Career and accomplishments
2002-2007: Effects of hand position on the ocular exploration of the peripersonnal space
Thura D., Hadj-Bouziane F., Meunier M. and Boussaoud D.
Institut de Neurosciences Cognitives de la Méditerranée (INCM), CNRS & Aix-Marseille Université, Marseille (France)
The central aim of my Ph.D. research was to test the effects of the signals coding hand position in space (i.e. vision and proprioception) on the neuronal activities related to target selection, saccade preparation and execution in the monkey frontal eye field (FEF). At the behavioral level, it’s now well documented that eye and hand are tightly coupled both in space and time. Accordingly, neurophysiological studies showed that several areas devoted to the planning of arm movements are influenced by eye position signals. Nonetheless, less is known about how oculomotor structures integrate hand position signals.
Effects of hand position on saccadic reaction times. First, I demonstrated that visible and invisible hand position strongly affects saccadic reaction times, both in humans and monkeys (Thura et al., J Neurophysiol, 2008).
Hand position and fontal eye field neuronal activity in monkey. Then, using extracellular single-unit recordings in two monkeys trained to execute visually guided saccades (I used two different techniques to acquire and record eye movements: the scleral search coil and the infra-red camera), I brought evidence that hand position signals are integrated by FEF visual and saccadic neurons during target selection for saccade execution. Interestingly, neurons were more likely to be modulated if saccade's target was located in the peri-personal space of the monkey (Thura et al., Behav Brain Res, 2008; Thura et al., Cerebral Cortex, 2011).
Hand position and dynamical orientation of visuospatial attention. Finally, recent data tend to demonstrate that hand affects the dynamical properties of visuospatial attentional exploration when a visual target, located in the peri-personal space, is selected for upcoming saccades. FEF visual neurons might participate to the neural mechanisms allowing this bias. Behavioral experiments in humans and neurophysiological recordings in monkey are performed in order to better understand the implication of hand position in the temporal mechanism of visuospatial attention orientation (Thura et al., Society for Neuroscience Abstract, 2007).
2006-2007: Neuronal bases of Learning by Observation
Belmalih A., Thura D., Isbaine F., Brovelli A., Demolliens M., Meunier M. and Boussaoud D.
Institut de Neurosciences Cognitives de la Méditerranée (INCM), CNRS & Aix-Marseille Université, Marseille (France)
Numerous animal species, including human and non human primates, can learn either through their own experience (trial and error learning, TE) or through the observation of conspecifics. Here, we address the question of how the brain of an observer encodes the outcome of others behavior, with particular focus on error and success signals.
Two monkeys were trained, always together, on a visuospatial task where they had to map 2 or 4 visual stimuli presented at the center of a touch screen on 2 or 4 spatial targets occupying the screen corners. The experimental design allowed the monkeys to face each other, and to have access to the touch screen displayed face up between them. Only one of the two monkeys had access to the touch screen at a time (actor), the other could observe but not reach the screen (observer). Monkeys alternated between the two roles. To control that the observer actually looked at to the touch screen, its gaze direction was monitored using an infrared camera. Neural activity was recorded from the dorso-lateral prefrontal cortex of the observer, an area known to play a key role in the processing of action and its outcome during learning. The actor initiated a trial by putting his hand on a lever close to him. Then a visual cue was presented together with 2 or 4 targets. Pressing the correct target led to a success signal (green ring), followed after a delay by a reward (a drop of fruit juice); Incorrect response led to an error signal (red ring) and no reward was given. We used several conditions to investigate the neural activity: the observation of familiar or new cue-target associations, the execution of new associations that were (observational learning) or were not (TE) observed before. Gaze data showed that the observer did monitor the actor's behavior.
Behavioral data showed that the observation led to performance improvement of up to 13% for associations learned after observation relative to those learned by TE. Preliminary neuronal data indicate that the dorso-lateral prefrontal cortex contains at least two populations of neurons encoding error and/or success signals. One responds similarly to the outcomes of own and other's actions (cognitive mirror properties). The other differentiates the source of the outcome, responding either to signals triggered by the monkey's own actions or only to those triggered by the other monkey's actions.
Taken together, the neuronal properties provide preliminary evidence that the monkey prefrontal cortex integrates signals about own and others’ successful and erroneous behavior. (Belmalih et al., SfN meeting abstract, 2010)
2008-2012: Human decisions in noisy and changing conditions
Thura D., Beauregard-Racine J., Fradet C-W. and Cisek P.
Département de physiologie, Université de Montréal, Montréal (Qc), Canada
In recent years, significant progress has been made toward understanding the neural basis of primate decision-making. Most decision-making studies and models have suggested that simple decisions are made through a process of ‘bounded integration’, in which neurons integrate sensory evidence until a threshold is reached. However, nearly all of the results supporting this theory have been obtained in tasks where sensory evidence was constant during the course of each trial. In this particular situation, behavioral and neural data are also compatible with a model in which there is no slow integration of sensory evidence, but instead a combination of the low-pass filtered evidence with a growing ‘urgency’ signal. In a recent study, Cisek et al. (2009) presented human subjects with a task (‘tokens’ task) in which evidence changed over time. Results were more consistent with the ‘urgency-gating’ model than with “integrator” models, but it was not clear whether this was task-dependent.
Here, 32 human subjects performed a new task, conceptually similar to the ‘tokens’ task but perceptually close to the well-known ‘motion direction discrimination’ task, in which subjects perform two-alternative perceptual decisions about the direction of motion in a dynamic random-dot display. In our task, each trial began with a centrally-located visual stimulus consisting of 200 dots moving in random directions for 500ms. After, every 200ms, a small portion of the dots began to move coherently in one of the two directions, for a total of 15 steps. The subject’s task was to select, as soon as they felt confident enough, the target corresponding to the net direction of motion they predicted to see at the end of the trial. Once the choice was made, the remaining steps of coherence were reduced to a 20ms duration.
