Syllabus: AINS6003 Deep Learning & Neural Networks

Course Identity

Field	Value
Course	AINS6003 Deep Learning & Neural Networks
Program	Aurnova Master of Science in Artificial Intelligence
Track	Core
Credits	3 graduate credits
Format	8-week online graduate course with weekly instructor interaction
Primary environment	Google Colab for first-run labs; GitHub Codespaces for repository-based or instructor-guided work

Catalog Description

AINS6003 Deep Learning & Neural Networks explores neural network architectures—including multilayer perceptrons, convolutional, recurrent, and transformer models—with emphasis on training techniques, optimization, regularization, and applications in vision, speech, and generative AI. The course aligns with Aurnova AIN6003 (3 credits, online, 8-week block).

Student Audience and Prerequisites

AINS6002 or equivalent exposure to machine learning concepts is recommended. Prior Python experience is helpful but not assumed as a gate; Colab labs provide guided execution and interpretation practice. The course is designed for graduate students preparing to lead, evaluate, manage, or apply AI work in professional settings. Students are not expected to be computer science majors. Technical work is scaffolded so students can run a notebook, observe output, change one controlled variable, and explain the evidence in professional language.

Course-Level Learning Outcomes

By the end of this course, students will be able to:

• CO1: Analyze professional problems in Deep Learning & Neural Networks and formulate AI use cases with explicit stakeholders, decision boundaries, data assumptions, and success criteria.

• CO2: Execute or interpret reproducible notebook-based investigations that demonstrate core Deep Learning & Neural Networks methods using guided Python/Colab workflows.

• CO3: Evaluate model, workflow, or governance evidence for accuracy, validity, uncertainty, bias, security, privacy, and operational limitations appropriate to core MSAI contexts.

• CO4: Produce professional artifacts for Deep Learning & Neural Networks, including briefs, model cards, risk registers, evaluation memos, implementation plans, or executive recommendations.

• CO5: Apply responsible AI, academic integrity, data stewardship, accessibility, and human-oversight expectations to course work and proposed deployments.

• CO6: Communicate AI findings, limitations, tradeoffs, and next-step recommendations to technical and nontechnical stakeholders using clear graduate-level evidence.

Credit-Hour and Contact-Hour Rationale

This 3-credit accelerated graduate course expects approximately 135 total student learning hours across the 8-week term. The course design allocates approximately 45 hours to instructor-led lecture, narrated content, guided lab demonstration, discussion, feedback, and synchronous or asynchronous interaction. Approximately 90 additional hours are allocated to reading, notebook practice, applied assignments, revision, and the final portfolio artifact. Each module contains enough slide and narration material for a 90-minute class session, plus labs and assignments that extend learning beyond direct instruction.

Weekly Schedule and Major Deliverables

Week	Module Topic	Essential Question	Primary Deliverable
1	From neurons to multilayer networks	How does a stack of differentiable units approximate complex functions from data?	Module 1 Assignment: Baseline MLP design brief
2	Backpropagation and automatic differentiation	How does the chain rule enable efficient learning in deep networks?	Module 2 Assignment: Gradient trace and autograd check
3	Optimization, loss, and regularization	How do we train deep networks reliably when loss surfaces are non-convex and data are noisy?	Module 3 Assignment: Optimizer and regularization comparison
4	Convolutional neural networks for vision	Why do convolutions, pooling, and translation equivariance matter for image understanding?	Module 4 Assignment: CNN architecture memo
5	Sequence models: RNNs and LSTMs	How do recurrent architectures model temporal structure and long-range dependencies?	Module 5 Assignment: Sequence modeling design note
6	Attention and transformers	How does self-attention replace fixed recurrence for language and multimodal tasks?	Module 6 Assignment: Attention mechanism analysis
7	Generative models and applications	What distinguishes discriminative training from generative modeling?	Module 7 Assignment: Discriminative vs generative comparison
8	GPU workflows, scale, and deployment	How do practitioners move from notebook experiments to reproducible GPU training pipelines?	Module 8 Assignment: Reproducible training workflow checklist

Assessment Plan

Component	Weight	Evidence
Module applied assignments	35%	Eight short professional artifacts scored with module rubrics.
Notebook labs and reflections	20%	Guided Colab/Codespaces labs with before-after interpretation and limitation statements.
Mid-course synthesis brief	15%	Integrative memo or design artifact covering Modules 1-4.
Final applied portfolio artifact	20%	Cumulative artifact demonstrating evidence, risk reasoning, and stakeholder communication.
Participation and professional engagement	10%	Discussion, peer review, instructor check-ins, and revision responsiveness.

Passing performance requires both cumulative course performance of 70% or higher and submission of the final applied portfolio artifact. Graduate-level performance requires accurate vocabulary, evidence-based reasoning, responsible limitation statements, and professional communication. Students may revise selected artifacts when the instructor determines that revision supports learning outcomes and academic integrity.

