Impact of On-chip Interconnect on In-memory Acceleration of Deep Neural Networks

With the widespread use of Deep Neural Networks (DNNs), machine learning algorithms have evolved in two diverse directions—one with ever-increasing connection density for better accuracy and the other with more compact sizing for energy efficiency. The increase in connection density increases on-chip data movement, which makes efficient on-chip communication a critical function of the DNN accelerator. The contribution of this work is threefold. First, we illustrate that the point-to-point (P2P)-based interconnect is incapable of handling a high volume of on-chip data movement for DNNs. Second, we evaluate P2P and network-on-chip (NoC) interconnect (with a regular topology such as a mesh) for SRAM- and ReRAM-based in-memory computing (IMC) architectures for a range of DNNs. This analysis shows the necessity for the optimal interconnect choice for an IMC DNN accelerator. Finally, we perform an experimental evaluation for different DNNs to empirically obtain the performance of the IMC architecture with both NoC-tree and NoC-mesh. We conclude that, at the tile level, NoC-tree is appropriate for compact DNNs employed at the edge, and NoC-mesh is necessary to accelerate DNNs with high connection density. Furthermore, we propose a technique to determine the optimal choice of interconnect for any given DNN. In this technique, we use analytical models of NoC to evaluate end-to-end communication latency of any given DNN. We demonstrate that the interconnect optimization in the IMC architecture results in up to 6 × improvement in energy-delay-area product for VGG-19 inference compared to the state-of-the-art ReRAM-based IMC architectures.

Control of Transcranial Direct Current Stimulation Duration by Assessing Functional Connectivity of Near-Infrared Spectroscopy Signals

Transcranial direct current stimulation (tDCS) has been shown to create neuroplasticity in healthy and diseased populations. The control of stimulation duration by providing real-time brain state feedback using neuroimaging is a topic of great interest. This study presents the feasibility of a closed-loop modulation for the targeted functional network in the prefrontal cortex. We hypothesize that we cannot improve the brain state further after reaching a specific state during a stimulation therapy session. A high-definition tDCS of 1[Formula: see text]mA arranged in a ring configuration was applied at the targeted right prefrontal cortex of 15 healthy male subjects for 10[Formula: see text]min. Functional near-infrared spectroscopy was used to monitor hemoglobin chromophores during the stimulation period continuously. The correlation matrices obtained from filtered oxyhemoglobin were binarized to form subnetworks of short- and long-range connections. The connectivity in all subnetworks was analyzed individually using a new quantification measure of connectivity percentage based on the correlation matrix. The short-range network in the stimulated hemisphere showed increased connectivity in the initial stimulation phase. However, the increase in connection density reduced significantly after 6[Formula: see text]min of stimulation. The short-range network of the left hemisphere and the long-range network gradually increased throughout the stimulation period. The connectivity percentage measure showed a similar response with network theory parameters. The connectivity percentage and network theory metrics represent the brain state during the stimulation therapy. The results from the network theory metrics, including degree centrality, efficiency, and connection density, support our hypothesis and provide a guideline for feedback on the brain state. The proposed neuro-feedback scheme is feasible to control the stimulation duration to avoid overdosage.

Structured patterns of activity in pulse-coupled oscillator networks with varied connectivity

Identifying coordinated activity within complex systems is essential to linking their structure and function. We study collective activity in networks of pulse-coupled oscillators that have variable network connectivity and integrate-and-fire dynamics. Starting from random initial conditions, we see the emergence of three broad classes of behaviors that differ in their collective spiking statistics. In the first class (“temporally-irregular”), all nodes have variable inter-spike intervals, and the resulting firing patterns are irregular. In the second (“temporally-regular”), the network generates a coherent, repeating pattern of activity in which all nodes fire with the same constant inter-spike interval. In the third (“chimeric”), subgroups of coherently-firing nodes coexist with temporally-irregular nodes. Chimera states have previously been observed in networks of oscillators; here, we find that the notions of temporally-regular and chimeric states encompass a much richer set of dynamical patterns than has yet been described. We also find that degree heterogeneity and connection density have a strong effect on the resulting state: in binomial random networks, high degree variance and intermediate connection density tend to produce temporally-irregular dynamics, while low degree variance and high connection density tend to produce temporally-regular dynamics. Chimera states arise with more frequency in networks with intermediate degree variance and either high or low connection densities. Finally, we demonstrate that a normalized compression distance, computed via the Lempel-Ziv complexity of nodal spike trains, can be used to distinguish these three classes of behavior even when the phase relationship between nodes is arbitrary.

NB-IoT and LTE-M towards massive MTC: Complete performance evaluation for 5G mMTC

Since the emerging 5G wireless network is expected to significantly revolutionize thefield of communication, its standardization and design should regard the internet ofthings (IoT) among the main orientations. Also, emerging IoT applications introducenew requirements other than throughput to support massive machine-type commu-nication (mMTC) where small data packets are occasionally sent. Therefore, moreimportance is attached to coverage, latency, power consumption, and connection den-sity. For this purpose, the third generation partnership project (3GPP) has introducedtwo novel cellular IoT technologies supporting mMTC, known as NB-IoT and LTE-M. This paper aims to determine the system configuration and deployment required forNB-IoT and LTE-M technologies to fully meet the 5G mMTC requirements in termsof coverage, throughput, latency, battery life, and connection density. An overview ofthese technologies and their design principles is also described. A complete evalua-tion of NB-IoT and LTE-M performance against 5G mMTC requirements is presented,and it is shown that these requirements can be met but only under certain conditionsregarding system configuration and deployment. This is followed by a performancecomparative analysis, which is mainly conducted to determine the limits and suitableuse cases of each technology.

