Efficient scaling of large language models with mixture of experts and 3D analog in-memory computing.

Large language models (LLMs), with their remarkable generative capacities, have greatly impacted a range of fields, but they face scalability challenges due to their large parameter counts, which result in high costs for training and inference. The trend of increasing model sizes is exacerbating these challenges, particularly in terms of memory footprint, latency and energy consumption. Here we explore the deployment of 'mixture of experts' (MoEs) networks-networks that use conditional computing to keep computational demands low despite having many parameters-on three-dimensional (3D) non-volatile memory (NVM)-based analog in-memory computing (AIMC) hardware.

Hardware-aware training for large-scale and diverse deep learning inference workloads using in-memory computing-based accelerators.

Analog in-memory computing-a promising approach for energy-efficient acceleration of deep learning workloads-computes matrix-vector multiplications but only approximately, due to nonidealities that often are non-deterministic or nonlinear. This can adversely impact the achievable inference accuracy. Here, we develop an hardware-aware retraining approach to systematically examine the accuracy of analog in-memory computing across multiple network topologies, and investigate sensitivity and robustness to a broad set of nonidealities.

Using convolutional neural network to analyze brain MRI images for predicting functional outcomes of stroke.

Nowadays, the physicians usually predict functional outcomes of stroke based on clinical experiences and big data, so we wish to develop a model to accurately identify imaging features for predicting functional outcomes of stroke patients. Using magnetic resonance imaging of ischemic and hemorrhagic stroke patients, we developed and trained a VGG-16 convolutional neural network (CNN) to predict functional outcomes after 28-day hospitalization. A total of 44 individuals (24 men and 20 women) were recruited from Taoyuan General Hospital and China Medical University Hsinchu Hospital to enroll in the study.

Optimised weight programming for analogue memory-based deep neural networks.

Analogue memory-based deep neural networks provide energy-efficiency and per-area throughput gains relative to state-of-the-art digital counterparts such as graphics processing units. Recent advances focus largely on hardware-aware algorithmic training and improvements to circuits, architectures, and memory devices. Optimal translation of software-trained weights into analogue hardware weights-given the plethora of complex memory non-idealities-represents an equally important task.

Development of a Flexible Metamaterial Film with High EM Wave Absorptivity by Numerical and Experimental Methods.

The present study is intended to develop and test a cost-effective and efficient printing method for fabricating flexible metamaterial film with high electromagnetic wave absorptivity. The film can be easily applied to the surfaces with curved aspects. Firstly, numerical parametric study of the absorption characteristics of the film is performed for the range of frequency varying from 2.

Three-Dimensional Printing of a Hybrid Bioceramic and Biopolymer Porous Scaffold for Promoting Bone Regeneration Potential.

In this study, we proposed a three-dimensional (3D) printed porous (termed as 3DPP) scaffold composed of bioceramic (beta-tricalcium phosphate (β-TCP)) and thermoreversible biopolymer (pluronic F-127 (PF127)) that may provide bone tissue ingrowth and loading support for bone defect treatment. The investigated scaffolds were printed in three different ranges of pore sizes for comparison (3DPP-1: 150−200 μm, 3DPP-2: 250−300 μm, and 3DPP-3: 300−350 μm). The material properties and biocompatibility of the 3DPP scaffolds were characterized using scanning electron microscopy, X-ray diffractometry, contact angle goniometry, compression testing, and cell viability assay.

An Innovative Customized Biomimetic Hydrogel for Drug Screening Application Potential: Biocompatibility and Cell Invasion Ability.

The ability of Pluronic F127 (PF127) conjugated with tetrapeptide Gly-Arg-Gly-Asp (GRGD) as a sequence of Arg-Gly-Asp (RGD) peptide to form the investigated potential hydrogel (hereafter referred to as 3DG bioformer (3BE)) to produce spheroid, biocompatibility, and cell invasion ability, was assessed in this study. The fibroblast cell line (NIH 3T3), osteoblast cell line (MG-63), and human breast cancer cell line (MCF-7) were cultured in the 3BE hydrogel and commercial product (Matrigel) for comparison. The morphology of spheroid formation was evaluated via optical microscopy.

Toward Software-Equivalent Accuracy on Transformer-Based Deep Neural Networks With Analog Memory Devices.

