Phenotypic variation of transcriptomic cell types in mouse motor cortex.

Cortical neurons exhibit extreme diversity in gene expression as well as in morphological and electrophysiological properties1,2. Most existing neural taxonomies are based on either transcriptomic3,4 or morpho-electric5,6 criteria, as it has been technically challenging to study both aspects of neuronal diversity in the same set of cells7. Here we used Patch-seq8 to combine patch-clamp recording, biocytin staining, and single-cell RNA sequencing of more than 1,300 neurons in adult mouse primary motor cortex, providing a morpho-electric annotation of almost all transcriptomically defined neural cell types. We found that, although broad families of transcriptomic types (those expressing Vip, Pvalb, Sst and so on) had distinct and essentially non-overlapping morpho-electric phenotypes, individual transcriptomic types within the same family were not well separated in the morpho-electric space. Instead, there was a continuum of variability in morphology and electrophysiology, with neighbouring transcriptomic cell types showing similar morpho-electric features, often without clear boundaries between them. Our results suggest that neuronal types in the neocortex do not always form discrete entities. Instead, neurons form a hierarchy that consists of distinct non-overlapping branches at the level of families, but can form continuous and correlated transcriptomic and morpho-electrical landscapes within families.

Fusarium graminearum rapid alkalinization factor peptide negatively regulates plant immunity and cell growth via the FERONIA receptor kinase.

The plant rapid alkalinization factor (RALF) peptides function as key regulators in cell growth and immune responses through the receptor kinase FERONIA (FER). In this study, we report that the transcription factor FgPacC binds directly to the promoter of FgRALF gene, which encodes a functional homologue of the plant RALF peptides from the wheat head blight fungus Fusarium graminearum (FgRALF). More importantly, FgPacC promotes fungal infection via host immune suppression by activating the expression of FgRALF. The FgRALF peptide also exhibited typical activities of plant RALF functions, such as inducing plant alkalinization and inhibiting cell growth, including wheat (Triticum aestivum), tomato (Solanum lycopersicum) and Arabidopsis thaliana. We further identified the wheat receptor kinase FERONIA (TaFER), which is capable of restoring the defects of the A. thaliana FER mutant. In addition, we found that FgRALF peptide binds to the extracellular malectin-like domain (ECD) of TaFER (TaFERECD) to suppress the PAMP-triggered immunity (PTI) and cell growth. Overexpression of TaFERECD in A. thaliana confers plant resistance to F. graminearum and protects from FgRALF-induced cell growth inhibition. Collectively, our results demonstrate that the fungal pathogen-secreted RALF mimic suppresses host immunity and inhibits cell growth via plant FER receptor. This establishes a novel pathway for the development of disease-resistant crops in the future without compromising their yield potential.

Responses of Chemosensory Perception to Stimulation of the Human Brain.

OBJECTIVE: Significant advances have been made in our understanding of the neural substrates of human chemosensory processing, involving the piriform cortex, insula, and orbitofrontal cortex. However, the important and challenging issues are to localize the brain regions with high anatomic precision that can causally produce chemosensory perception and further delineate the topography of different classifications of chemosensory perception. METHODS: We quantitatively measured subjective responses of chemosensory perception to intracranial electrical stimulation over the brain in neurosurgical patients (n = 302) with medically refractory epilepsy. RESULTS: The chemosensory perceptions including olfaction, gustation, and chemesthesis were elicited in 21 of 302 patients (7%). Chemosensory responses were evoked in 53 (0.2%) of 21,661 stimulated sites. The highest response rate (1.8%) was in the insula (37/2,051 stimulated sites from 15/163 patients). The chemosensory perception emerged predominantly during stimulation of the insula along the central sulcus axis. Notably, there existed a distinct pattern that the anteroventral insula predominately represented orthonasal olfaction, whereas different chemosensory modalities converged in the mid-dorsal insula. INTERPRETATION: This study provided a detailed characterization of chemosensory perception across the brain, especially in the insula. These results suggest that the cortex along the banks of the central sulcus of the insula may play a role in producing the supramodal sensation of flavor. It also indicates that dysfunction of the central insula should be considered during the evaluation of chemosensory-related epileptic seizures. ANN NEUROL 2023;93:175-183.

