Many engineered and biological systems must trade off performance and energy use, and the brain is no exception. While there are theories on how activity levels are controlled in biological networks through feedback control (homeostasis), it is not clear what the effects on population coding are, and therefore how performance and energy can be traded off. In this talk we will consider this tradeoff in auto-encoding networks, in which there is a clear definition of performance (the coding loss). We first show how SNNs follow a characteristic trade-off curve between activity levels and coding loss, but that standard networks need to be retrained to achieve different tradeoff points. We next formalize this tradeoff with a joint loss function incorporating coding loss (performance) and activity loss (energy use). From this loss we derive a class of spiking networks which coordinates its spiking to minimize both the activity and coding losses -- and as a result can dynamically adjust its coding precision and energy use. The network utilizes several known activity control mechanisms for this --- threshold adaptation and feedback inhibition --- and elucidates their potential function within neural circuits. Using geometric intuition, we demonstrate how these mechanisms regulate coding precision, and thereby performance. Lastly, we consider how these insights could be transferred to trained SNNs. Overall, this work addresses a key energy-coding trade-off which is often overlooked in network studies, expands on our understanding of homeostasis in biological SNNs, as well as provides a clear framework for considering performance and energy use in artificial SNNs.

Cortical circuits process information by rich recurrent interactions between excitatory neurons and inhibitory interneurons. One of the prime functions of interneurons is to stabilize the circuit by feedback inhibition, but the level of specificity on which inhibitory feedback operates is not fully resolved. We hypothesized that inhibitory circuits could enable separate feedback control loops for different synaptic input streams, by means of specific feedback inhibition to different neuronal compartments. To investigate this hypothesis, we adopted an optimization approach. Leveraging recent advances in training spiking network models, we optimized the connectivity and short-term plasticity of interneuron circuits for compartment-specific feedback inhibition onto pyramidal neurons. Over the course of the optimization, the interneurons diversified into two classes that resembled parvalbumin (PV) and somatostatin (SST) expressing interneurons. The resulting circuit can be understood as a neural decoder that inverts the nonlinear biophysical computations performed within the pyramidal cells. Our model provides a proof of concept for studying structure-function relations in cortical circuits by a combination of gradient-based optimization and biologically plausible phenomenological models

The success of deep learning sparked interest in whether the brain learns by using similar techniques for assigning credit to each synaptic weight for its contribution to the network output. However, the majority of current attempts at biologically-plausible learning methods are either non-local in time, require highly specific connectivity motives, or have no clear link to any known mathematical optimization method. Here, we introduce Deep Feedback Control (DFC), a new learning method that uses a feedback controller to drive a deep neural network to match a desired output target and whose control signal can be used for credit assignment. The resulting learning rule is fully local in space and time and approximates Gauss-Newton optimization for a wide range of feedback connectivity patterns. To further underline its biological plausibility, we relate DFC to a multi-compartment model of cortical pyramidal neurons with a local voltage-dependent synaptic plasticity rule, consistent with recent theories of dendritic processing. By combining dynamical system theory with mathematical optimization theory, we provide a strong theoretical foundation for DFC that we corroborate with detailed results on toy experiments and standard computer-vision benchmarks.

Maintaining average activity level within a set-point range constitutes a fundamental property of central neural circuits. Accumulated evidence suggests that firing rate distributions and their means represent physiological variables regulated by homeostatic systems during sleep-wake cycle in central neural circuits. While intracellular Ca2+ has long been hypothesized as a feedback control signal, the source of Ca2+ and the molecular machinery enabling network-wide homeostatic responses remain largely unknown. I will present our hypothesis and framework on identifying homeostatic regulators in neural circuits. Next, I will show our new results on the role of mitochondria in the regulation of activity set-points and feedback responses. Finally, I will provide an evidence on state-dependent dysregulation of activity set-points at the presymptomatic disease stage in familial Alzheimer’s models.

General anesthesia is a drug-induced, reversible condition comprised of five behavioral states: unconsciousness, amnesia (loss of memory), antinociception (loss of pain sensation), akinesia (immobility), and hemodynamic stability with control of the stress response. Our work shows that a primary mechanism through which anesthetics create these altered states of arousal is by initiating and maintaining highly structured oscillations. These oscillations impair communication among brain regions. We illustrate this effect by presenting findings from our human studies of general anesthesia using high-density EEG recordings and intracranial recordings. These studies have allowed us to give a detailed characterization of the neurophysiology of loss and recovery of consciousness due to propofol. We show how these dynamics change systematically with different anesthetic classes and with age. As a consequence, we have developed a principled, neuroscience-based paradigm for using the EEG to monitor the brain states of patients receiving general anesthesia. We demonstrate that the state of general anesthesia can be rapidly reversed by activating specific brain circuits. Finally, we demonstrate that the state of general anesthesia can be controlled using closed loop feedback control systems. The success of our research has depended critically on tight coupling of experiments, signal processing research and mathematical modeling.

Brain-machine interfaces (BMIs) change how the brain sends and receives information from the environment, opening new ways to probe brain function. For instance, motor BMIs allow us to precisely define and manipulate the sensorimotor loop which has enabled new insights into motor control and learning. In this talk, I’ll first present an example study where sensory-motor loop manipulations in BMI allowed us to probe feed-forward and feedback control mechanisms in ways that are not possible in the natural motor system. This study shed light on sensorimotor processing, and in turn led to state-of-the-art neural interface performance. I’ll then survey recent work that highlights the likelihood that BMIs, much like natural motor learning, engages multiple distributed learning mechanisms that can be carefully interrogated with BMI.

Animals rely on an internal sense of body position and movement to effectively control motor behaviour. This sense of proprioception is mediated by diverse populations of internal mechanosensory neurons distributed throughout the body. My lab is trying to understand how proprioceptive stimuli are detected by sensory neurons, integrated and transformed in central circuits, and used to guide motor output. We approach these questions using genetic tools, in vivo two-photon imaging, and patch-clamp electrophysiology in Drosophila. We recently found that the axons of fly leg proprioceptors are organized into distinct functional projections that contain topographic representations of specific kinematic features: one group of axons encodes tibia position, another encodes movement direction, and a third encodes bidirectional movement and vibration frequency. Whole-cell recordings from downstream neurons reveal that position, movement, and directional information remain segregated in central circuits. These feedback signals then converge upon motor neurons that control leg muscles during walking. Overall, our findings reveal how a low-dimensional stimulus – the angle of a single leg joint – is encoded by a diverse population of mechanosensory neurons. Specific proprioceptive parameters are initially processed by parallel pathways, but are ultimately integrated to influence motor output. This architecture may help to maximize information transmission, processing speed, and robustness, which are critical for feedback control of the limbs during adaptive locomotion.