I did two secondments:
- June and July 2017: Institut de Neurociències and Departament de Cognició i Desenvolupament, Universitat de Barcelona (Supervisor: Prof. Joan López-Moliner). During those two months we extended our behavioral study on saccadic reaction time estimation to a manual response task.
- August 2017: Department of Biomedical Engineering and Zlotowski Center for Neuroscience, Ben-Gurion University of the Negev (Supervisor: Prof. Opher Donchin). The purpose for this secondment was to improve my skills on computational modeling especially on Bayesian algorithms implementation.
Secondment at the Universitat de Barcelona.
Previous studies showed how environmental contingencies can modulate saccadic reaction time distribution. This implies that one should perceive her own reaction time to solve the credit assignment problem in order to correctly allocate responses given environmental properties.
During the secondment, we extended the paradigm we used to probe this hypothesis on saccadic movements to manual responses.
In particular, in the saccadic reaction time task we trained ten participants in an adaptive procedure (“staircase paradigm” or “up and down procedure”). In each trial they had to saccade to a stepping target and then in a 2-AFC task, they had to choose the number representing the actual saccadic reaction time (SRT) while the second number was a made-up value which proportionally differed from the value representing SRT. The relative difference between the two alternatives was computed either by adding or subtracting one of the percentage values of a decreasing fixed staircase range.
The procedure that we used to conduct our experiment in Barcelona was the same but instead of performing saccadic movements the participants (n=10) were involved in a reaching task, in which they had to reach toward a target appearing on a tablet. We used two different manipulations: in one condition, participants had to estimate their reaction times, and in another condition, they had to estimate the duration of their movements. In order to facilitate and speed up learning, the made-up value was determined by a QUEST procedure instead of a fixed staircase. In addition, a double staircase has been used.
A feedback was provided in both conditions after each response.
Psychometric functions have been computed for each participant and the probability of correct responses for each percentage of the staircases plotted as a function of the percentage difference. Results showed that in both conditions performance was accurate with the 75% of correct responses corresponding to percentages ranging from around 7 to 35. I particular, movement time discrimination was more accurate than reaction time perception with thresholds ranging from 7 to 21 in the movement time discrimination task and from 10 to 35 in the reaction time discrimination.
This indicates that our participants can discriminate very small reaction time differences, providing support for the possibility that the credit assignment problem may be solved even for short reaction times.
During this secondment I managed to work with arm movement dataset, I learned how to collect these types of data and how to analyze them by programming in R. I also improved my knowledge on function fitting.
Secondment at the Ben-Gurion University of the Negev.
The secondment in Beer-Sheba was mainly motivated by the necessity of having some knowledge about Bayesian modeling and apply it on my study.
As previously said, in two experiments participants were engaged in a 2-AFC task in which, after a movement, they had to choose the option representing the actual reaction time in milliseconds. The main criticism about this procedure is that participants can choose the answer according to an estimated (by using the feedback information) mean and variance of their latency distribution. Obviously as long as the made up number is set outside the estimated distribution the percentage of correct responses is significantly higher when compared to the trials where the two numbers are both inside the area of the estimated distribution. This problem can be at least partially overtaken by doing a Bayesian analysis.
During my staying I’ve done additional analysis on my results and I started to study the basis on Bayesian modeling, I trained on coding some algorithms like the Markov chain Monte Carlo and I improved programming skill and tools usage.
Analyses on this topic are still in progress.
To conclude, I really enjoyed those two secondments and I am thankful to Joan López-Moliner and Opher Donchin for having given me the opportunity to join their amazing teams and for having shared their knowledge with me. It was a very professionally and personally speaking fulfilling experience thanks to which I’ve learned a lot of things and improved my chances to succeed in my career.