Active secretion of IL-33 from astrocytes is dependent on TMED10 and promotes central nervous system homeostasis.

Interleukin-33 (IL-33), secreted by astrocytes, regulates the synapse development in the spinal cord and hippocampus and suppresses autoimmune disease in the central nervous system (CNS). However, the mechanism of unconventional protein secretion of this cytokine remains unclear. In this study, we found that IFN-γ promotes the active secretion of IL-33 from astrocytes, and the active secretion of IL-33 from cytoplasm to extracellular space was dependent on interaction with transmembrane emp24 domain 10 (TMED10) via the IL-1 like cytokine domain in astrocytes. Knockout of Il-33 or its receptor St2 induced hippocampal astrocyte activation and depressive-like disorder in naive mice, as well as increased spinal cord astrocyte activation and polarization to a neurotoxic reactive subtype and aggravated passive experimental autoimmune encephalomyelitis (EAE). Our results have identified that IL-33 is actively secreted by astrocytes through the unconventional protein secretion pathway facilitated by TMED10 channels. This process helps maintain CNS homeostasis by inhibiting astrocyte activation.

Bidirectional relationship between late-onset epilepsy (LOE) and dementia: A systematic review and meta-analysis of cohort studies.

OBJECTIVE: To explore the bidirectional relationship of late-onset epilepsy (LOE) with dementia and Alzheimer's disease (AD). METHODS: Using the common electronic databases, including PubMed, Cochrane Library databases and EMBASE, we systematically reviewed published cohort studies that assessed the risk of LOE in individuals comorbid with dementia or AD, and those with dementia or AD comorbid with LOE that had been published up to 31 March 2023. The data extraction process was carried out independently by two authors. The summary adjusted relative ratio (aRR) was calculated by employing Rev Man 5.3 for the inclusion of studies. To investigate the origins of heterogeneity, we conducted both subgroup and sensitivity analyses. In the presence of heterogeneity, a random-effects model was employed. To evaluate potential publication bias, we utilized the funnel plot and conducted Begg's and Egger's tests. RESULTS: We included 20 eligible studies in the final analysis after a rigorous screening process. Pooled results indicated that LOE was association with an increased risk of all-cause dementia (aRR: 1.34, 95% confidence interval [CI]: 1.13-1.59) and AD (aRR: 2.49, 95% CI: 1.16-5.32). In addition, the pooled effect size for LOE associated with baseline AD and all-cause dementia were 3.51 (95% CI: 3.47-3.56) and 2.53 (95% CI: 2.39-2.67), respectively. Both sensitivity and subgroup analyses showed that these positive correlations persisted. According to the results of the Egger's and Begg's tests, as well as visual inspection of funnel plots, none of the studies appeared to be biased by publication. CONCLUSION: The findings suggested that LOE is a potential risk factor for dementia and AD, and vice versa, dementia and AD are both potential risk indicators for LOE. Since there is substantial heterogeneity among the cohorts analyzed and more cohort studies should be conducted to confirm the correlations found in the current study.

Prevalence and unfavorable outcome of frailty in older adults with gastric cancer: a systematic review and meta-analysis.

