Memory Reduction Research Articles

Tensor networks enable the calculation of turbulence probability distributions.

Predicting the dynamics of turbulent fluids has been an elusive goal for centuries. Even with modern computers, anything beyond the simplest turbulent flows is too chaotic and multiscaled to be directly simulatable. An alternative is to treat turbulence probabilistically, viewing flow properties as random variables distributed according to joint probability density functions (PDFs). Such PDFs are neither chaotic nor multiscale, yet remain challenging to simulate due to their high dimensionality. Here, we overcome the dimensionality problem by encoding turbulence PDFs as highly compressed "tensor networks" (TNs). This enables single CPU core simulations that would otherwise be impractical even with supercomputers: for a 5 + 1 dimensional PDF of a chemically reactive turbulent flow, we achieve reductions in memory and computational costs by factors of [Formula: see text] and [Formula: see text], respectively, compared to standard finite-difference algorithms. A future path is opened toward something heretofore thought infeasible: directly simulating high-dimensional PDFs of both turbulent flows and other chaotic systems that can usefully be described probabilistically.

Morroniside ameliorates sevoflurane anesthesia-induced cognitive dysfunction in aged mice through modulating the TLR4/NF-κB pathway.

Morroniside (Mor) is a bioactive compound in Cornus officinalis with anti-inflammatory, neuroprotective and antioxidant properties. Prolonged use of the anesthetic sevoflurane (Sev) has been connected to the development postoperative cognitive dysfunction (POCD). This research aims to elucidate the mechanism of action of Mor to improve cognitive impairment. A model of cognitive dysfunction induced by Sev was established in aged mice and tested for behavioral analysis using the water maze experiment. Histopathological changes and neuronal apoptosis in mouse hippocampus were observed by hematoxylin and eosin (HE) staining, Nissl staining, and TUNEL staining. ELISA and qRT-PCR determined the levels of inflammatory factors. Phenotypic transformation of microglia in hippocampal tissue was assessed by immunofluorescence, flow cytometry, and qRT-PCR. The interaction between Mor and TLR4 was analyzed using molecular docking. Western blot identified the levels of apoptosis-related proteins, synapse-related proteins, and TLR4/NF-κB pathway proteins. Inhalation of Sev caused a notable reduction in learning and spatial memory in old mice, which was dose-dependently ameliorated by Mor. Mor inhibited neuroinflammation, modulated the polarization state of hippocampal microglia, promoted their polarization to M2 type, alleviated Sev-induced hippocampal tissue damage and neuronal apoptosis. Notably, Mor can bind well with TLR4 and reduce TLR4-positive expression. Mor blocked Sev-induced TLR4/NF-κB pathway activation in hippocampal tissues, and the TLR4 agonist CRX-527 attenuated the effect of Mor. In conclusion, Mor blocked the TLR4/NF-κB pathway, reducing hippocampal tissue damage and neuroinflammation caused by Sev, which in turn improving cognitive impairment in aged mice.

Loss of the mitochondrial protein mitoNEET in mice is associated with cognitive impairments and increased neuroinflammation.

Mitochondrial dysfunction is implicated in several neurodegenerative diseases associated with memory and cognitive deficits, including Alzheimer's disease. Changes in bioenergetic function results in reactive oxygen species, oxidative damage and consequently neuroinflammation, which contributes to neuronal cell loss. In this study, we evaluated the impact of the loss of the redox active [2Fe-2S] mitochondrial-associated protein mitoNEET (CISD1) on neuroinflammation and cognition using an age-appropriate preclinical model. While associations between neuroinflammation and poor cognitive impacts have been shown in recent work, little has been done to assess whether loss of mitoNEET is associated with changes in neuroinflammatory markers or negative cognitive-behavioral outcomes. Using 9-11-month-old mitoNEET knockout (CISD1-/-) and wild-type mice, we conducted a battery of cognitive tests to assess the impact of mitoNEET loss on performance. We then histologically evaluated the effect of absence of mitoNEET on markers of neuroinflammation in the aged brain. We found loss of mitoNEET in mice was associated with a significant reduction in willingness to explore within an open field and impaired short-term spatial working memory in the Y-maze. We also found a significant reduction in novel object recognition memory that was gene-dependent and accompanied by reduced c-fos expression in hippocampus and cortical regions. Our findings indicate that mitoNEET loss is significantly associated with impairments in cognitive-behavioral and neuroinflammatory outcomes; specifically, learning and memory, anxiety-like behaviors, neuroinflammation, and neural activation. This is the first study to demonstrate cognitive-associated behavioral deficits with neuroinflammation in the mitoNEET knockout mouse model.

