peer review: technical demo, experiment report, and future work.
What to cover: the core concept, why it matters, what good usage looks like, and where learners are likely to misunderstand it.
Demonstration: give one simple example, one realistic example, and one failure or limitation example.
Evidence of learning: learners explain the topic in their own words and apply it to a small artifact or decision.

Practice and evidence of learning

Learners complete or discuss: Submit code, report, and reproducibility checklist.
Learners produce: Submit code, report, and reproducibility checklist.
Instructor checks for accuracy, practical usefulness, clear assumptions, appropriate human review, and fit with the course audience.
Learners revise once after feedback so the module contributes to the final project, portfolio, or playbook.

Minimum coverage before moving on

Learners can explain the module vocabulary without relying on tool-generated text.
Learners have seen one worked example, one hands-on application, and one limitation or failure case.
Learners know what must be verified, what data must be protected, and who remains accountable for the output.

Core labs and builds

From-scratch ML lab: linear/logistic regression and gradient descent.
Deep learning lab: PyTorch training loop, checkpoints, and metric logging.
Transformer lab: toy attention, tokenizer choices, and generation experiments.
AI systems lab: model serving, latency measurement, cost estimate, and monitoring plan.

Capstone

A research-style applied AI project. Students must define a task, reproduce at least one baseline, implement a meaningful improvement or analysis, run ablations, evaluate limitations, and submit a reproducible repository and concise technical paper.

Assessment design

Weekly notebooks graded for correctness, interpretation, and reproducibility.
Paper discussion memos and peer critiques.
Midterm practical exam covering supervised learning and deep learning implementation.
Final project with code review, experiment report, presentation, and model/system card.

Recommended tools and datasets

Python, NumPy, pandas, scikit-learn, PyTorch, Hugging Face, Weights & Biases or MLflow, Git, Docker basics, Jupyter, cloud GPU credits if available.

Instructor notes

Undergraduates should learn to reason from problem to data to method to evaluation. Avoid turning the course into a tour of APIs. Make students implement enough internals to debug models and read papers confidently.

Instructor Build Checklist

Prepare one short demo for each module and one learner activity that creates a saved artifact.
Prepare examples that match the audience, local context, and likely tools learners can access.
Add a verification step to every AI-generated output: factual check, source check, data sensitivity check, and quality review.
Keep a running portfolio folder so each module contributes to the final project or learner playbook.
Reserve time for reflection on what the learner did, what AI did, what was checked, and what remains uncertain.