Modeling with Machine Learning for Computer Science

Spring2025 Student Body Chart: 6.C011/6.C511

Example: we wish to realize a diagnostic helper for a physician
Modeling is not: simply applying an off-the-shelf method to a problem (e.g., using a LLM to generate candidate diagnosis for a physician to consider in response to a patient description)
Modeling is not: developing a new method to address a generic problem (e.g., improving transformer efficiency, making them dynamically configurable)
Modeling is about relating methods and tasks (e.g., figuring out how that LLM can be made fair or robust, what fine-tuning data to use, how to incorporate physician feedback, etc)

Frame the task/capability you are after
ask the right questions, formalize the problem as a learning to predict/control task
Solve (data + method + optimization)
select/tailor/design a machine learning method to solve the task with the given data; specify the hypothesis class, objective function, optimization algorithm
Assess/understand the results
interpret/analyze the results, whether the method worked/when it is likely to work

6.C011/C511 - Course Overview