Neural knowledge assembly in humans and neural networks.

Human understanding of the world can change rapidly when new information comes to light, such as when a plot twist occurs in a work of fiction. This flexible "knowledge assembly" requires few-shot reorganization of neural codes for relations among objects and events. However, existing computational theories are largely silent about how this could occur. Here, participants learned a transitive ordering among novel objects within two distinct contexts before exposure to new knowledge that revealed how they were linked. Blood-oxygen-level-dependent (BOLD) signals in dorsal frontoparietal cortical areas revealed that objects were rapidly and dramatically rearranged on the neural manifold after minimal exposure to linking information. We then adapt online stochastic gradient descent to permit similar rapid knowledge assembly in a neural network model.

Individual Alpha Frequency Determines the Impact of Bottom-Up Drive on Visual Processing.

Endogenous alpha oscillations propagate from higher-order to early visual cortical regions, consistent with the observed modulation of these oscillations by top-down factors. However, bottom-up manipulations also influence alpha oscillations, and little is known about how these top-down and bottom-up processes interact to impact behavior. To address this, participants performed a detection task while viewing a stimulus flickering at multiple alpha band frequencies. Bottom-up drive at a participant's endogenous alpha frequency either impaired or enhanced perception, depending on the frequency, but not amplitude, of their endogenous alpha oscillation. Fast alpha drive impaired perceptual performance in participants with faster endogenous alpha oscillations, while participants with slower oscillations displayed enhanced performance. This interaction was reflected in slower endogenous oscillatory dynamics in participants with fast alpha oscillations and more rapid dynamics in participants with slow endogenous oscillations when receiving high-frequency bottom-up drive. This central tendency may suggest that driving visual circuits at alpha band frequencies that are away from the peak alpha frequency improves perception through dynamical interactions with the endogenous oscillation. As such, studies that causally manipulate neural oscillations via exogenous stimulation should carefully consider interacting effects of bottom-up drive and endogenous oscillations on behavior.

Neural state space alignment for magnitude generalization in humans and recurrent networks.

A prerequisite for intelligent behavior is to understand how stimuli are related and to generalize this knowledge across contexts. Generalization can be challenging when relational patterns are shared across contexts but exist on different physical scales. Here, we studied neural representations in humans and recurrent neural networks performing a magnitude comparison task, for which it was advantageous to generalize concepts of "more" or "less" between contexts. Using multivariate analysis of human brain signals and of neural network hidden unit activity, we observed that both systems developed parallel neural "number lines" for each context. In both model systems, these number state spaces were aligned in a way that explicitly facilitated generalization of relational concepts (more and less). These findings suggest a previously overlooked role for neural normalization in supporting transfer of a simple form of abstract relational knowledge (magnitude) in humans and machine learning systems.

If deep learning is the answer, what is the question?

Neuroscience research is undergoing a minor revolution. Recent advances in machine learning and artificial intelligence research have opened up new ways of thinking about neural computation. Many researchers are excited by the possibility that deep neural networks may offer theories of perception, cognition and action for biological brains. This approach has the potential to radically reshape our approach to understanding neural systems, because the computations performed by deep networks are learned from experience, and not endowed by the researcher. If so, how can neuroscientists use deep networks to model and understand biological brains? What is the outlook for neuroscientists who seek to characterize computations or neural codes, or who wish to understand perception, attention, memory and executive functions? In this Perspective, our goal is to offer a road map for systems neuroscience research in the age of deep learning. We discuss the conceptual and methodological challenges of comparing behaviour, learning dynamics and neural representations in artificial and biological systems, and we highlight new research questions that have emerged for neuroscience as a direct consequence of recent advances in machine learning.

Fluctuations in instantaneous frequency predict alpha amplitude during visual perception.

Rhythmic neural activity in the alpha band (8-13 Hz) is thought to have an important role in the selective processing of visual information. Typically, modulations in alpha amplitude and instantaneous frequency are thought to reflect independent mechanisms impacting dissociable aspects of visual information processing. However, in complex systems with interacting oscillators such as the brain, amplitude and frequency are mathematically dependent. Here, we record electroencephalography in human subjects and show that both alpha amplitude and instantaneous frequency predict behavioral performance in the same visual discrimination task. Consistent with a model of coupled oscillators, we show that fluctuations in instantaneous frequency predict alpha amplitude on a single trial basis, empirically demonstrating that these metrics are not independent. This interdependence suggests that changes in amplitude and instantaneous frequency reflect a common change in the excitatory and inhibitory neural activity that regulates alpha oscillations and visual information processing.

Ethanol alters local cellular levels of (3α,5α)-3-hydroxypregnan-20-one (3α,5α-THP) independent of the adrenals in subcortical brain regions.

The neuroactive steroid (3α,5α)-3-hydroxypregnan-20-one (3α,5α-THP or allopregnanolone) is a positive modulator of GABAA receptors synthesized in the brain, adrenal glands, and gonads. In rats, ethanol activates the hypothalamic-pituitary-adrenal axis and elevates 3α,5α-THP in plasma, cerebral cortex, and hippocampus. In vivo, these effects are dependent on both the pituitary and adrenal glands. In vitro, however, ethanol locally increases 3α,5α-THP in hippocampal slices, in the absence of adrenal influence. Therefore, it is not known whether ethanol can change local brain levels of 3α,5α-THP in vivo, independent of the adrenals. To directly address this controversy, we administered ethanol (2 g/kg) or saline to rats that underwent adrenalectomy (ADX) or received sham surgery and performed immunohistochemistry for 3α,5α-THP. In the medial prefrontal cortex (mPFC), ethanol increased 3α,5α-THP after sham surgery, compared with saline controls, with no ethanol-induced change in 3α,5α-THP following ADX. In subcortical regions, 3α,5α-THP was increased independent of adrenals in the CA1 pyramidal cell layer, dentate gyrus polymorphic layer, bed nucleus of the stria terminalis, and paraventricular nucleus of the hypothalamus. Furthermore, ethanol decreased 3α,5α-THP labeling in the nucleus accumbens shore and central nucleus of the amygdala, independent of the adrenal glands. These data indicate that ethanol dynamically regulates local 3α,5α-THP levels in several subcortical regions; however, the adrenal glands contribute to 3α,5α-THP elevations in the mPFC. Using double immunofluorescent labeling we determined that adrenal dependence of 3α,5α-THP induction by ethanol is not due to a lack of colocalization of 3α,5α-THP with the cholesterol transporters steroidogenic acute regulatory protein (StAR) or translocator protein (TSPO).