Sparsified time-dependent Fourier neural operators for fusion simulations

This paper presents a sparsified Fourier neural operator for coupled time-dependent partial differential equations (ST-FNO) as an efficient machine learning surrogate for fluid and particle-based fusion codes such as NIMROD (Non-Ideal Magnetohydrodynamics with Rotation - Open Discussion) and GTC (Gyrokinetic Toroidal Code). ST-FNO leverages the structures in the governing equations and utilizes neural operators to represent Green's function-like numerical operators in the corresponding numerical solvers. Once trained, ST-FNO can rapidly and accurately predict dynamics in fusion devices compared with first-principle numerical algorithms. In general, ST-FNO represents an efficient and accurate machine learning surrogate for numerical simulators for multi-variable nonlinear time-dependent partial differential equations, with the proposed architectures and loss functions. The efficacy of ST-FNO has been demonstrated using quiescent H-mode simulation data from NIMROD and kink-mode simulation data from GTC. The ST-FNO H-mode results show orders of magnitude reduction in memory and central processing unit usage in comparison with the numerical solvers in NIMROD when computing fields over a selected poloidal plane. The ST-FNO kink-mode results achieve a factor of 2 reduction in the number of parameters compared to baseline FNO models without accuracy loss.

Ruxolitinib Treatment Ameliorates Clinical, Immunological, and Transcriptomic Aberrations in Patients with STAT3 Gain-of-Function Disease

BackgroundSTAT3 gain-of-function (GOF) disease presents with lymphoproliferation, autoimmunity, and failure to thrive. While JAK inhibitors have alleviated symptoms, their effects on disease pathogenesis remain unclear. ObjectiveWe prospectively investigated the clinical, immunological, and transcriptomic responses of four STAT3 GOF patients under long-term ruxolitinib treatment. MethodsWe conducted clinical and immunological evaluations at baseline and after ruxolitinib treatment at 3, 8, 12, and over 12 months. Our assessments included circulating T follicular helper cells (cTFH), regulatory T (Treg) cells, cytokine levels, and proliferation assays. Furthermore, we investigated the transcriptomic changes with treatment and conducted T-cell receptor sequencing. ResultsRuxolitinib achieved substantial control over the clinical manifestations. Post-treatment evaluations demonstrated a notable increase in naive CD4+ and CD8+ T cell populations, alongside a significant reduction in effector memory T cells. Additionally, there was a decrease in cTFH cells and double-negative T cells. Treg cell percentages and their canonical markers, already reduced before treatment, further declined with ruxolitinib. The treatment did not alter IFN-γ, IL-17A, IL-10, IL-4 cytokine productions of CD4+ T cells. Importantly, ruxolitinib effectively normalized the previously dysregulated transcriptome profile in peripheral blood mononuclear cells, bringing it closer to that of healthy controls. This normalization was most striking in the downregulation of STAT3-targeted genes, interferon-related genes, myeloid cell activation, and cytotoxic effector CD8+ T-cell genes, with effects persisting for up to 12 months. Self-reactive T-cell indices based on TCR repertoire analysis revealed potential auto-reactive cell clones in the patient samples. ConclusionRuxolitinib reversed cellular and transcriptomic signatures, enhancing our understanding of the disease's pathophysiology and highlighting essential immunological markers for precise monitoring.

Late-Life Alcohol Exposure Does Not Exacerbate Age-Dependent Reductions in Mouse Spatial Memory and Brain TFEB Activity.