To distinguish the models, we embedded in a full pseudorandom sequence some trials whose first 6 motion steps provided either a perceptual bias for or against the correct target. Consistent with the ‘urgency-gating’ model, decision times as well as success probabilities of the majority of subjects were not significantly influenced by these early biases, suggesting that noisy sensory evidence in favor of a given choice was not integrated with a long time-constant. Additional analyses suggested that the level of certainty at which most subjects made their decisions decreased significantly during the course of a trial, suggestingbthat these long decisions were mostly driven by the evidence-irrelevant, motor-related urgency signal (Thura et al., 2012). In this paper we also analytically demonstrate that the urgency-gating model allows subjects to maximise what they care about the most in such tasks: the rate of reward.
Our results suggest that humans form decisions by comparing to a threshold the product of the momentary information provided by the environment with a growing signal related to elapsed time (‘urgency’).
2009-2014: Temporal mechanisms of decision-making in the dorsal premotor and primary cortex of the macaque monkey
Thura D. and Cisek P.
Département de physiologie, Université de Montréal, Montréal (Qc), Canada
Many decision-making studies and models have suggested that simple decisions are made through a process of “bounded integration”, in which neurons accumulate sensory evidence until a threshold is reached. However, when human subjects were presented with tasks in which evidence changed over time (Cisek et al., 2009; Thura et al., 2012), their behavior was better explained with an “urgency-gating” model, in which the current evidence is multiplied by a growing “urgency” signal. In the present study, we examine neural correlates of this process in monkey frontal cortex (PMd and M1).
Two monkeys were trained to perform a two-alternative reach decision task (the “tokens” task). The animal began each trial by moving a handle into a central circle in which 15 small circular tokens were randomly arranged. Next, the tokens began to jump, one-by-one every 200 ms, from the central circle to one or the other of two target circles placed 180° apart. The monkey’s task was to move the cursor, as soon as he felt sufficiently confident, to the target that he believed would ultimately receive the majority of the tokens. Once the choice was made, the timing of the remaining token movements was accelerated to either 50ms (fast block) or to 150ms (slow block) in separate blocks.
Analysis of behavioral variables (performance, decision time and probability of success at decision time) shows that the monkey acts very similarly to humans. For instance, they behave more hastily in the fast blocks and more conservatively in the slow blocks. Moreover, in trials incorporating an early bias in favor of or against the correct target, decision times as well as success probabilities are not influenced by these early biases in a way predicted by most “integrator” models. One hundred and seventy eight PMd and seventy four M1 task related neurons were recorded while monkeys were performing the “tokens” task. Among this population, 99 cells predicted the monkey’s choices (pre-decision target selectivity) and reached a fixed threshold at the time of decisions. Prior to the choice commitment, these decision-related activities reflect the profile of evidence presented to the monkey with target selectivity emerging earlier in easy than ambiguous trials. Consistent with behavioral data, analyses show that neural activity at a given moment is not significantly influenced by the information presented earlier in the trial, but is amplified as time is passing.
Overall, our results suggest that neural activity in premotor and motor cortex combines current sensory information provided by the environment with a growing urgency signal, and decisions are made when this quantity reaches a threshold (Thura and Cisek, Neuron, 2014).
2010-present: Neural bases of speed-accuracy trade-off adjustments during decision making and movement execution
Thura D., Cos I., Trung J. and Cisek P.
Département de neurosciences, Université de Montréal, Montréal (Qc), Canada
By looking at monkeys' behavior in the tokens task (see description above), we first noticed that animals' decision policy is compatible with the urgency gating model: in both blocks (the two temporal conditions: slow versus fast), long decisions are made with less confidence compared to short decisions. This suggests that decisions are increasingly influenced by a motor signal that pushes monkeys to decide and act under uncertainty. Here, we investigate the hypothesis that the brain adapts its speed-accuracy trade-off (SAT) by adjusting the urgency signal as a function of the task context.
Between blocks, we found that monkeys' decisions were usually shorter in the fast blocks compared to the slow blocks. Interestingly, this difference tends to vanish as time is passing, probably because monkeys realize that the amount of time potentially saved decreases as time elapses in a given trial. Monkeys thus appear to adjust their SAT policy as a function of the timing parameters of the task in order to optimize their rate of reward.
Then, we looked at the reaching movements performed by the monkeys to report their decisions. Within each block, movement duration decreases as a function of decision duration, and movement duration are shorter in the fast blocks for the shortest decisions. This is consistent with the idea that movement duration is influenced by the same context-dependent urgency signals that we estimated from animal’s decision policy.
To summary, we demonstrate for the first time a direct context-dependent correlation between two phenomena traditionally considered separate: accuracy criterion for decisions and movement duration. We interpret this correlation as the result of the influence of the urgency level on the decision process that also affects, as other potential factors, the way movement is performed to report the decision. We believe that such shared mechanism allows animals to ultimately increase their reward rate (Thura et al., 2014).
We are actually conducting single cell recordings in monkey PMd, M1 and GP (see a video of a GPi neuron activity recorded during the tokens task) to assess the neural bases of the SAT adjustment between blocks during both decision-making and movement execution. We are also recording local field potentials in both cortex and basal ganglia to look at some frequency-specific activities related to SAT adjustment and/or choice commitment during dynamic decision-making (see a video of a simultaneous recording of spkes and LFPs in GPi).