Grading Scale

Grade	Range	Interpretation
A	93-100	Excellent graduate performance; evidence is rigorous, well communicated, and professionally defensible.
A-	90-92	Strong graduate performance with minor gaps in depth, precision, or integration.
B+	87-89	Good performance; work is complete and evidence-based but not consistently advanced.
B	83-86	Satisfactory graduate performance; meets core expectations with some limitations.
B-	80-82	Marginal graduate performance; recurring gaps require attention.
C	70-79	Minimal passing performance; significant weaknesses in evidence, accuracy, or communication.
F	Below 70	Does not demonstrate required course outcomes.

Required Readings, Standards, and Professional Resources

The instructor may update editions, links, or equivalent institutional resources before the term begins. Readings combine scholarly foundations, professional standards, and applied guidance.

Before the course opens, the instructor or program designee should verify library access, open-access links, edition currency, accessibility of required readings, and any instructor-selected substitutions. This verification should be retained with the program reading evidence described in CastaliaInstitute/MSAI/accreditation/AINS-scholarship-and-standards-audit.md.

Module	Topic	Required Reading or Standard	Applied Professional Resource
1	From neurons to multilayer networks	Goodfellow, Bengio, and Courville, Deep Learning, multilayer perceptrons.	Goodfellow, Bengio, and Courville, Deep Learning, backpropagation.
2	Backpropagation and automatic differentiation	Goodfellow, Bengio, and Courville, Deep Learning, backpropagation.	Goodfellow, Bengio, and Courville, Deep Learning, optimization for training deep models.
3	Optimization, loss, and regularization	Goodfellow, Bengio, and Courville, Deep Learning, optimization for training deep models.	Goodfellow, Bengio, and Courville, Deep Learning, convolutional networks.
4	Convolutional neural networks for vision	Goodfellow, Bengio, and Courville, Deep Learning, convolutional networks.	Jurafsky and Martin, sequence modeling and neural language model chapters.
5	Sequence models: RNNs and LSTMs	Jurafsky and Martin, sequence modeling and neural language model chapters.	Vaswani et al., Attention Is All You Need, 2017.
6	Attention and transformers	Vaswani et al., Attention Is All You Need, 2017.	Goodfellow et al., Generative Adversarial Nets, 2014, plus instructor-selected generative AI reading.
7	Generative models and applications	Goodfellow et al., Generative Adversarial Nets, 2014, plus instructor-selected generative AI reading.	PyTorch documentation, performance tuning and reproducibility guidance.
8	GPU workflows, scale, and deployment	PyTorch documentation, performance tuning and reproducibility guidance.	Goodfellow, Bengio, and Courville, Deep Learning, multilayer perceptrons.

Learning Technology and Beginner Support

Students should start notebook labs in Google Colab unless the instructor specifies Codespaces. Colab lowers setup burden and supports students who are new to Python. Students should first run the notebook unchanged, then change one small value and explain what changed. Codespaces is used when an activity requires the full repository, multiple files, tests, or GitHub workflow evidence. The course assesses interpretation, evidence, limitations, and professional judgment; it does not assume that every student enters as a programmer.

Before launch, the instructor or program designee should verify the accessibility and student-support checklist in CastaliaInstitute/MSAI/accreditation/AINS-accessibility-and-student-support-audit.md, including Colab link access, alternative activity paths, assistive-technology concerns, and support escalation routes.

Academic Integrity and Permitted AI Assistance

Students must submit their own analysis, decisions, explanations, and reflections. Use of generative AI tools is permitted for brainstorming, debugging, summarizing public documentation, and improving prose when allowed by the instructor, but students must disclose material AI assistance and remain responsible for factual accuracy, citations, code behavior, and final claims. Fabricated citations, hidden AI-generated submissions, unauthorized collaboration, or submission of work the student cannot explain may be treated as academic misconduct.

Accessibility and Accommodations

Aurnova is expected to provide reasonable accommodations consistent with institutional policy and applicable law. Students who need accommodations should contact the designated institutional office and notify the instructor as early as possible. Course materials should be provided in accessible formats when feasible, including readable HTML pages, downloadable notebooks, text-based instructions, and alternatives for activities that require specific hardware or external accounts.

Data Privacy, Safety, and Responsible Use

Unless explicitly approved by the instructor and institution, students must not upload confidential, regulated, proprietary, patient, student, financial, security-sensitive, or personally identifiable information into Colab, Codespaces, public repositories, or third-party AI tools. Labs use synthetic or public proxy data. Students must document data assumptions, privacy risks, and operational limits before recommending real-world use.

Late Work, Participation, and Instructor Interaction

Students are expected to engage weekly with readings, lecture material, labs, and applied deliverables. Late work may receive reduced credit unless prior arrangements are approved or institutional policy applies. The instructor should provide regular substantive interaction through announcements, discussion, feedback, office hours or appointments, assignment feedback, and timely grading. Participation is evaluated by evidence of preparation, constructive discussion, peer feedback, and revision effort rather than by technical bravado.

Accreditation Evidence Location

The companion accreditation packet for this course is available in accreditation.md. It maps course outcomes to program outcomes, module assessments, direct evidence, and continuous-improvement expectations.