BACKGROUND AND AIM: Previous studies reported inconsistent results on the prevalence and prognostic implications of frailty among older adults with gastric cancer. This systematic review synthesized available literature pertaining on this topic to establish the prevalence and unfavorable outcomes of frailty in older adults with gastric cancer. METHODS: A comprehensive search was conducted across multiple English databases including PubMed, Cochrane Library, CINAHL, Embase, and Web of Science as well as Chinese databases, namely, CNKI, Wan Fang, and CBM, from inception to July 4, 2023, to identify potential studies. Data related to the incidence of frailty and its unfavorable outcomes in older adults with gastric cancer were extracted. RevMan5.3 and R 4.2.2 were used to evaluate pooled prevalence, hazard ratios (HR), and 95% confidence interval (CI). RESULTS: This review comprehensively selected 13 studies, comprising 9 cohort studies and 4 cross-sectional studies, on 44,117 older adults diagnosed with gastric cancer. The incidence of frailty among older adults with gastric cancer ranged from 10 to 71%. The pooled prevalence of frailty was 29% (95% CI 0.21-0.39). Frailty was found to be associated with an elevated risk of postoperative complications (HR = 1.99, 95% CI 1.45-2.73), prolonged postoperative hospital stay (HR = 2.68, 95% CI 2.38-3.02), likelihood of readmission (HR = 3.28, 95% CI 1.77-6.08), and an increased mortality risk (HR = 1.60, 95% CI 1.36-1.90). CONCLUSIONS: Frailty was associated with a poor prognosis in older adults with gastric cancer. Clinical medical staff should focus on the frailty of older adults with gastric cancer, conduct large-scale, multicenter, and prospective studies and early screening of patients, and provide guidance for the implementation of prevention and treatment strategies.

Inhibition of histone acetyltransferase GCN5 by a transcription factor FgPacC controls fungal adaption to host-derived iron stress.

Poaceae plants can locally accumulate iron to suppress pathogen infection. It remains unknown how pathogens overcome host-derived iron stress during their successful infections. Here, we report that Fusarium graminearum (Fg), a destructive fungal pathogen of cereal crops, is challenged by host-derived high-iron stress. Fg infection induces host alkalinization, and the pH-dependent transcription factor FgPacC undergoes a proteolytic cleavage into the functional isoform named FgPacC30 under alkaline host environment. Subsequently FgPacC30 binds to a GCCAR(R = A/G)G element at the promoters of the genes involved in iron uptake and inhibits their expression, leading to adaption of Fg to high-iron stress. Mechanistically, FgPacC30 binds to FgGcn5 protein, a catalytic subunit of Spt-Ada-Gcn5 Acetyltransferase (SAGA) complex, leading to deregulation of histone acetylation at H3K18 and H2BK11, and repression of iron uptake genes. Moreover, we identified a protein kinase FgHal4, which is highly induced by extracellular high-iron stress and protects FgPacC30 against 26S proteasome-dependent degradation by promoting FgPacC30 phosphorylation at Ser2. Collectively, this study uncovers a novel inhibitory mechanism of the SAGA complex by a transcription factor that enables a fungal pathogen to adapt to dynamic microenvironments during infection.

G-quadruplex structural variations in human genome associated with single-nucleotide variations and their impact on gene activity.

G-quadruplexes (G4s) formed by guanine-rich nucleic acids play a role in essential biological processes such as transcription and replication. Besides the >1.5 million putative G-4-forming sequences (PQSs), the human genome features >640 million single-nucleotide variations (SNVs), the most common type of genetic variation among people or populations. An SNV may alter a G4 structure when it falls within a PQS motif. To date, genome-wide PQS-SNV interactions and their impact have not been investigated. Herein, we present a study on the PQS-SNV interactions and the impact they can bring to G4 structures and, subsequently, gene expressions. Based on build 154 of the Single Nucleotide Polymorphism Database (dbSNP), we identified 5 million gains/losses or structural conversions of G4s that can be caused by the SNVs. Of these G4 variations (G4Vs), 3.4 million are within genes, resulting in an average load of >120 G4Vs per gene, preferentially enriched near the transcription start site. Moreover, >80% of the G4Vs overlap with transcription factor-binding sites and >14% with enhancers, giving an average load of 3 and 7.5 for the two regulatory elements, respectively. Our experiments show that such G4Vs can significantly influence the expression of their host genes. These results reveal genome-wide G4Vs and their impact on gene activity, emphasizing an understanding of genetic variation, from a structural perspective, of their physiological function and pathological implications. The G4Vs may also provide a unique category of drug targets for individualized therapeutics, health risk assessment, and drug development.

