Brain-inspired global-local learning incorporated with neuromorphic computing

There are two principle approaches for learning in artificial intelligence: error-driven global learning and neuroscience-oriented local learning. Integrating them into one network may provide complementary learning capabilities for versatile learning scenarios. At the same time, neuromorphic computing holds great promise, but still needs plenty of useful algorithms and algorithm-hardware co-designs to fully exploit its advantages. Here, we present a neuromorphic global-local synergic learning model by introducing a brain-inspired meta-learning paradigm and a differentiable spiking model incorporating neuronal dynamics and synaptic plasticity. It can meta-learn local plasticity and receive top-down supervision information for multiscale learning. We demonstrate the advantages of this model in multiple different tasks, including few-shot learning, continual learning, and fault-tolerance learning in neuromorphic vision sensors. It achieves significantly higher performance than single-learning methods. We further implement the model in the Tianjic neuromorphic platform by exploiting algorithm-hardware co-designs and prove that the model can fully utilize neuromorphic many-core architecture to develop hybrid computation paradigm.

Young and Aged Neuronal Tissue Dynamics With a Simplified Neuronal Patch Cellular Automata Model

Realistic single-cell neuronal dynamics are typically obtained by solving models that involve solving a set of differential equations similar to the Hodgkin-Huxley (HH) system. However, realistic simulations of neuronal tissue dynamics —especially at the organ level, the brain— can become intractable due to an explosion in the number of equations to be solved simultaneously. Consequently, such efforts of modeling tissue- or organ-level systems require a lot of computational time and the need for large computational resources. Here, we propose to utilize a cellular automata (CA) model as an efficient way of modeling a large number of neurons reducing both the computational time and memory requirement. First, a first-order approximation of the response function of each HH neuron is obtained and used as the response-curve automaton rule. We then considered a system where an external input is in a few cells. We utilize a Moore neighborhood (both totalistic and outer-totalistic rules) for the CA system used. The resulting steady-state dynamics of a two-dimensional (2D) neuronal patch of size 1, 024 × 1, 024 cells can be classified into three classes: (1) Class 0–inactive, (2) Class 1–spiking, and (3) Class 2–oscillatory. We also present results for different quasi-3D configurations starting from the 2D lattice and show that this classification is robust. The numerical modeling approach can find applications in the analysis of neuronal dynamics in mesoscopic scales in the brain (patch or regional). The method is applied to compare the dynamical properties of the young and aged population of neurons. The resulting dynamics of the aged population shows higher average steady-state activity 〈a(t → ∞)〉 than the younger population. The average steady-state activity 〈a(t → ∞)〉 is significantly simplified when the aged population is subjected to external input. The result conforms to the empirical data with aged neurons exhibiting higher firing rates as well as the presence of firing activity for aged neurons stimulated with lower external current.

On the Neuronal Dynamics of Aesthetic Experience: Evidence from Electroencephalographic Oscillatory Dynamics

Aesthetic experiences have an influence on many aspects of life. Interest in the neural basis of aesthetic experiences has grown rapidly in the past decade, and fMRI studies have identified several brain systems supporting aesthetic experiences. Work on the rapid neuronal dynamics of aesthetic experience, however, is relatively scarce. This study adds to this field by investigating the experience of being aesthetically moved by means of ERP and time–frequency analysis. Participants' electroencephalography (EEG) was recorded while they viewed a diverse set of artworks and evaluated the extent to which these artworks moved them. Results show that being aesthetically moved is associated with a sustained increase in gamma activity over centroparietal regions. In addition, alpha power over right frontocentral regions was reduced in high- and low-moving images, compared to artworks given intermediate ratings. We interpret the gamma effect as an indication for sustained savoring processes for aesthetically moving artworks compared to aesthetically less-moving artworks. The alpha effect is interpreted as an indication of increased attention for aesthetically salient images. In contrast to previous works, we observed no significant effects in any of the established ERP components, but we did observe effects at latencies longer than 1 sec. We conclude that EEG time–frequency analysis provides useful information on the neuronal dynamics of aesthetic experience.

Neuronal Dynamics of Pain in Parkinson’s Disease

Pain is an important non-motor symptom of Parkinson’s disease (PD). It negatively impacts the quality of life. However, the pathophysiological mechanisms underlying pain in PD remain to be elucidated. This study sought to use electroencephalographic (EEG) coherence analysis to compare neuronal synchronization in neuronal networks between patients with PD, with and without pain. Twenty-four patients with sporadic PD were evaluated for the presence of pain. Time-frequency and coherence analyses were performed on their EEG data. Whole-brain and regional coherence were calculated and compared between pain-positive and pain-negative patients. There was no significant difference in the whole-brain coherence between the pain-positive and pain-negative groups. However, temporal–temporal coherence differed significantly between the two groups (p = 0.031). Our findings indicate that aberrant synchronization of inter-temporal regions is involved in PD-related pain. This will further our understanding of the mechanisms underlying pain in PD.

Neuronal Dynamics and miRNA Signaling Differ between SH-SY5Y APPSwe and PSEN1 Mutant iPSC-Derived AD Models upon Modulation with miR-124 Mimic and Inhibitor

Neuronal miRNA dysregulation may have a role in the pathophysiology of Alzheimer’s disease (AD). miRNA(miR)-124 is largely abundant and a critical player in many neuronal functions. However, the lack of models reliably recapitulating AD pathophysiology hampers our understanding of miR-124’s role in the disease. Using the classical human SH-SY5Y-APP695 Swedish neuroblastoma cells (SH-SWE) and the PSEN1 mutant iPSC-derived neurons (iNEU-PSEN), we observed a sustained upregulation of miR-124/miR-125b/miR-21, but only miR-124 was consistently shuttled into their exosomes. The miR-124 mimic reduced APP gene expression in both AD models. While miR-124 mimic in SH-SWE neurons led to neurite outgrowth, mitochondria activation and small Aβ oligomer reduction, in iNEU-PSEN cells it diminished Tau phosphorylation, whereas miR-124 inhibitor decreased dendritic spine density. In exosomes, cellular transfection with the mimic predominantly downregulated miR-125b/miR-21/miR-146a/miR-155. The miR-124 inhibitor upregulated miR-146a in the two experimental cell models, while it led to distinct miRNA signatures in cells and exosomes. In sum, though miR-124 function may be dependent on the neuronal AD model, data indicate that keeping miR-124 level strictly controlled is crucial for proper neuronal function. Moreover, the iNEU-PSEN cellular model stands out as a useful tool for AD mechanistic studies and perhaps for the development of personalized therapeutic strategies.

Imaging Voltage with Microbial Rhodopsins

Membrane potential is the critical parameter that reflects the excitability of a neuron, and it is usually measured by electrophysiological recordings with electrodes. However, this is an invasive approach that is constrained by the problems of lacking spatial resolution and genetic specificity. Recently, the development of a variety of fluorescent probes has made it possible to measure the activity of individual cells with high spatiotemporal resolution. The adaptation of this technique to image electrical activity in neurons has become an informative method to study neural circuits. Genetically encoded voltage indicators (GEVIs) can be used with superior performance to accurately target specific genetic populations and reveal neuronal dynamics on a millisecond scale. Microbial rhodopsins are commonly used as optogenetic actuators to manipulate neuronal activities and to explore the circuit mechanisms of brain function, but they also can be used as fluorescent voltage indicators. In this review, we summarize recent advances in the design and the application of rhodopsin-based GEVIs.