Opportunistic Face Encoding in the Brain: the Role of the Categorization Task

Abstract

- Introduction: In this study we demonstrate that the uncertainty of a categorization task modulates the information that the observer encodes from the stimulus. We demonstrate this in the context of a face categorization task and show that task uncertainty causes a greater extraction of information. - Methods: Two observers completed 8000 trials of condition 1 (EXPECT) and 8000 trials of condition 2 (NOEXPECT). In the EXPECT condition, observers were aware of the categorization task to be performed (gender or expressiveness judgment) prior to presentation of the sparsely sampled face stimulus (see [1] for details of the stimuli). In the NOEXPECT condition this task information came afterwards (Figure 1). Concurrently we measured high density EEG, EOG and behavioural response. For each sensor and time point we computed a classification image representing the visual information underlying modulations in neuronal activity at that sensor and time point (for full details see [2]). We establish the feature sensitivity value for three visual features (the left eye, the right eye and the mouth, established as the critical information for gender and expressiveness tasks [1,2,3]) as the percentage overlap between each classification image and a feature reference template [2,3]. Projecting each feature sensitivity value as a component in an RGB color space (see Figure 2) allows for the visualization of the processing of both individual and combinations of multiple features within the brain. To measure the effects of expectation on information encoding we compute the feature sensitivity measure on all trials (collapsing GENDER and EXNEX categorizations) and compare the EXPECT and NOEXPECT conditions. To reduce the dimensionality of sensitivity data we sum the sensitivity measures across the EEG channels, resulting in a 1-dimensional temporally varying information measure for each feature in each condition. To compare information processing across EXPECT and NOEXPECT, we subtract the former from the latter to produce the curves shown in figures 2 and 3 (black curve represents the sum of all features). Their interpretation is simple: a positive value for a feature reveals that it is more encoded in the brain in the NOEXPECT condition; a negative value reveals that it is more encoded in the EXPECT condition. - Results: The above information measures are shown for two participants in Figures 2 and 3 for the time region from 140ms to 240ms after stimulus presentation. In each, the top row gives the RGB coding measure for the NOEXPECT condition, the bottom row gives the EXPECT condition and the middle row is RGB temporal information measure. For both participants, more significant visual information is encoded in the NOEXPECT condition than in the EXPECT condition. - Conclusions: The results obtained support our initial hypothesis that when the observer knows in advance the relevant task, they will adopt an opportunistic strategy extracting only the minimum information required from the stimulus. When task uncertainty is introduced the observer is forced to adopt a more generic strategy, and encode more visual information from the stimulus, maintaining potentially redundant information in memory until the task uncertainty is removed.