Burst-Enhanced Super-Resolution Network (BESR).

Multi-frame super-resolution (MFSR) leverages complementary information between image sequences of the same scene to increase the resolution of the reconstructed image. As a branch of MFSR, burst super-resolution aims to restore image details by leveraging the complementary information between noisy sequences. In this paper, we propose an efficient burst-enhanced super-resolution network (BESR). Specifically, we introduce Geformer, a gate-enhanced transformer, and construct an enhanced CNN-Transformer block (ECTB) by combining convolutions to enhance local perception. ECTB efficiently aggregates intra-frame context and inter-frame correlation information, yielding an enhanced feature representation. Additionally, we leverage reference features to facilitate inter-frame communication, enhancing spatiotemporal coherence among multiple frames. To address the critical processes of inter-frame alignment and feature fusion, we propose optimized pyramid alignment (OPA) and hybrid feature fusion (HFF) modules to capture and utilize complementary information between multiple frames to recover more high-frequency details. Extensive experiments demonstrate that, compared to state-of-the-art methods, BESR achieves higher efficiency and competitively superior reconstruction results. On the synthetic dataset and real-world dataset of BurstSR, our BESR achieves PSNR values of 42.79 dB and 48.86 dB, respectively, outperforming other MFSR models significantly.

Simple Binding and Dissociation of a Sialoglycoprotein Using Boronic Acid-Modified Functional Interfaces on Microparticles.

An overexpression of sialic acid is an indicator of metastatic cancer, and selective detection of sialic acid shows potential for cancer diagnosis. Boronic acid is a promising candidate for this purpose because of its ability to specifically bind to sialic acid under acidic conditions. Notably, the binding strength can be easily modulated by adjusting the pH, which allows for a simple dissociation of the bound sialic acid. In this study, we developed 5-boronopicolinic acid (5-BPA)-modified magnetic particles (BMPs) to selectively capture sialic acid biomolecules. We successfully captured fetuin, a well-known sialoglycoprotein, on BMPs at >104 molecules/particle using an acetate buffer (pH 5.0). Facile dissociation then occurred when the system was changed to a pH 7.6 phosphate buffer. This capture-and-release process could be repeated at least five times. Moreover, this system could enrich fetuin by more than 20 times. In summary, BMPs are functional particles for facile purification and concentration through the selective capture of sialic acid proteins and can improve detection sensitivity compared with conventional methods. This technology shows potential for the detection of sialic acid overexpression by biological particles.

Short-term postmortem interval estimation by detection of apoptosis-related protein in skin.

Time-of-death extrapolation has always been one of the most important issues in forensic practice. For a complicated case in which a corpse is destroyed with little evidence, judging the time of death of the deceased is a major challenge, which also enables criminals to escape legal sanctions. To find a method to roughly judge the time of death of a corpse with only a small amount of skin tissue, in this study, we established an early death model by using mice; furthermore, the postmortem interval was estimated by determining the protein and mRNA levels of Bax and Bcl-2 in the skin. In this process, 0 h after death was used as the control group, and the expression levels of Bax and Caspase-3 reached the maximum value at 8-12 h, while Bcl-2, as an inhibitor of apoptosis protein, peaked after 24 h. The mRNA expression levels of related proteins in postmortem skin tissues were also different. The results of these data indicate that the protein and mRNA levels of Bax and Bcl-2 in the skin have potential application in early time-of-death estimation.

Confined Target-Triggered Hot Spots for In Situ SERS Analysis of Intranuclear Genotoxic Markers.