Alcohol consumption is believed to affect Alzheimer's disease (AD) risk, but the contributing mechanisms are not well understood. A potential mediator of the proposed alcohol-AD connection is autophagy, a degradation pathway that maintains organelle and protein homeostasis. Autophagy is regulated through the activity of Transcription factor EB (TFEB), which promotes lysosome and autophagy-related gene expression. The purpose of this study is to explore whether chronic alcohol consumption worsens the age-related decline in TFEB-mediated lysosomal biogenesis in the brain and exacerbates cognitive decline associated with aging. To explore the effect of alcohol on brain TFEB and autophagy, we exposed young (3-month-old) and aged (23-month-old) mice to two alcohol-feeding paradigms and assessed biochemical, transcriptome, histology, and behavioral endpoints. In young mice, alcohol decreased hippocampal nuclear TFEB staining but increased SQSTM1/p62, LC3-II, ubiquitinated proteins, and phosphorylated Tau. Hippocampal TFEB activity was lower in aged mice than it was in young mice, and Gao-binge alcohol feeding did not worsen the age-related reduction in TFEB activity. Morris Water and Barnes Maze spatial memory tasks were used to characterize the effects of aging and chronic alcohol exposure (mice fed alcohol for 4 weeks). The aged mice showed worse spatial memory acquisition in both tests. Alcohol feeding slightly impaired spatial memory in the young mice, but had little effect or even slightly improved spatial memory acquisition in the aged mice. In conclusion, aging produces greater reductions in brain autophagy flux and impairment of spatial memory than alcohol consumption.

Multi-Resolution Real-Time Deep Pose-Space Deformation

We present a hard-real-time multi-resolution mesh shape deformation technique for skeleton-driven soft-body characters. Producing mesh deformations at multiple levels of detail is very important in many applications in computer graphics. Our work targets applications where the multi-resolution shapes must be generated at fast speeds ("hard-real-time", e.g., a few milliseconds at most and preferably under 1 millisecond), as commonly needed in computer games, virtual reality and Metaverse applications. We assume that the character mesh is driven by a skeleton, and that high-quality character shapes are available in a set of training poses originating from a high-quality (slow) rig such as volumetric FEM simulation. Our method combines multi-resolution analysis, mesh partition of unity, and neural networks, to learn the pre-skinning shape deformations in an arbitrary character pose. Combined with linear blend skinning, this makes it possible to reconstruct the training shapes, as well as interpolate and extrapolate them. Crucially, we simultaneously achieve this at hard real-time rates and at multiple mesh resolution levels. Our technique makes it possible to trade deformation quality for memory and computation speed, to accommodate the strict requirements of modern real-time systems. Furthermore, we propose memory layout and code improvements to boost computation speeds. Previous methods for realtime approximations of quality shape deformations did not focus on hard real-time, or did not investigate the multi-resolution aspect of the problem. Compared to a "naive" approach of separately processing each hierarchical level of detail, our method offers a substantial memory reduction as well as computational speedups. It also makes it possible to construct the shape progressively level by level and interrupt the computation at any time, enabling graceful degradation of the deformation detail. We demonstrate our technique on several examples, including a stylized human character, human hands, and an inverse-kinematics-driven quadruped animal.

Association of chronic arsenic exposure with cellular immune profile in MINIMat adolescents: A birth cohort in Bangladesh

Chronic arsenic exposure is known to affect the immune system. We aimed to evaluate the association between arsenic exposure and immune cell profile in 15 years old adolescents (n=389) in rural Bangladesh, with chronic exposure to groundwater arsenic. Single blood and urine were collected. Urinary arsenic (U-As) concentration was measured using atomic absorption spectrometry. Peripheral blood mononuclear cells (PBMC) were analyzed by flow cytometry. Non-linear association was found between U-As (median, 24.9 µg/L) and immune cells with a cut-off at U-As 20 µg/L. U-As (<20 µg/L) were significantly associated with increases in CD8+T (21 %), naïve CD8+T (42 %) and early B cells (40 %), and classical monocytes (55 %), but reduction in CD3+T cells (37%) and intermediate-monocytes (56 %). U-As (>20 µg/L) were associated with a 3 % reduction in memory B cells. Arsenic exposure was associated with altered immune cell profile in adolescents likely rendering them vulnerable to adverse health effects in later life.

