Alice Wang

Natural intelligences (NIs) thrive in a dynamic world - they learn quickly, sometimes with only a few samples. In contrast, artificial intelligences (AIs) typically learn with a prohibitive number of training samples and computational power. What design principle difference between NI and AI could contribute to such a discrepancy? Here, we investigate the role of weight polarity: development processes initialize NIs with advantageous polarity configurations; as NIs grow and learn, synapse magnitudes update, yet polarities are largely kept unchanged. We demonstrate with simulation and image classification tasks that if weight polarities are adequately set a priori, then networks learn with less time and data. We also explicitly illustrate situations in which a priori setting the weight polarities is disadvantageous for networks. Our work illustrates the value of weight polarities from the perspective of statistical and computational efficiency during learning.

Why do brains have inhibitory connections? Why do deep networks have negative weights? We believe representing functions is the primary role of both (i) the brain in natural intelligence, and (ii) deep networks in artificial intelligence. Our answer to why there are inhibitory/negative weights is: to learn more functions. We prove that, in the absence of negative weights, neural networks with non-decreasing activation functions are not universal approximators. While this may be an intuitive result to some, to the best of our knowledge, there is no formal theory, in either machine learning or neuroscience, that demonstrates why negative weights are crucial in the context of representation capacity. Further, we provide insights on the geometric properties of the representation space that non-negative deep networks cannot represent. We expect these insights will yield a deeper understanding of more sophisticated inductive priors imposed on the distribution of weights that lead to more efficient biological and machine learning.

Area V4 is the first object-specific processing stage in the ventral visual pathway, just as area MT is the first motion-specific processing stage in the dorsal pathway. For almost 50 years, coding of object shape in V4 has been studied and conceived in terms of flat pattern processing, given its early position in the transformation of 2D visual images. Here, however, in awake monkey recording experiments, we found that roughly half of V4 neurons are more tuned and responsive to solid, 3D shape-in-depth, as conveyed by shading, specularity, reflection, refraction, or disparity cues in images. Using 2-photon functional microscopy, we found that flat- and solid-preferring neurons were segregated into separate modules across the surface of area V4. These findings should impact early shape-processing theories and models, which have focused on 2D pattern processing. In fact, our analyses of early object processing in AlexNet, a standard visual deep network, revealed a similar distribution of sensitivities to flat and solid shape in layer 3. Early processing of solid shape, in parallel with flat shape, could represent a computational advantage discovered by both primate brain evolution and deep-network training.