Exploring the Role of Valence in Conscious Perception: Insights from Similarity Judgments and Deep Learning Models

Recent theories claim that valence plays an important role in conscious perception (eg. Barrett & Bar, 2009). Inspired by these theories, we tested how valence judgments relate to similarity judgments and whether they correlate with different stages of processing in deep neural networks (DNNs). To support the hypothesis that all perception is valenced, we focused on micro-valence, (Lebrecht et al., 2012). Forty-seven participants provided similarity and valence judgments for 120 images of everyday objects using the odd-one-out task and the birthday task (Lebrecht et al., 2012). For the same images, we extracted activations from the layers of DNNs trained to classify objects. Representation similarity analysis (Hebart et al. 2020) and multidimensional scaling analyses highlight the role of valence in the similarity space. Further, DNN analysis showed that perceptual features of the stimuli contributed to both valence and similarity processing. Most importantly, valence correlated with activations in the early DNN layers, suggesting a role of low-level visual features in valence computation. These results indicate that valence computation may occur in early visual processing. They also show that valence is involved in similarity judgments, corroborating recent claims for the functional role of conscious experience (Cleeremans & Tallon-Baudry, 2022).