While models are central to scientific practice, modeling remains a surprisingly individual process that relies on the inclinations and ingenuity of the investigator. In this course we will attempt to identify what constitutes a good model by examining general perspectives on modeling practice and specific case studies of successful models, as well as modeling our own data sets.
Thursdays, 10-11am, via Zoom
gagarwal at berkeley dot edu
This course is ideally suited for those who are immersed in a particular scientific problem and are seeking to model a data set. Class meetings will include discussions of readings and students’ research projects. Students will also meet in small groups with an instructor over the semester to develop models.
Aims and Learning outcomes
This course has two aims. First, it seeks to clarify the role of modeling in neuroscience, and practices that promote the development of good models. Second, it seeks to help students apply these lessons to model data sets of their personal interest. By the end of the semester, students will be more comfortable with finding, developing, and evaluating models.
Students are expected to read the assigned literature, complete assignments, and participate actively in discussions. In addition, they will choose a data set to model over the course of the semester, and meet regularly with a small group and instructor to develop the model. You will work ~2 hours / week outside of class, but much of this is aimed to support your personal research efforts. These also apply to auditors, unless you discuss otherwise with an instructor.
We would like to create a meeting space that promotes easy communication, respectful disagreement, and collaboration. Here are some provisional guidelines to aid this:
- Doors open at 10 and we start promptly at 10:05. The class officially ends at 11, but the room will remain open afterwards for informal discussion.
- Everyone is expected to participate. You can do so through questions or comments, or by actively listening to others.
- Keep your video on so we may reduce the anonymity of virtually meeting.
- If you disagree with, or do not understand something, please feel free to interject. These detours are often invaluable.
Some presentations and discussions will be recorded for future reference, but will not be posted in any public forums.
Weeks 1-3: What makes a good model?
We will read perspectives that specify best practices, common shortcomings, and fundamental limitations of the modeling process. We will reach out to paper authors with any unresolved questions.
Weeks 4-5: What am I modeling?
Students will present a scientific problem they are studying or are interested in, an experiment designed to study that problem, and candidate modeling approaches for the resulting data.
Weeks 6-8: History of a model
We will review a series of influential papers on the Drift Diffusion Model, starting from its invocation to explain behavioral data, the eventual discovery of a neural substrate, and a revisionist critique.
Weeks 9-11: Models from the inside out
Each week we will read a paper that presents a model, discuss its strengths and weaknesses, and meet with one of the authors. Papers will be chosen to reflect different scales of investigation (e.g. cellular, neural populations, behavioral, computational). Authors will be invited to share their experience in searching for and choosing a model, tradeoffs they encountered, and lessons they learned in the process.
Weeks 12-15: Project presentations
Students will present their project, stating the scientific problem, as well as the challenges and choices they confronted during the modeling process.