The γH2AX is a type of confined target in nuclei which is highly expressed around the damaged DNA during genotoxicity and has therefore been identified as a marker of genotoxicity. Convenient and intuitive in situ real-time detection of γH2AX is crucial for an accurate assessment of genotoxicity. Selective and nondestructive surface-enhanced Raman spectroscopy (SERS) is suitable to achieve this goal. However, the detection of substances in the nucleus by SERS is still limited due to the contradiction of probes between the nuclei entry efficiency and signal enhancement. This study utilized the characteristics of γH2AX as a confined target and constructed a γH2AX immunosensor based on gold nanoprobes with a small size (15 nm), which was modified with the TAT nuclear targeting peptide to ensure high nuclei entry efficiency. Once DNA damage was induced, the local overexpression of γH2AX further recruited the probe through immune recognition, so that hot spots could be assembled in situ to generate strong Raman signals, which were applied to evaluate the genotoxicity of drug impurities. This study proposed a novel SERS detection strategy, characterized by confined target-induced size conversion and hot spot formation, for in situ real-time analysis of intranuclear targets at the single-living-cell level, which intelligently simplified the structure of SERS probes and the operation process.

Evaluation of parametrial infiltration in patients with IB-IIB cervical cancer by a radiomics model integrating features from tumoral and peritumoral regions in 18 F-fluorodeoxy glucose positron emission tomography/MR images.

Parametrial infiltration (PMI) is an essential factor in staging and planning treatment of cervical cancer. The purpose of this study was to develop a radiomics model for accessing PMI in patients with IB-IIB cervical cancer using features from 18 F-fluorodeoxy glucose (18 F-FDG) positron emission tomography (PET)/MR images. In this retrospective study, 66 patients with International Federation of Gynecology and Obstetrics stage IB-IIB cervical cancer (22 with PMI and 44 without PMI) who underwent 18 F-FDG PET/MRI were divided into a training dataset (n = 46) and a testing dataset (n = 20). Features were extracted from both the tumoral and peritumoral regions in 18 F-FDG PET/MR images. Single-modality and multimodality radiomics models were developed with random forest to predict PMI. The performance of the models was evaluated with F1 score, accuracy, and area under the curve (AUC). The Kappa test was used to observe the differences between PMI evaluated by radiomics-based models and pathological results. The intraclass correlation coefficient for features extracted from each region of interest (ROI) was measured. Three-fold crossvalidation was conducted to confirm the diagnostic ability of the features. The radiomics models developed by features from the tumoral region in T2 -weighted images (F1 score = 0.400, accuracy = 0.700, AUC = 0.708, Kappa = 0.211, p = 0.329) and the peritumoral region in PET images (F1 score = 0.533, accuracy = 0.650, AUC = 0.714, Kappa = 0.271, p = 0.202) achieved the best performances in the testing dataset among the four single-ROI radiomics models. The combined model using features from the tumoral region in T2 -weighted images and the peritumoral region in PET images achieved the best performance (F1 score = 0.727, accuracy = 0.850, AUC = 0.774, Kappa = 0.625, p < 0.05). The results suggest that 18 F-FDG PET/MRI can provide complementary information regarding cervical cancer. The radiomics-based method integrating features from the tumoral and peritumoral regions in 18 F-FDG PET/MR images gave a superior performance for evaluating PMI.

Integrated bulk and single-cell RNA-sequencing reveals SPOCK2 as a novel biomarker gene in the development of congenital pulmonary airway malformation.

BACKGROUND: Congenital pulmonary airway malformation (CPAM) is the most frequent pulmonary developmental malformation and the pathophysiology remains poorly understood. This study aimed to identify the characteristic gene expression patterns and the marker genes essential to CPAM. METHODS: Tissues from the cystic area displaying CPAM and the area of normal appearance were obtained during surgery. Bulk RNA sequencing (RNA-seq) and single-cell RNA sequencing (scRNA-seq) were performed for integrating analysis. Iterative weighted gene correlation network analysis (iWGCNA) was used to identify specifically expressed genes to CPAM. RESULTS: In total, 2074 genes were significantly differentially expressed between the CPAM and control areas. Of these differentially expressed genes (DEGs), 1675 genes were up-regulated and 399 genes were down-regulated. Gene ontology analysis revealed these DEGs were specifically enriched in ciliated epithelium and involved in immune response. We also identified several CPAM-related modules by iWGCNA, among them, P15_I4_M3 module was the most influential module for distinguishing CPAMs from controls. By combining the analysis of the expression dataset from RNA-seq and scRNA-seq, SPOCK2, STX11, and ZNF331 were highlighted in CPAM. CONCLUSIONS: Through our analysis of expression datasets from both scRNA-seq and bulk RNA-seq of tissues obtained from patients with CPAM, we identified the characteristic gene expression patterns associated with the condition. Our findings suggest that SPOCK2 could be a potential biomarker gene for the diagnosis and therapeutic target in the development of CPAM, whereas STX11 and ZNF331 might serve as prognostic markers for this condition. Further investigations with larger samples and function studies are necessary to confirm the involvement of these genes in CPAM.