Case Report: An MRI Traumatic Brain Injury Longitudinal Case Study at 7 Tesla: Pre- and Post-injury Structural Network and Volumetric Reorganization and Recovery

Importance: A significant limitation of many neuroimaging studies examining mild traumatic brain injury (mTBI) is the unavailability of pre-injury data.Objective: We therefore aimed to utilize pre-injury ultra-high field brain MRI and compare a collection of neuroimaging metrics pre- and post-injury to determine mTBI related changes and evaluate the enhanced sensitivity of high-resolution MRI.Design: In the present case study, we leveraged multi-modal 7 Tesla MRI data acquired at two timepoints prior to mTBI (23 and 12 months prior to injury), and at two timepoints post-injury (2 weeks and 8 months after injury) to examine how a right parietal bone impact affects gross brain structure, subcortical volumetrics, microstructural order, and connectivity.Setting: This research was carried out as a case investigation at a single primary care site.Participants: The case participant was a 38-year-old female selected for inclusion based on a mTBI where a right parietal impact was sustained.Main outcomes: The main outcome measurements of this investigation were high spatial resolution structural brain metrics including volumetric assessment and connection density of the white matter connectome.Results: At the first scan timepoint post-injury, the cortical gray matter and cerebral white matter in both hemispheres appeared to be volumetrically reduced compared to the pre-injury and subsequent post-injury scans. Connectomes produced from whole-brain diffusion-weighted probabilistic tractography showed a widespread decrease in connectivity after trauma when comparing mean post-injury and mean pre-injury connection densities. Findings of reduced fractional anisotropy in the cerebral white matter of both hemispheres at post-injury time point 1 supports reduced connection density at a microstructural level. Trauma-related alterations to whole-brain connection density were markedly reduced at the final scan timepoint, consistent with symptom resolution.Conclusions and Relevance: This case study investigates the structural effects of traumatic brain injury for the first time using pre-injury and post-injury 7 Tesla MRI longitudinal data. We report findings of initial volumetric changes, decreased structural connectivity and reduced microstructural order that appear to return to baseline 8 months post-injury, demonstrating in-depth metrics of physiological recovery. Default mode, salience, occipital, and executive function network alterations reflect patient-reported hypersomnolence, reduced cognitive processing speed and dizziness.

Subsystem macroarchitecture of the intrinsic midbrain neural network and its tectal and tegmental subnetworks

The midbrain is the smallest of three primary vertebrate brain divisions. Here we use network science tools to reveal the global organizing principles of intramidbrain axonal circuitry before adding extrinsic connections with the remaining nervous system. Curating the experimental neuroanatomical literature yielded 17,248 connection reports for 8,742 possible connections between the 94 gray matter regions forming the right and left midbrain. Evidence for the existence of 1,676 connections suggests a 19.2% connection density for this network, similar to that for the intraforebrain network [L. W. Swanson et al., Proc. Natl. Acad. Sci. U.S.A. 117, 31470–31481 (2020)]. Multiresolution consensus cluster analysis parceled this network into a hierarchy with 6 top-level and 30 bottom-level subsystems. A structure–function model of the hierarchy identifies midbrain subsystems that play specific functional roles in sensory–motor mechanisms, motivation and reward, regulating complex reproductive and agonistic behaviors, and behavioral state control. The intramidbrain network also contains four bilateral region pairs designated putative hubs. One pair contains the superior colliculi of the tectum, well known for participation in visual sensory–motor mechanisms, and the other three pairs form spatially compact right and left units (the ventral tegmental area, retrorubral area, and midbrain reticular nucleus) in the tegmentum that are implicated in motivation and reward mechanisms. Based on the core hypothesis that subsystems form functionally cohesive units, the results provide a theoretical framework for hypothesis-driven experimental analysis of neural circuit mechanisms underlying behavioral responses mediated in part by the midbrain.

Research on the Spatial Structure of the European Union’s Tourism Economy and Its Effects

With the implementation of European integration policies such as the single market, the Euro and the Schengen Visa, the EU member states are developing closer economic ties through tourism, and their level of tourism integration is constantly improving. Taking the 28 EU member states as research objects, this paper constructs a tourism economic connection network among them, measures the strength of their tourism economic connections from 1995 to 2018 by using the modified gravity model and social network method, and analyzes the spatial structure characteristics and effects of the EU tourism economy. The results are as follows: (1) The tourism economic ties of EU member states are growing increasingly close, enhancing network stability. (2) Germany, France, Italy, Austria and the United Kingdom are the top five countries in the degree centrality and closeness centrality rankings, meaning that they are located in the center of the network and have great influence, and the network is becoming increasingly concentrated. Germany, Italy, Sweden, Austria and France play an important intermediary role in the network, and the centrality of most member states has increased. (3) The core areas are mainly concentrated in Western Europe, Southern Europe, Mediterranean mainland countries and Central Europe, while the marginal areas are mainly concentrated in Eastern Europe, Northern Europe and Mediterranean island countries; the network connection density of the core area, the network connection density of the marginal area, and the network connection density between the core and marginal area overall show an increasing trend. (4) Improvements in the complete network connectedness and a reduction in graph efficiency can significantly reduce differences in EU tourism economic development levels and improve spatial equity.