Can multisensory training aid visual learning? A computational investigation

The Development of Multisensory Integration at the Neuronal Level

Multisensory integration is a fundamental function of the brain. In the typical adult, multisensory neurons' response to paired multisensory (e.g., audiovisual) cues is significantly more robust than the corresponding best unisensory response in many brain regions. Synthesizing sensory signals from multiple modalities can speed up sensory processing and improve the salience of outside events or objects. Despite its significance, multisensory integration is testified to be not a neonatal feature of the brain. Neurons' ability to effectively combine multisensory information does not occur rapidly but develops gradually during early postnatal life (for cats, 4-12 weeks required). Multisensory experience is critical for this developing process. If animals were restricted from sensing normal visual scenes or sounds (deprived of the relevant multisensory experience), the development of the corresponding integrative ability could be blocked until the appropriate multisensory experience is obtained. This section summarizes the extant literature on the development of multisensory integration (mainly using cat superior colliculus as a model), sensory-deprivation-induced cross-modal plasticity, and how sensory experience (sensory exposure and perceptual learning) leads to the plastic change and modification of neural circuits in cortical and subcortical areas.

The developmental trajectory of object recognition robustness: Children are like small adults but unlike big deep neural networks

In laboratory object recognition tasks based on undistorted photographs, both adult humans and deep neural networks (DNNs) perform close to ceiling. Unlike adults', whose object recognition performance is robust against a wide range of image distortions, DNNs trained on standard ImageNet (1.3M images) perform poorly on distorted images. However, the last 2 years have seen impressive gains in DNN distortion robustness, predominantly achieved through ever-increasing large-scale datasets-orders of magnitude larger than ImageNet. Although this simple brute-force approach is very effective in achieving human-level robustness in DNNs, it raises the question of whether human robustness, too, is simply due to extensive experience with (distorted) visual input during childhood and beyond. Here we investigate this question by comparing the core object recognition performance of 146 children (aged 4-15 years) against adults and against DNNs. We find, first, that already 4- to 6-year-olds show remarkable robustness to image distortions and outperform DNNs trained on ImageNet. Second, we estimated the number of images children had been exposed to during their lifetime. Compared with various DNNs, children's high robustness requires relatively little data. Third, when recognizing objects, children-like adults but unlike DNNs-rely heavily on shape but not on texture cues. Together our results suggest that the remarkable robustness to distortions emerges early in the developmental trajectory of human object recognition and is unlikely the result of a mere accumulation of experience with distorted visual input. Even though current DNNs match human performance regarding robustness, they seem to rely on different and more data-hungry strategies to do so.

Enriched learning: behavior, brain, and computation

The presence of complementary information across multiple sensory or motor modalities during learning, referred to as multimodal enrichment, can markedly benefit learning outcomes. Why is this? Here, we integrate cognitive, neuroscientific, and computational approaches to understanding the effectiveness of enrichment and discuss recent neuroscience findings indicating that crossmodal responses in sensory and motor brain regions causally contribute to the behavioral benefits of enrichment. The findings provide novel evidence for multimodal theories of enriched learning, challenge assumptions of longstanding cognitive theories, and provide counterevidence to unimodal neurobiologically inspired theories. Enriched educational methods are likely effective not only because they may engage greater levels of attention or deeper levels of processing, but also because multimodal interactions in the brain can enhance learning and memory.

Decoding sounds depicting hand-object interactions in primary somatosensory cortex

Neurons, even in the earliest sensory regions of cortex, are subject to a great deal of contextual influences from both within and across modality connections. Recent work has shown that primary sensory areas can respond to and, in some cases, discriminate stimuli that are not of their target modality: for example, primary somatosensory cortex (SI) discriminates visual images of graspable objects. In the present work, we investigated whether SI would discriminate sounds depicting hand-object interactions (e.g. bouncing a ball). In a rapid event-related functional magnetic resonance imaging experiment, participants listened attentively to sounds from 3 categories: hand-object interactions, and control categories of pure tones and animal vocalizations, while performing a one-back repetition detection task. Multivoxel pattern analysis revealed significant decoding of hand-object interaction sounds within SI, but not for either control category. Crucially, in the hand-sensitive voxels defined from an independent tactile localizer, decoding accuracies were significantly higher for hand-object interactions compared to pure tones in left SI. Our findings indicate that simply hearing sounds depicting familiar hand-object interactions elicit different patterns of activity in SI, despite the complete absence of tactile stimulation. These results highlight the rich contextual information that can be transmitted across sensory modalities even to primary sensory areas.