Recent advances in deep learning have been driven by ever-increasing model sizes, with networks growing to millions or even billions of parameters. Such enormous models call for fast and energy-efficient hardware accelerators. We study the potential of Analog AI accelerators based on Non-Volatile Memory, in particular Phase Change Memory (PCM), for software-equivalent accurate inference of natural language processing applications.

Measurement of optic nerve sheath diameter by ultrasound in healthy term neonates.

Background: Optic nerve sheath diameter (ONSD) ultrasound is a noninvasive and repeatable tool to dynamically evaluate intracranial pressure with high diagnostic accuracy; however, data in neonates are scarce. The aim of this study was to determine the reference value of ONSD and potential influencing factors in healthy term neonates.

Methods: We retrospectively reviewed 250 full-term neonates who underwent cranial ultrasound as part of selective newborn screening over a 2-year period.

Risk of Thromboembolism in Non-Valvular Atrial Fibrillation With or Without Clinical Hyperthyroidism.

Background: Patients with hyperthyroidism have higher risk of atrial fibrillation (AF). However, the risk of thromboembolic event in patients with hyperthyroidism-related AF is controversial.

Objectives: The aim of the study was to examine the risk of thromboembolic events in AF patients with/without hyperthyroidism.

hnRNPK S379 phosphorylation participates in migration regulation of triple negative MDA-MB-231 cells.

We have previously identified a novel Aurora-A-mediated Serine 379 (S379) phosphorylation of a poly(C)-binding protein, hnRNPK, the overexpression of which is frequently observed in various cancers. It is known that the oncogenic Aurora-A kinase promotes the malignancy of cancer cells. This study aims to investigate the unexplored functions of hnRNPK S379 phosphorylation using MDA-MB-231 cells, a triple negative breast cancer cell that has amplification of the Aurora-A kinase gene.

Recreational nitrous oxide abuse related subacute combined degeneration of the spinal cord in adolescents - A case series and literature review.

Background: Nitrous oxide (NO) is a commonly used inhaled anesthetic in outpatient dental procedures. However, the increasing recreational use of NO may result in vitamin B deficiency-related neurologic and psychiatric symptoms. The aim of this study was to demonstrate the clinical features of chronic NO abuse in pediatric patients.

Training fully connected networks with resistive memories: impact of device failures.

Hardware accelerators based on two-terminal non-volatile memories (NVMs) can potentially provide competitive speed and accuracy for the training of fully connected deep neural networks (FC-DNNs), with respect to GPUs and other digital accelerators. We recently proposed [S. Ambrogio et al.

Equivalent-accuracy accelerated neural-network training using analogue memory.

Neural-network training can be slow and energy intensive, owing to the need to transfer the weight data for the network between conventional digital memory chips and processor chips. Analogue non-volatile memory can accelerate the neural-network training algorithm known as backpropagation by performing parallelized multiply-accumulate operations in the analogue domain at the location of the weight data. However, the classification accuracies of such in situ training using non-volatile-memory hardware have generally been less than those of software-based training, owing to insufficient dynamic range and excessive weight-update asymmetry.

Perovskite Quantum Dots and Their Application in Light-Emitting Diodes.

Perovskite quantum dots (PQDs) attract significant interest in recent years because of their unique optical properties, such as tunable wavelength, narrow emission, and high photoluminescence quantum efficiency (PLQY). Recent studies report new types of formamidinium (FA) PbBr PQDs, PQDs with organic-inorganic mixed cations, divalent cation doped colloidal CsPb M Br PQDs (M = Sn , Cd , Zn , Mn ) featuring partial cation exchange, and heterovalent cation doped into PQDs (Bi ). These PQD analogs open new possibilities for optoelectronic devices.

High-Performance CsPb Sn Br Perovskite Quantum Dots for Light-Emitting Diodes.

All inorganic CsPbBr perovskite quantum dots (QDs) are potential emitters for electroluminescent displays. We have developed a facile hot-injection method to partially replace the toxic Pb with highly stable Sn . Meanwhile, the absolute photoluminescence quantum yield of CsPb Sn Br increased from 45 % to 83 % with Sn substitution.

Gadolinium-doped iron oxide nanoparticles induced magnetic field hyperthermia combined with radiotherapy increases tumour response by vascular disruption and improved oxygenation.

