The goal of the monthly ML+X event is to allow researchers across different domains to unite based on common ML methodologies. This event will feature a series of 10 minute lightning talks that adhere to a specific methodological theme (e.g., clustering, computer vision, deep learning, etc.). Anyone who is applying machine learning in their work is welcome to present. Please see below for the presentation format requirements. If you have any questions about the application process, please send an email to Chris Endemann (
endemann@wisc.edu).
Presentation Format Requirements:
- Each presenter will be given 10min to present and 2min to answer questions.
- If domain-specific terminology is used throughout your presentation, please either define your terms near the start of the presentation or opt to use plain language
- During the presentation, the following questions must be addressed:
1. Data Description: What does your training data look like (e.g., dimensionality, number of observations, variables/features present, depiction of data, etc.)?
2. Data Validation: What steps were taken to ensure the training data was of high quality and representative of future observations? Is the training data representative of a larger population or is it biased in some way? If the data is labeled, how confident are you in the labels?
3. Modeling Goal: What is the end-goal of modeling the data (e.g. classify/predict future observations, understand which features contribute most to the final model's predictions, relate the model's fitted parameters to real-world mechanisms, etc.)?
4. Model Selection: What final model was selected for your analysis? How/why was that model selected over others? What is the ratio between the number of fitted parameters in the final model and the total number of observations used to train it?
5. Model Validation: As George Box is quoted saying, "All models are wrong, but some are useful." How did you validate your model and measure its performance?
6. Model Utility: Was the modeling end-goal achieved? If not, why might this model have underperformed for this application?
7. Software Tools: What were three software tools/packages or other ML tools that were essential to this work?