Benchmarking In-Sensor Machine Learning Computing: An Extension to the MLCommons-Tiny Suite

This paper proposes a new benchmark specifically designed for in-sensor digital machine learning computing to meet an ultra-low embedded memory requirement. With the exponential growth of edge devices, efficient local processing is essential to mitigate economic costs, latency, and privacy concerns associated with the centralized cloud processing. Emerging intelligent sensors equipped with computing assets to run neural network inferences and embedded in the same package, which hosts the sensing elements, present new challenges due to their limited memory resources and computational skills. This benchmark evaluates models trained with Quantization Aware Training (QAT) and compares their performance with Post-Training Quantization (PTQ) across three use cases: Human Activity Recognition (HAR) by means of the SHL dataset, Physical Activity Monitoring (PAM) by means of the PAMAP2 dataset, and superficial electromyography (sEMG) regression with the NINAPRO DB8 dataset. The results demonstrate the effectiveness of QAT over PTQ in most scenarios, highlighting the potential for deploying advanced AI models on highly resource-constrained sensors. The INT8 versions of the models always outperformed their FP32, regarding memory and latency reductions, except for the activations for CNN. The CNN model exhibited reduced memory usage and latency with respect to its Dense counterpart, allowing it to meet the stringent 8KiB data RAM and 32 KiB program RAM limits of the ISPU. The TCN model proved to be too large to fit within the memory constraints of the ISPU, primarily due to its greater capacity in terms of number of parameters, designed for processing more complex signals like EMG. This benchmark aims to guide the development of efficient AI solutions for In-Sensor Machine Learning Computing, fostering innovation in the field of Edge AI benchmarking, such as the one conducted by the MLCommons-Tiny working group.

ASIP Performance Enhancement by Hazard Control through Scoreboard.

The application-specific instruction set processor (ASIP) has been gradually accepted in AI, communication, media, game and industry control. The digital signal processor (DSP) is a typical ASIP, whose benefits include high performance in specific domains, low power consumption, high flexibility and low silicon consumption. One of the challenges for DSP design is to handle problems induced by datapath acceleration. The datapath acceleration (instruction fusion, black box instructions) induces control complexities. To most efficiently utilize hardware, control challenges can be summarized as RAW (Read-After-Write) handling, hardware hazard handling, and WAW (Write-After-Write) handling. Both an advanced compiler and hardware hazard handler can be used as solutions. In this paper, we introduced both solutions and exposed the benefits from the hardware solution. The benefits include utilizing low silicon to achieve higher performance and program memory reduction on chip. In summary, our solution only uses 0.91% extra silicon area yet achieves 32.75% performance improvement. So, the overall performance-to-cost ratio could be evaluated as 32%.

Effect of a cash transfer intervention on memory decline and dementia probability in older adults in rural South Africa

Evidence on cash transfers as a population-level intervention to support healthy cognitive aging in low-income settings is sparse. We assessed the effect of a cash transfer intervention on cognitive aging outcomes in older South African adults. We leveraged the overlap in the sampling frames of a Phase 3 randomized cash transfer trial [HIV Prevention Trial Network (HPTN) 068, 2011-2015] and an aging cohort [Health and Aging in Africa: A Longitudinal Study of an INDEPTH Community (HAALSI), 2014-2022] in rural Mpumalanga Province, South Africa. In 2011/12, young women and their primary caregivers were randomly assigned 1:1 to receive a monthly cash transfer or control. In 2014/2015, 862 adults aged 40+ y living in trial households were enrolled in the HAALSI cohort, with cognitive data collected in three waves over 7 y. We estimated the impact of the intervention on rate of memory decline and dementia probability scores. Memory decline in the cash transfer arm was 0.03 SD units (95% CI: 0.002, 0.05) slower per year than in the control arm. Dementia probability scores were three percentage points lower in the cash transfer arm than the control arm (β = -0.03; 95% CI: -0.05, -0.001). Effects were consistent across subgroups. A modestly sized household cash transfer delivered over a short period in mid- to later-life led to a meaningful slowing of memory decline and reduction in dementia probability 7 y later. Cash transfer programs could help stem the tide of new dementia cases in economically vulnerable populations in the coming decades.

Cognitive reserve as residual variance in cognitive performance: Latent dimensionality, correlates, and dementia prediction.

