Performance Ceiling Research Articles

Low‐Cost Surrogate Modeling for Expedited Data Acquisition of Reconfigurable Metasurfaces

AbstractIn recent years, machine learning (ML) and deep learning (DL) have been widely used to break the metasurface’s performance ceiling. However, the existing data‐driven ML and DL methods usually require the availability of vast amounts of training data to ensure their stable and accurate performance. The process of acquiring these data is high‐cost due to the need for numerous full‐wave electromagnetic (EM) simulations. Here, we propose a low‐cost surrogate model to generate these data efficiently. The proposed model employs microwave network theory to separate meta‐elements into four independent components. Through integration with transmission line theory, we derive the EM responses of meta‐elements using analytical representation with the active device equivalent impedance and dielectric as design variables. Two typical phase‐modulation active meta‐elements are employed to verify the accuracy of our macromodel in comparison with full‐wave EM simulations. Based on the developed macromodel, the superior prediction ability is further presented to illustrate the performance of meta‐elements with various active devices and dielectric substrates. The proposed macromodel is a feasible and general method to rapidly obtain the necessary training data of active meta‐elements, which holds a great potential to significantly reduce the designing time of ML and DL models for the active metasurfaces.

Fused Low-Earth-Orbit GNSS

Traditional Global Navigation Satellite System (GNSS) immunity to interference may be approaching a practical performance ceiling. Greater gains are possible outside traditional GNSS orbits and spectrum. GNSS from low Earth orbit (LEO) has long been viewed as promising but expensive, requiring large constellations for rapid navigation solutions. The recent emergence of commercial broadband LEO mega-constellations invites study on dual-purposing these for both communications-their primary mission-and a secondary positioning, navigation, and timing (PNT) service. Operating at shorter wavelengths than traditional GNSS, these constellations would permit highly directive, relatively compact receiver antennas. PNT-specific on-orbit resources would not be required: the transmitters, antennas, clocks, and spectrum of the hosting broadband network would suffice for PNT. Non-cooperative use of LEO signals for PNT is an option, but cooperation with the constellation operator (“fusion” with its communications mission) eases the burden of tracking a dense, low-altitude constellation from the ground and enables a receiver to produce single-epoch stand-alone PNT solutions. This paper proposes such a cooperative concept, termed fused LEO GNSS. Viability hinges on opportunity cost, or the burden a secondary PNT mission imposes on the communications constellation operator. This is assessed in terms of time-space-bandwidth product and energy budget. It is shown that a near-instantaneous-fix PNT service over ±60° latitude (covering 99.8% of the world's population) with positioning performance superior to traditional GNSS pseudoranging would cost less than 1.6% of downlink capacity for the largest of the new constellations, SpaceX's Starlink. This allocation is comparable to adding one user consuming 5.7 Mbps of broadband service to each cell.

Optimizing image captioning: The effectiveness of vision transformers and VGG networks for remote sensing

This study presents a comprehensive evaluation of two prominent deep learning models, Vision Transformer (ViT) and VGG16, within the domain of image captioning for remote sensing data. By leveraging the BLEU score, a widely accepted metric for assessing the quality of text generated by machine learning models against a set of reference captions, this research aims to dissect and understand the capabilities and performance nuances of these models across various sample sizes: 25, 50, 75, and 100 samples. Our findings reveal that the Vision Transformer model generally outperforms the VGG16 model across all evaluated sample sizes, achieving its peak performance at 50 samples with a BLEU score of 0.5507. This performance shows that ViT benefits from its ability to capture global dependencies within the data, providing a more nuanced understanding of the images. However, the performance slightly decreases as the sample size increases beyond 50, indicating potential challenges in scalability or overfitting to the training data. Conversely, the VGG16 model shows a different performance trajectory, starting with a lower BLEU score for smaller sample sizes but demonstrating a consistent improvement as the sample size increases, culminating in its highest BLEU score of 0.4783 for 100 samples. This pattern suggests that VGG16 may require a larger dataset to adequately learn and generalize from the data, although it achieves a more modest performance ceiling compared to ViT. Through a detailed analysis of these findings, the study underscores the strengths and limitations of each model in the context of image captioning. The Vision Transformer's superior performance highlights its potential for applications requiring high accuracy in text generation from images. In contrast, the gradual improvement exhibited by VGG16 suggests its utility in scenarios where large datasets are available, and scalability is a priority. This study contributes to the ongoing discourse in the AI community regarding the selection and optimization of deep learning models for complex tasks such as image captioning, offering insights that could guide future research and application development in this field.

