Learning From Negative Results#
I posted previously a little about my MS thesis. It’s finally completely done and fully published.
My project started out with the objective to build better models for human emotion, using a standard set of features derived from peripheral physiology signals. These signals are considered to be indirect, but still valuable, measurements of activity in the brain. Having more detailed models for how emotion exists in the brain could have impact on research in psychopathologies. Prior work with standard statistical analysis showed some differences in the measurements that correlated with the different emotions. The problem that I was to address was that the measurements were only weak predictors of the specific emotions.
The main idea was to incorporate time or more sophisticated classifiers in order to improve performance. Unfortunately, as i tried new methods, none gave a noteworthy improvement. None even gave diagnostic information that gave me ideas for how to build an alternate model that would work. However, I did come across interesting questions as I tried to diagnose the issue. I was able to apply more detailed analysis to the results I had to provide new insight that was valuable.
I also learned enough that I have ideas for future research. I still need to re-align these ideas with a slightly new project direction, but I feel much more prepared to choose a good research direction.