Cognitive reserve (CR) is typically operationalized as episodic memory residualized on brain health indices. The dimensionality of more generalized models of CR has rarely been examined. In a sample of N = 113 dementia-free older adults (ages 62-86 years at MRI scan; 58.4% women), the domain-specific representation of general cognition (COG) before vs. after residualization on brain indices (brain volume loss, cerebral blood flow, white matter hyperintensities) was compared (i.e., COG vs. CR). COG and CR were assessed by 15 tasks spanning five domains: processing speed, verbal memory, visuospatial memory, fluid reasoning, and vocabulary. Measurement invariance and item-construct representation were tested in a series of structural factor analyses. COG and CR were then examined in relation to 22 risk and protective factors and dementia status at time of death. Item-factor loadings differed such that CR more strongly emphasized fluid reasoning. More years of education, higher occupational class, more hobbies/interests, and fewer difficulties with personal mobility similarly predicted better COG and CR. Only the sub-domain of visuospatial memory (both before and after residualization) was associated with conversion to dementia by end-of-life (r = -.30; p = .01). Results provide tentative support for the role of fluid reasoning (intelligence) as a potential compensatory factor for age- and/or neuropathology-related reductions in processing speed and memory. Intellectually stimulating work, efforts to preserve personal mobility, and a diversity of hobbies and interests may attenuate age- and/or pathology-related reductions in cognitive functioning prior to dementia onset.

Optimizing Spatial Complexity of the Hard Decision Decoder Based on Hash and Syndrome Decoding (HSDec)

In this article, we propose an optimized version of the Hard Decision Decoder based on Hash and Syndrome Decoding (HSDec) decoder, named Reduced Memory Space of HSDec (RMS-HSDec), which uses less memory space. In this article, we aim to reduce the spatial complexity of the HSDec decoding algorithm while preserving its error correction capabilities. Our methodology involves allocating only the essential memory space for correctable error patterns and optimizing the hashing mechanism to effectively handle potential collisions. While maintaining the integrity of error correction, this new method guarantees memory reduction rates of over 96 % for the BCH(63, 39, 9) code and over 84 % for the QR(47, 24, 11) code compared to HSDec. Simulations were conducted to evaluate the performance of RMS-HSDec on various BCH and QR codes over AWGN and Rayleigh channels. The results demonstrated significant memory reduction rates and coding gains ranging from 0.8 dB to 2.8 dB over the AWGN channel and from 14 dB to 32 dB over the Rayleigh channel, confirming the robustness of the algorithm under different channel conditions. Comparative analyses showed that RMS-HSDec maintains competitive performance with existing decoders while offering effective error correction. These findings confirm the robustness of the RMS-HSDec algorithm under different channel conditions. Overall, the proposed decoder proves to be an effective solution, optimizing memory usage without compromising error correction capabilities, making it ideal for high-density data applications and environments with limited memory resources.

SpQuant-SNN: ultra-low precision membrane potential with sparse activations unlock the potential of on-device spiking neural networks applications.

Spiking neural networks (SNNs) have received increasing attention due to their high biological plausibility and energy efficiency. The binary spike-based information propagation enables efficient sparse computation in event-based and static computer vision applications. However, the weight precision and especially the membrane potential precision remain as high-precision values (e.g., 32 bits) in state-of-the-art SNN algorithms. Each neuron in an SNN stores the membrane potential over time and typically updates its value in every time step. Such frequent read/write operations of high-precision membrane potential incur storage and memory access overhead in SNNs, which undermines the SNNs' compatibility with resource-constrained hardware. To resolve this inefficiency, prior works have explored the time step reduction and low-precision representation of membrane potential at a limited scale and reported significant accuracy drops. Furthermore, while recent advances in on-device AI present pruning and quantization optimization with different architectures and datasets, simultaneous pruning with quantization is highly under-explored in SNNs. In this work, we present SpQuant-SNN, a fully-quantized spiking neural network with ultra-low precision weights, membrane potential, and high spatial-channel sparsity, enabling the end-to-end low precision with significantly reduced operations on SNN. First, we propose an integer-only quantization scheme for the membrane potential with a stacked surrogate gradient function, a simple-yet-effective method that enables the smooth learning process of quantized SNN training. Second, we implement spatial-channel pruning with membrane potential prior, toward reducing the layer-wise computational complexity, and floating-point operations (FLOPs) in SNNs. Finally, to further improve the accuracy of low-precision and sparse SNN, we propose a self-adaptive learnable potential threshold for SNN training. Equipped with high biological adaptiveness, minimal computations, and memory utilization, SpQuant-SNN achieves state-of-the-art performance across multiple SNN models for both event-based and static image datasets, including both image classification and object detection tasks. The proposed SpQuant-SNN achieved up to 13× memory reduction and >4.7× FLOPs reduction with < 1.8% accuracy degradation for both classification and object detection tasks, compared to the SOTA baseline.