FedA3I: Annotation Quality-Aware Aggregation for Federated Medical Image Segmentation against Heterogeneous Annotation Noise

Federated learning (FL) has emerged as a promising paradigm for training segmentation models on decentralized medical data, owing to its privacy-preserving property. However, existing research overlooks the prevalent annotation noise encountered in real-world medical datasets, which limits the performance ceilings of FL. In this paper, we, for the first time, identify and tackle this problem. For problem formulation, we propose a contour evolution for modeling non-independent and identically distributed (Non-IID) noise across pixels within each client and then extend it to the case of multi-source data to form a heterogeneous noise model (i.e., Non-IID annotation noise across clients). For robust learning from annotations with such two-level Non-IID noise, we emphasize the importance of data quality in model aggregation, allowing high-quality clients to have a greater impact on FL. To achieve this, we propose Federated learning with Annotation quAlity-aware AggregatIon, named FedA3I, by introducing a quality factor based on client-wise noise estimation. Specifically, noise estimation at each client is accomplished through the Gaussian mixture model and then incorporated into model aggregation in a layer-wise manner to up-weight high-quality clients. Extensive experiments on two real-world medical image segmentation datasets demonstrate the superior performance of FedA3I against the state-of-the-art approaches in dealing with cross-client annotation noise. The code is available at https://github.com/wnn2000/FedAAAI.

Preresidency research output among US neurological surgery residents.

Research productivity is often used to evaluate candidates for neurosurgery residency. Official annual reports describe the mean total number of research products of successful applicants for each match cycle; however, the average number of indexed publications, the highest-valued research product, is not reported separately from other research products. The primary objectives of this study were to describe the distribution of preresidency indexed publication quantity among successful neurosurgery applicants from 2017 to 2021 and determine whether any change in publication quantity across application cycles existed. Secondary objectives included determining the rate at which the average publication quantity is increasing across application cycles, whether this increase is driven by high-output applicants alone, and if a performance ceiling has been reached. US doctor of medicine seniors applying to the senior author's institution between 2017 and 2021 and who successfully matched into any US neurosurgery program were included. Publication quantities were extracted using Scopus. Additional variables were extracted from residency applications. Mean (SD) and median (IQR) publication quantities were used to describe the distribution and compare across years. Applicants were ranked by descending publication count and divided into quartiles. Averages within each quartile were compared with respective quartiles across years. Averages of the top 10% most productive applicants were compared across years to determine if a performance ceiling existed. Overall, 93.2% of matched applicants were captured. The mean and median total numbers of publications for applicants who matched from 2017 to 2021 were 5.6 ± 8.3 and 3.0 (1.0, 7.0), respectively. The mean and median numbers of publications increased from 3.7 ± 5.3 and 2.0 (0.0, 5.0) in 2016-2017 to 8.1 ± 10.0 and 5.0 (2.0, 11.0) in 2020-2021 (p < 0.001). The distribution of publication quantity was right-skewed. Multivariable analysis determined the application year to be independently and positively correlated with publication quantity (β 1.07 [95% CI 0.71-1.42], p < 0.001). All quartiles observed an increased average number of publications across years (p < 0.001). The mean and median numbers for the top 10% increased from 15.8 ± 8.7 and 13.0 (10.8, 15.5) in 2016-2017, respectively, to 31.3 ± 16.0 and 25.0 (21.0, 35.5) in 2020-2021 (p < 0.001). Indexed publications account for a small portion of the total research products that successful neurosurgery candidates list on applications. A high number of publications is not necessary for candidates to match, with approximately 50% of all applicants who successfully matched having ≤ 5 publications and 25% having ≤ 2 publications. The average preresidency publication quantity has been increasing yearly among neurosurgery applicants. This increase was present across the applicant pool. Additionally, no performance ceiling was observed.

Block-Wise Partner Learning for Model Compression.

