Meta-learning synaptic plasticity and memory addressing for continual familiarity detection.

Over the course of a lifetime, we process a continual stream of information. Extracted from this stream, memories must be efficiently encoded and stored in an addressable manner for retrieval. To explore potential mechanisms, we consider a familiarity detection task in which a subject reports whether an image has been previously encountered. We design a feedforward network endowed with synaptic plasticity and an addressing matrix, meta-learned to optimize familiarity detection over long intervals. We find that anti-Hebbian plasticity leads to better performance than Hebbian plasticity and replicates experimental results such as repetition suppression. A combinatorial addressing function emerges, selecting a unique neuron as an index into the synaptic memory matrix for storage or retrieval. Unlike previous models, this network operates continuously and generalizes to intervals it has not been trained on. Our work suggests a biologically plausible mechanism for continual learning and demonstrates an effective application of machine learning for neuroscience discovery.

Periodic flashing coordinated reset stimulation paradigm reduces sensitivity to ON and OFF period durations.

Pathological synchronization in the basal ganglia network has been considered an important component of Parkinson's disease pathophysiology. An established treatment for some patients with Parkinson's disease is deep brain stimulation, in which a tonic high-frequency pulse train is delivered to target regions of the brain. In recent years, a novel neuromodulation paradigm called coordinated reset stimulation has been proposed, which aims to reverse the pathological synchrony by sequentially delivering short high-frequency bursts to distinct sub-regions of the pathologically synchronized network, with an average intra-burst interval for each sub-region corresponding to period of the pathological oscillation. It has further been proposed that the resultant desynchronization can be enhanced when stimulation is interrupted periodically, and that it is particularly beneficial to precisely tune the stimulation ON and OFF time-windows to the underlying pathological frequency. Pre-clinical and clinical studies of coordinated reset stimulation have relied on these proposals for their stimulation protocols. In this study, we present a modified ON-OFF coordinated reset stimulation paradigm called periodic flashing and study its behavior through computational modeling using the Kuramoto coupled phase oscillator model. We demonstrate that in contrast to conventional coordinated reset stimulation, the periodic flashing variation does not exhibit a need for precise turning of the ON-OFF periods to the pathological frequency, and demonstrates desynchronization for a wide range of ON and OFF periods. We provide a mechanistic explanation for the previously observed sensitivities and demonstrate that they are an artifact of the specific ON-OFF cycling paradigm used. As a practical consequence, the periodic flashing paradigm simplifies the tuning of optimal stimulation parameters by decreasing the dimension of the search space. It also suggests new, more flexible ways of delivering coordinated reset stimulation.

Time domain diffuse correlation spectroscopy: modeling the effects of laser coherence length and instrument response function.

Diffuse correlation spectroscopy (DCS) is an optical technique that non-invasively quantifies an index of blood flow (BFi) by measuring the temporal autocorrelation function of the intensity fluctuations of light diffusely remitted from the tissue. Traditional DCS measurements use continuous wave (CW) lasers with coherence lengths longer than the photon path lengths in the sample to ensure that the diffusely remitted light is coherent and generates a speckle pattern. Recently, we proposed time domain DCS (TD-DCS) to allow measurements of the speckle fluctuations for specific path lengths of light through the tissue, which has the distinct advantage of permitting an analysis of selected long path lengths of light to improve the depth sensitivity of the measurement. However, compared to CW-DCS, factors including the instrument response function (IRF), the detection gate width, and the finite coherence length need to be considered in the model analysis of the experimental data. Here we present a TD-DCS model describing how the intensity autocorrelation functions measured for different path lengths of light depend on the coherence length, pulse width of the laser, detection gate width, IRF, BFi, and optical properties of the scattering sample. Predictions of the model are compared with experimental results using a homogeneous liquid phantom sample that mimics human tissue optical properties. The BFis obtained from the TD-DCS model for different path lengths of light agree with the BFi obtained from CW-DCS measurements, while the standard simplified model underestimates the BFi by a factor of ∼2. This Letter establishes the theoretical foundation of the TD-DCS technique and provides guidance for future BFi measurements in tissue.

Time-domain diffuse correlation spectroscopy.

Physiological monitoring of oxygen delivery to the brain has great significance for improving the management of patients at risk for brain injury. Diffuse correlation spectroscopy (DCS) is a rapidly growing optical technology able to non-invasively assess the blood flow index (BFi) at the bedside. The current limitations of DCS are the contamination introduced by extracerebral tissue and the need to know the tissue's optical properties to correctly quantify the BFi. To overcome these limitations, we have developed a new technology for time-resolved diffuse correlation spectroscopy. By operating DCS in the time domain (TD-DCS), we are able to simultaneously acquire the temporal point-spread function to quantify tissue optical properties and the autocorrelation function to quantify the BFi. More importantly, by applying time-gated strategies to the DCS autocorrelation functions, we are able to differentiate between short and long photon paths through the tissue and determine the BFi for different depths. Here, we present the novel device and we report the first experiments in tissue-like phantoms and in rodents. The TD-DCS method opens many possibilities for improved non-invasive monitoring of oxygen delivery in humans.

Analysis of complex neural circuits with nonlinear multidimensional hidden state models.

A universal need in understanding complex networks is the identification of individual information channels and their mutual interactions under different conditions. In neuroscience, our premier example, networks made up of billions of nodes dynamically interact to bring about thought and action. Granger causality is a powerful tool for identifying linear interactions, but handling nonlinear interactions remains an unmet challenge. We present a nonlinear multidimensional hidden state (NMHS) approach that achieves interaction strength analysis and decoding of networks with nonlinear interactions by including latent state variables for each node in the network. We compare NMHS to Granger causality in analyzing neural circuit recordings and simulations, improvised music, and sociodemographic data. We conclude that NMHS significantly extends the scope of analyses of multidimensional, nonlinear networks, notably in coping with the complexity of the brain.