Investigation on the Corrosion Resistance Properties of Cobalt-Chromium-Molybdenum Alloy Artificial Human Body Components with Robust Biomimetic Superhydrophobic and Slippery Surfaces Based on Laser Texturing.

Cobalt-chromium-molybdenum alloy is used as a material for artificial human body components such as artificial hip joint and artificial denture and is often affected by electrochemical corrosion in human body fluids and saliva, which leads to inflammatory reactions and damage to the surrounding tissues as well as loosening and failure of the body components themselves. Few studies have been conducted to prepare corrosion-resistant coatings on the surface of Co28Cr6Mo. In this study, we used laser texturing to process a bionic 3D micronanocomposite structure on the surface of Co28Cr6Mo and quickly prepared a superhydrophobic and slippery surface coating with excellent corrosion resistance using polydimethylsiloxane solution and silicone oil modification. This surface had ultralow surface adhesion and good robustness of durability and abrasion resistance, reducing bacterial colonization or tissue adhesion and solving the problem of the lack of stability of the superhydrophobic surface. Microgrid grooves and layered nanoparticles were structurally responsible for the variation in wettability. The formation mechanism and composition of the prepared coatings were further analyzed. Electrochemical corrosion experiments were conducted on the surface in simulating body fluid and saliva environments, which showed the enhanced corrosion resistance of the prepared surface in the human body. These findings can further develop the surface functional modification of Co28Cr6Mo, accelerating basic and applied research studies on artificial human components.

Enhanced Curie temperature and conductivity of van der Waals ferromagnet MgV2S4via electrostatic doping.

A van der Waals intrinsic ferromagnet with double magnetic atom layers is of great interest for both revealing fundamental physics and exploring promising applications in low-dimensional spintronics. Here, the magnetic and electronic properties of the van der Waals ferromagnet MgV2S4 monolayer are studied under electrostatic doping using first-principles calculations. A MgV2S4 monolayer presents the desired physical properties such as that of being a half-semiconductor with a direct bandgap of 1.21 eV and a ferromagnetic ground state, and having a high Curie temperature of 462 K. Unlike the robust ferromagnetic ground state, magnetic anisotropy and Curie temperature are sensitive to electrostatic doping. Meanwhile, the transition from a semiconductor to a half-metal and the significant improvement in conductivity under electrostatic doping make the MgV2S4 monolayer a promising candidate for low-dimensional spintronic field-effect transistors.

Hydrogel capsule-based digital quantitative polymerase chain reaction.

Droplet digital PCR (ddPCR) is accurate in nucleic acid quantification owing to its linearity and high sensitivity. Amplification of nucleic acid in droplets, however, is limited by the stability of droplets against thermal cycling. While the use of fluorinated oil or supplementation of surfactant could improve the stability of droplets, this process has also greatly increased the cost of ddPCR and limited post-PCR analysis. Here, we report a novel method known as gel capsule-based digital PCR (gc-dPCR) which includes a method to prepare hydrogel capsules encapsulating the PCR reaction mix, conducting PCR reaction, and readout by either quantitative PCR (qPCR) system or fluorescence microplate reader. We have compared the developed method to vortex ddPCR. Our approach results in higher fluorescence intensity compared to ddPCR suggesting higher sensitivity of the system. As hydrogel capsules are more stable than droplets in fluorinated oil throughout thermal cycling, all partitions can be quantified, thus preventing loss of information from low-concentration samples. The new approach should extend to all droplet-based PCR methods. It has greatly improved ddPCR by increasing droplets stability and sensitivity, and reducing the cost of ddPCR, which help to remove the barrier of ddPCR in settings with limited resources.