A combinatorial model reduction method for the finite element analysis of wind instruments

AbstractA high‐fidelity finite element model is proposed for the complete simulation of the time‐harmonic acoustic propagation in wind instruments. The challenge is to meet the extremely high accuracy required by professional musicians, in a complex domain, for all fingerings and over a wide frequency range, within an affordable computational time. Several modelling assumptions are made to limit the numerical complexity of the problem while preserving all relevant physics. A dedicated high‐performance solution strategy is also proposed, based on partitioning, condensation and model order reduction, exploiting the combinatorial nature of wind instrument fingerings. Finally, the proposed approach is applied to the simulation of an alto saxophone. An order of magnitude reduction in memory and computational cost is achieved.

Immunophenotyping and Therapeutic Insights from Chronic Mucocutaneous Candidiasis Cases with STAT1 Gain-of-Function Mutations.

Heterozygous STAT1 Gain-of-Function (GOF) mutations are the most common cause of chronic mucocutaneous candidiasis (CMC) among Inborn Errors of Immunity. Clinically, these mutations manifest as a broad spectrum of immune dysregulation, including autoimmune diseases, vascular disorders, and malignancies. The pathogenic mechanisms of immune dysregulation and its impact on immune cells are not yet fully understood. In treatment, JAK inhibitors have shown therapeutic effectiveness in some patients. We analyzed clinical presentations, cellular phenotypes, and functional impacts in five Taiwanese patients with STAT1 GOF. We identified two novel GOF mutations in 5 patients from 2 Taiwanese families, presenting with symptoms of CMC, late-onset rosacea, and autoimmunity. The enhanced phosphorylation and delayed dephosphorylation were displayed by the patients' cells. There are alterations in both innate and adaptive immune cells, including expansion of CD38+HLADR +CD8+ T cells, a skewed activated Tfh cells toward Th1, reduction of memory, marginal zone and anergic B cells, all main functional dendritic cell lineages, and a reduction in classical monocyte. Baricitinib showed therapeutic effectiveness without side effects. Our study provides the first comprehensive clinical and molecular characteristics in STAT1 GOF patient in Taiwan and highlights the dysregulated T and B cells subsets which may hinge the autoimmunity in STAT1 GOF patients. It also demonstrated the therapeutic safety and efficacy of baricitinib in pediatric patient. Further research is needed to delineate how the aberrant STAT1 signaling lead to the changes in cellular populations as well as to better link to the clinical manifestations of the disease.

DELTA: Memory-Efficient Training via Dynamic Fine-Grained Recomputation and Swapping

To accommodate the increasingly large-scale models within limited-capacity GPU memory, various coarse-grained techniques, such as recomputation and swapping, have been proposed to optimize memory usage. However, these methods have encountered limitations, either in terms of inefficient memory reduction or diminished training performance. In response to this, our paper introduces DELTA, an innovative approach for memory-efficient large-scale model training that combines fine-grained memory optimization and prefetching technology to reduce memory usage while maintaining high training throughput concurrently. Initially, we formulate the problem of memory-throughput joint optimization as an easy-solving 0/1 Knapsack problem. Leveraging this formalization, we use an improving polynomial complexity heuristic algorithm to address the problem effectively. Furthermore, we introduce a novel bidirectional prefetching technology into dynamic memory management, which significantly accelerates the model training when compared to relying solely on recomputation or swapping. Finally, DELTA offers users an automated training execution library, eliminating the need for manual configuration or specialized expertise. Experimental results demonstrate the effectiveness of DELTA in reducing GPU memory consumption. Compared to state-of-the-art methods, DELTA achieves substantial memory savings ranging from 40% to 72%, while maintaining comparable convergence performance for various models, including ResNet-50, ResNet-101, and BERT-Large. Notably, DELTA enables the training of GPT2-Large and GPT2-XL with batch sizes increased by 5.5 × and 6 ×, respectively, showcasing its versatility and practicality in enabling large-scale model training on GPU hardware.

