Università degli studi di Modena e Reggio Emilia

In affective computing, machines should detect and recognize human emotions, and adapt their behavior to them. Nowadays these objectives are mainly achieved through the exploitation of machine learning techniques. These techniques are employed to process different emotional-related features and to produce classification labels or coordinates in a valence-arousal space. This approach, however, ignores the neurophysiological processes governing implicit emotional states and, consequently, suffers from substantial limitations. Moreover, machine learning methods employed today do not benefit from the knowledge of emotional dynamics for properly adapting themselves. In this manuscript, starting from recent neuroscience and computational theories, we show how a simple mathematical description of processes governing implicit emotional dynamics can be developed. Moreover, we discuss how our mathematical models can be exploited to improve tracking, estimation and active modulation of human emotions.

It is well known from the technical literature that non-conscious perception of emotional stimuli affects behavior, perception, and even decision making [e.g., see Ref. (1) for a comprehensive review]. Non-conscious perception can be obtained by inducing sensory unawareness, e.g., through backward masking and binocular rivalry (1). Experiments adopting such paradigms have evidenced that non-consciously perceived emotional stimuli elicit activity in the amygdala, superior colliculus, basal ganglia, and pulvinar. More specifically, it has been shown that a subcortical fast route exists between the thalamus and the amygdala, which, in turn, project onto different cortical and subcortical structures [e.g., onto the nucleus accumbens, NAcc, when appetitive stimuli are perceived (2)]. These findings agree with the hypothesis about amygdala functionality proposed by LeDoux (3, 4). In fact, LeDoux has hypothized the existence of a thalamic pathway to the amygdala; such a pathway would allow to automatically detect evolutionary prepared visual stimuli (such as emotional faces, sexual-related stimuli, spiders, snakes, and injuries). Note that this model is also supported by other results acquired by different researchers that have employed masking in normal participants (5, 6) or have observed brain activity in patients affected by cortical blindness (7, 8). According to this model about amygdala functionality, the superior colliculus stimulates the pulvinar nucleus of the thalamus, which then arouses the amygdala (4, 9, 10). This suggests that salient features representing biologically prepared stimuli could be stored in the amygdala since birth. From an evolutionary perspective, this can be related to the fact that fast and implicit (or unconscious) reactions are needed in dangerous and highly dynamical environments. Moreover, even ontogenetic stimuli (e.g., weapons) are encoded within the amygdala through implicit learning during life (11, 12). These data evidence the importance of subcortical regions associated with implicit emotional processing. In fact, since the brain structure works like a hierarchical network (13) in which the limbic system represents a lower hierarchical level with respect to the higher cortical structure, it is likely that the overall perception and emotional appraisal are influenced by low-level evaluations. More specifically, the signals coming from lower and higher hierarchical levels determine prediction errors (or error signals) at intermediate levels; such error signals propagate through the entire hierarchical structure, determining cognitive perception, causes attributions, emotional evaluations, actions, and behaviors (14). Hence, if subcortical limbic-brainstem regions are defective, all the network hierarchy functioning will be compromised. As a matter of fact, a dysfunction in the limbic-brainstem regions is associated with various psychiatric disorders with higher cognitive deficits including autism, schizophrenia, posttraumatic stress disorders (PTSD), attention deficits/hyperactivity disorder (ADHD), neurosis, phobia, and others.