Improvement of Sensitivity and Speed of Virus Sensing Technologies Using nm- and µm-Scale Components.

Various viral diseases can be widespread and cause severe disruption to global society. Highly sensitive virus detection methods are needed to take effective measures to prevent the spread of viral infection. This required the development of rapid virus detection technology to detect viruses at low concentrations, even in the biological fluid of patients in the early stages of the disease or environmental samples. This review describes an overview of various virus detection technologies and then refers to typical technologies such as beads-based assay, digital assay, and pore-based sensing, which are the three modern approaches to improve the performance of viral sensing in terms of speed and sensitivity.

A Hardware-Friendly High-Precision CNN Pruning Method and Its FPGA Implementation.

To address the problems of large storage requirements, computational pressure, untimely data supply of off-chip memory, and low computational efficiency during hardware deployment due to the large number of convolutional neural network (CNN) parameters, we developed an innovative hardware-friendly CNN pruning method called KRP, which prunes the convolutional kernel on a row scale. A new retraining method based on LR tracking was used to obtain a CNN model with both a high pruning rate and accuracy. Furthermore, we designed a high-performance convolutional computation module on the FPGA platform to help deploy KRP pruning models. The results of comparative experiments on CNNs such as VGG and ResNet showed that KRP has higher accuracy than most pruning methods. At the same time, the KRP method, together with the GSNQ quantization method developed in our previous study, forms a high-precision hardware-friendly network compression framework that can achieve "lossless" CNN compression with a 27× reduction in network model storage. The results of the comparative experiments on the FPGA showed that the KRP pruning method not only requires much less storage space, but also helps to reduce the on-chip hardware resource consumption by more than half and effectively improves the parallelism of the model in FPGAs with a strong hardware-friendly feature. This study provides more ideas for the application of CNNs in the field of edge computing.

Adaptive Global Power-of-Two Ternary Quantization Algorithm Based on Unfixed Boundary Thresholds.

In the field of edge computing, quantizing convolutional neural networks (CNNs) using extremely low bit widths can significantly alleviate the associated storage and computational burdens in embedded hardware, thereby improving computational efficiency. However, such quantization also presents a challenge related to substantial decreases in detection accuracy. This paper proposes an innovative method, called Adaptive Global Power-of-Two Ternary Quantization Based on Unfixed Boundary Thresholds (APTQ). APTQ achieves adaptive quantization by quantizing each filter into two binary subfilters represented as power-of-two values, thereby addressing the accuracy degradation caused by a lack of expression ability of low-bit-width weight values and the contradiction between fixed quantization boundaries and the uneven actual weight distribution. It effectively reduces the accuracy loss while at the same time presenting strong hardware-friendly characteristics because of the power-of-two quantization. This paper extends the APTQ algorithm to propose the APQ quantization algorithm, which can adapt to arbitrary quantization bit widths. Furthermore, this paper designs dedicated edge deployment convolutional computation modules for the obtained quantized models. Through quantization comparison experiments with multiple commonly used CNN models utilized on the CIFAR10, CIFAR100, and Mini-ImageNet data sets, it is verified that the APTQ and APQ algorithms possess better accuracy performance than most state-of-the-art quantization algorithms and can achieve results with very low accuracy loss in certain CNNs (e.g., the accuracy loss of the APTQ ternary ResNet-56 model on CIFAR10 is 0.13%). The dedicated convolutional computation modules enable the corresponding quantized models to occupy fewer on-chip hardware resources in edge chips, thereby effectively improving computational efficiency. This adaptive CNN quantization method, combined with the power-of-two quantization results, strikes a balance between the quantization accuracy performance and deployment efficiency in embedded hardware. As such, valuable insights for the industrial edge computing domain can be gained.