We work on two complimentary types of data. The first is “in-house” dataset, which is collected within CIBR. This provides a great opportunity to work together with experimentalists to understand their scientific problems, observe how the data is collected, and receive immediate feedback on the algorithm performance. We often find that typical solutions found in textbooks and papers do not work right away, and thus developing a working analysis pipeline requires many trial-and-errors as well as combining multiple approaches. However, once successful, we can immediately see the impact of our work on accelerating neuroscience research. For example, what used to take hours by manual labor can now be automated, or high-throughput experiments that were previously infeasible are now within reach etc.
The other type of data we work on is publicly available neuroscience dataset. These datasets offer the advantage of allowing easier comparisons between different algorithms and provide opportunities to learn from others working on the same dataset. This allows us to refine our skills and knowledge, which is helpful when analyzing in-house data. We frequently participate in international neuroscience data analysis competitions. Previously, we have finished 3rd place on Visual Columns Challenge and 2nd place on Minimum Feedback Challenge, 11th place on VNC Matching Challenge, all organized by the FlyWire at Princeton University, and 2nd place on Brain-to-Text benchmark ‘24 organized by researchers at Stanford University. By making our solutions publicly available, we aim to contribute to the global research community and accelerate scientific progress.
TOP