Rapid prediction for the transient dispersion of leaked airborne pollutant in urban environment based on graph neural networks

Rapidly predicting airborne pollutant dispersion in urban is vital for ventilation design and evacuation planning. Computational fluid dynamics (CFD) simulations are commonly used to provide accurate predictions, but the computational cost is too high. Although graph neural networks (GNNs) provide fast predictions of flow fields by manipulating unstructured mesh on GPU, they suffer from high memory usage and accuracy decreases when applied to large-scale urban scenes. Moreover, it is difficult for GNNs to learn the coupled relationship between wind field and pollutant concentration field. We propose a multi-objective GNN model as CFD surrogate to rapidly predict the transient dispersion of airborne pollutant under the influence of complex wind field patterns in urban environment. Based on random urban layouts generated by a 2D bin packing algorithm, we employ a validated CFD model to construct a sample dataset of wind fields and concentration fields. We leverage graph pooling and multi-scale feature fusion to improve prediction accuracy, and subgraph partitioning of both wind field and concentration field to reduce GPU memory usage. The results show that our GNN model at its best runs 1–2 orders of magnitude faster than CFD simulation with accuracy evaluation metrics R2=0.92, and achieves 70 % GPU memory reduction.

Obesity as Inducer of Cognitive Function Decline via Dysbiosis of Gut Microbiota in Rats.

Diet-induced obesity is a global phenomenon that affects the population worldwide with manifestations at both the phenotypic and genotypic levels. Cognitive function decline is a major global health challenge. The relation between obesity and cognitive function is a debatable issue. The main goal of the current research was to study the implications of obesity on cognitive function and gut microbiota diversity and its impact on plasma and brain metabolic parameters in rats. Obesity was induced in rats by feeding on a high-fat (HF) or a high-fat/high-sucrose (HFHS) diet. The results reveal that both the HF (0.683) and HFHS (0.688) diets were effective as obesity inducers, which was confirmed by a significant increase in the body mass index (BMI). Both diet groups showed dyslipidemia and elevation of oxidative stress, insulin resistance (IR), and inflammatory markers with alterations in liver and kidney functions. Obesity led to a reduction in cognitive function through a reduction in short-term memory by 23.8% and 30.7% in the rats fed HF and HFHS diets, respectively, and learning capacity and visuo-spatial memory reduced by 8.9 and 9.7 s in the rats fed an HF or HFHS diet, respectively. Bacteroidetes, Firmicutes, Proteobacteria, Fusobacteria, and Spirochaetes phyla were detected. The Firmicutes/Bacteroidetes ratio (F/B) significantly decreased in the HF group, while it increased in the HFHS group compared to the normal control. The two species, Bacteroides acidifaciens and Bacteroides ovatus, which are associated with IR, were drastically compromised by the high-fat/high-sucrose diet. Some species that have been linked to reduced inflammation showed a sharp decrease in the HFHS group, while Prevotella copri, which is linked to carbohydrate metabolism, was highly enriched. In conclusion: Obesity led to cognitive impairment through changes in short-term and visuo-spatial memory. A metagenomic analysis revealed alterations in the abundance of some microbial taxa associated with obesity, inflammation, and insulin resistance in the HF and HFHS groups.

Olfactory Ecto-Mesenchymal Stem Cells in Modeling and Treating Alzheimer's Disease.

Alzheimer's disease (AD) is a condition in the brain that is marked by a gradual and ongoing reduction in memory, thought, and the ability to perform simple tasks. AD has a poor prognosis but no cure yet. Therefore, the need for novel models to study its pathogenesis and therapeutic strategies is evident, as the brain poorly recovers after injury and neurodegenerative diseases and can neither replace dead neurons nor reinnervate target structures. Recently, mesenchymal stem cells (MSCs), particularly those from the human olfactory mucous membrane referred to as the olfactory ecto-MSCs (OE-MSCs), have emerged as a potential avenue to explore in modeling AD and developing therapeutics for the disease due to their lifelong regeneration potency and facile accessibility. This review provides a comprehensive summary of the current literature on isolating OE-MSCs and delves into whether they could be reliable models for studying AD pathogenesis. It also explores whether healthy individual-derived OE-MSCs could be therapeutic agents for the disease. Despite being a promising tool in modeling and developing therapies for AD, some significant issues remain, which are also discussed in the review.