Frame the problem
Requirements 1 to 3
Pick a real, valuable problem, justify why it matters with evidence, and turn it into a well-defined machine-learning task with inputs, outputs, and constraints.
Suggested Weekly Plan for Project Execution
A suggested plan for running the project across a 13-week semester, mapping the project requirements to weeks, with the proposal, midterm, and final at weeks 5, 8, and 13, plus two extra weeks to polish and submit the repository. Treat it as a template, and adjust it to your own course schedule.
The thirteen-week plan, plus two to submit
Each week names the focus, the requirements it advances, and the milestone where relevant.
Research prior work and novelty
Requirements 4 and 5
Review recent papers and state-of-the-art methods, map the existing approaches and datasets, and state the gap and the novelty your project will add.
Plan and start the data
Requirement 6
Design a sound data acquisition or generation methodology and begin building the dataset. See the Synthetic Data and EDA guides.
EDA and the first baseline
Requirement 7
Run exploratory data analysis, fix the data problems it surfaces, and implement a first credible baseline. Draft the proposal slides.
Proposal presentation
Milestone · Proposal
Present the use case, problem statement, novelty, planned models and data, and evaluation plan. Follow the Project Proposal guide.
Build the models
Requirement 8
Implement several genuinely different approaches on the same data and metrics, including a trained or fine-tuned model and zero-shot or few-shot variants, not only prompting.
Experiments and rigor
Requirements 8 and 11
Run controlled experiments, tune hyperparameters and design choices, and start a systematic error analysis. Prepare the midterm.
Midterm presentation
Milestone · Midterm
Show the dataset and EDA, the baseline, initial measurable results, and an error analysis, with an honest completion plan. Follow the Midterm guide.
Act on feedback and iterate
Requirements 11 and 12
Address the instructor's feedback, improve the weak components, and rerun controlled experiments so each change is measured, not assumed.
Deepen and ablate
Requirements 11 and 13
Add ablation studies and robustness checks, and be ready to justify the non-trivial design decisions and the trade-offs behind them.
Consolidate the results
Requirements 9 and 10
Finalize the comparison, write clear result tables and figures, and get the repository into shape as you go.
Freeze and reproduce
Requirements 10 and 12
Lock the final results, make sure the pipeline reproduces end to end, and keep the code, data, and slides aligned. Prepare the final talk.
Final presentation
Milestone · Final
Deliver the full technical argument: definition, achievements, methodology, results, interpretation, conclusions, and demonstrated effort. Follow the Final Presentation guide.
Polish the repository
Repository, part 1
Clean the README, structure, and documentation; document the data and models; make training and evaluation reproducible with exact commands; and add the worked examples.
Final submission
Milestone · Submission
Run the GitHub Submission checklist: latest aligned artifacts, slides in PPT and PDF, no secrets, working links, and a result someone else can reproduce. Then submit.