* | Final project: Presentation
Objective(s)
Assemble and deliver a 15 minute presentation (plus five minutes for questions) that describes your term project.
In this presentation, you should include:
- Information about your team and each member’s role.
- An introduction, including relevant background descriptions, to the “big picture” question that you are seeking to address.
- A clearly articulated research question (e.g., “Can we predict XYZ from ABC?”).
- At least one slide that clearly describes your desired inputs and outputs to an ML model.
- Data descriptions, including characterization via statistics.
- An introduction to your model of choice. Why did you opt to go with this model? Where else is it used?
- A walkthrough of the training and execution of your model.
- Results, including an assessment of performance. How well did your model work? What remains to be done?
- Future directions.
Rubric
| Criterion | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|
| Timing & Execution | Unprepared delivery; significantly over or under allotted time. | Limited rehearsal evident; significant time management issues. | Some evidence of preparation, but delivery is uneven; slightly over or under time. | Mostly well-rehearsed with minor stumbles; stays on time or very close. | Clearly well-rehearsed; delivery polished; on time. |
| Clarity | Very difficult to understand; little to no effort made to communicate clearly. | Frequently unclear or disorganized; audience would struggle to understand the work. | Understandable in parts, but some sections are difficult to follow or poorly organized. | Mostly clear with minor moments of confusion; accessible to a general technical audience. | Exceptionally easy to follow for a non-expert; ideas flow logically and all key concepts are well-explained. |
| Formulation of a research question | No explicit research question, non-existent motivation, and no description of background or state of the art. | A research question is present but with little to no motivation (or vice versa); little to no description of background. | A well-developed research question and motivation; good background description including state of the art, though descriptions may be technical or difficult to follow. | A clear presentation of the research question, motivation, and existing work/background, which are all well-developed and detailed. | As in 4, with exceptional depth, accessibility, and synthesis of background material. |
| Design & deployment of a AI/ML Workflow | Little or no description of inputs and outputs; non-existent data characterization; no articulation of challenges; the audience has no idea how input data will translate to something useful. | Incomplete descriptions of inputs, outputs, and workflow; explanations are unclear. | Clear descriptions and characterizations of inputs and outputs, but may be missing critical information (e.g., where data come from). | A detailed and easy-to-follow presentation of all aspects of the ML workflow and associated inputs, outputs, and challenges. | As in 4, with exceptional clarity and insight into design decisions and tradeoffs. |
| Curation & characterization of dataset | No dataset presented. | Dataset(s) exist but processing and characterization are either non-existent or poorly done. | Datasets are well-organized and any interested party can go from raw data to the processed and characterized dataset used in the workflows. | Datasets are exceptionally curated: everything is detailed, properly attributed, and easy to follow. | As in 4, with outstanding documentation, provenance tracking, and accessibility for outsiders. |
| Assessment of AI/ML Performance | No assessment of algorithm performance. | An assessment of performance exists but either lacks critical analysis or is inappropriate. | An assessment of performance exists, is clearly described, and provides critical insights into what is working and what might be improved. | Multiple assessments of performance—each appropriate, well-detailed, and illuminating—exist. | As in 4, with exceptional rigor, breadth of metrics, and actionable insights. |
| Reproducibility of Results | No publicly available repository. | A repository exists but is missing critical elements (e.g., README.md); lack of organization; code is difficult to follow or fails to run. | A repository exists, is well-organized, and code executes properly. | The repository is easy to navigate and understand for a technically proficient (but non-subject-matter-expert) outsider. | As in 4, with exemplary documentation, environment setup instructions, and end-to-end reproducibility. |
Deadline
Presentations will be held in-class on March 12 and 14.