The gadolinium-doped iron oxide nanoparticles (GdIONP) with greater specific power adsorption rate (SAR) than FeO was developed and its potential application in tumour therapy and particle tracking were demonstrated in transgenic adenocarcinoma of the mouse prostate C1 (TRAMP-C1) tumours. The GdIONPs accumulated in tumour region during the treatment could be clearly tracked and quantified by T2-weighted MR imaging. The therapeutic effects of GdIONP-mediated hyperthermia alone or in combination with radiotherapy (RT) were also evaluated.

Early appendectomy shortens antibiotic course and hospital stay in children with early perforated appendicitis.

Background: The optimal management of perforated appendicitis in the pediatric population has been controversial. This study aimed to compare the therapeutic efficacy between conservative treatment (CS) and early appendectomy (EA) in pediatric perforated appendicitis, and to determine whether surgical intervention is an optimal treatment modality for early perforated appendicitis in children.

Methods: Patients treated between January 2012 and April 2014, aged 0-18 years, with an imaging-based diagnosis of perforated appendicitis were retrospectively reviewed.

Two-dimensional pattern formation using graphoepitaxy of PS-b-PMMA block copolymers for advanced FinFET device and circuit fabrication.

Directed self-assembly (DSA) of lamellar phase block-co-polymers (BCPs) can be used to form nanoscale line-space patterns. However, exploiting the potential of this process for circuit relevant patterning continues to be a major challenge. In this work, we propose a way to impart two-dimensional pattern information in graphoepitaxy-based lamellar phase DSA processes by utilizing the interactions of the BCP with the template pattern.

Large-scale plasmonic microarrays for label-free high-throughput screening.

Microarrays allowing simultaneous analysis of thousands of parameters can significantly accelerate screening of large libraries of pharmaceutical compounds and biomolecular interactions. For large-scale studies on diverse biomedical samples, reliable, label-free, and high-content microarrays are needed. In this work, using large-area plasmonic nanohole arrays, we demonstrate for the first time a large-scale label-free microarray technology with over one million sensors on a single microscope slide.

Comprehensive MDCT evaluation of patients with pulmonary hypertension: diagnosing underlying causes with the updated Dana Point 2008 classification.

Objective: Pulmonary hypertension is a challenge for imagers and clinicians, with a variety of possible underlying causes, each with its own specific treatment. Although the diagnosis is based on physiologic measurements, ECG-gated MDCT can play a vital role in elucidating underlying cardiac, vascular, and pulmonary causes.

Conclusion: A revised system for pulmonary hypertension, the Dana Point classification, can provide a template for review of the myriad causes of this complex condition.

Partial ring artifact on cardiac CT: image presentation and clinical implication.

In computed tomography (CT), the term "artifact" is applied to any systematic discrepancy between the CT numbers in the reconstructed image and the true attenuation coefficients of the object. A partial ring artifact in cardiac CT has a unique pattern and has not been previously reported in real clinical practice. In this report, we will demonstrate the images and animations of a cardiac CT which is affected by a partial ring artifact due to a broken detector module.

Parallel scanning-optical nanoscopy with optically confined probes.

We report the imaging of sub-diffraction limited features using an optical probe generated by focusing a round spot at one wavelength, lambda(1) = 405 nm, and a ring-shaped spot at a second wavelength, lambda(2) = 532 nm, onto a thin photochromic layer that coats the nanostructures. Illumination at lambda(2) turns the photochromic layer opaque to lambda(1) everywhere except at the centre of the ring, where the illumination at lambda(1) penetrates and probes the underlying nanostructure. We confirm that this optically confined probe increases image contrast and is able to resolve features smaller than the far-field diffraction limit.

Confining light to deep subwavelength dimensions to enable optical nanopatterning.

In the past, the formation of microscale patterns in the far field by light has been diffractively limited in resolution to roughly half the wavelength of the radiation used. Here, we demonstrate lines with an average width of 36 nanometers (nm), about one-tenth the illuminating wavelength lambda1 = 325 nm, made by applying a film of thermally stable photochromic molecules above the photoresist. Simultaneous irradiation of a second wavelength, lambda2 = 633 nm, renders the film opaque to the writing beam except at nodal sites, which let through a spatially constrained segment of incident lambda1 light, allowing subdiffractional patterning.