Despite the great potential of convolutional neural networks (CNNs) in various tasks, the resource-hungry nature greatly hinders their wide deployment in cost-sensitive and low-powered scenarios, especially applications in remote sensing. Existing model pruning approaches, implemented by a "subtraction" operation, impose a performance ceiling on the slimmed model. Self-knowledge distillation (Self-KD) resorts to auxiliary networks that are only active in the training phase for performance improvement. However, the knowledge is holistic and crude, and the learning-based knowledge transfer is mediate and lossy. Here, we propose a novel model-compression method, termed block-wise partner learning (BPL), which comprises "extension" and "fusion" operations and liberates the compressed model from the bondage of baseline. Different from the Self-KD, the proposed BPL creates a partner for each block for performance enhancement in training. For the model to absorb more diverse information, a diversity loss (DL) is designed to evaluate the difference between the original block and the partner. Besides, the partner is fused equivalently instead of being discarded directly. After training, we can simply adopt the fused compressed model that contains the enhancement information of partners but with fewer parameters and less inference cost. As validated using the UC Merced land-use, NWPU-RESISC45, and RSD46-WHU datasets, the BPL demonstrates superiority over other compared model-compression approaches. For example, it attains a substantial floating-point operations (FLOPs) reduction of 73.97% with only 0.24 accuracy (ACC.) loss for ResNet-50 on the UC Merced land-use dataset. The code is available at https://github.com/zhangxin-xd/BPL.

Breaking the performance ceiling for neoantigen immunogenicity prediction.

Breaking the performance ceiling for neoantigen immunogenicity prediction.

SegCoFusion: An Integrative Multimodal Volumetric Segmentation Cooperating With Fusion Pipeline to Enhance Lesion Awareness.

Multimodal volumetric segmentation and fusion are two valuable techniques for surgical treatment planning, image-guided interventions, tumor growth detection, radiotherapy map generation, etc. In recent years, deep learning has demonstrated its excellent capability in both of the above tasks, while these methods inevitably face bottlenecks. On the one hand, recent segmentation studies, especially the U-Net-style series, have reached the performance ceiling in segmentation tasks. On the other hand, it is almost impossible to capture the ground truth of the fusion in multimodal imaging, due to differences in physical principles among imaging modalities. Hence, most of the existing studies in the field of multimodal medical image fusion, which fuse only two modalities at a time with hand-crafted proportions, are subjective and task-specific. To address the above concerns, this work proposes an integration of multimodal segmentation and fusion, namely SegCoFusion, which consists of a novel feature frequency dividing network named FDNet and a segmentation part using a dual-single path feature supplementing strategy to optimize the segmentation inputs and suture with the fusion part. Furthermore, focusing on multimodal brain tumor volumetric fusion and segmentation, the qualitative and quantitative results demonstrate that SegCoFusion can break the ceiling both of segmentation and fusion methods. Moreover, the effectiveness of the proposed framework is also revealed by comparing it with state-of-the-art fusion methods on 2D two-modality fusion tasks, our method achieves better fusion performance than others. Therefore, the proposed SegCoFusion develops a novel perspective that improves the performance in volumetric fusion by cooperating with segmentation and enhances lesion awareness.

Is immunology doing well? A look at 100 immune-mediated inflammatory diseases for 100 years of the Journal.

It is now 60 years since Ian Mackay and Macfarlane Burnet published their seminal text "The Autoimmune Diseases" in which they examined the full scope of human inflammatory pathology as a manifestation of the underlying structure and function of the immune system. Here I revisit this approach to ask to what extent has the promise of Mackay and Burnet's work been exploited in clinical medicine as currently practiced. In other words, is immunology doing well? Despite spectacular headline contributions of immunology in clinical medicine, I present evidence suggesting a performance ceiling in our capacity to answer the relatively straightforward questions that patients frequently ask about their own diseases and find that this ceiling exists across almost all of the 100 immune-mediated inflammatory diseases examined. I propose that these questions are difficult, not so much because the immune system is overwhelmingly complex but rather that we have more to learn about the relatively simple agents and rules that may underpin self-organizing complex interacting systems as revealed in studies from other disciplines. The way that the immune system has evolved to exploit the ancient machinery determining three independent cell fate timers as described in this Journal would be a great place to start to decode the self-organizing principles that underpin the emergent pathology that we observe in the clinic.

CsSn0.5Ge0.5I3‐Based Lead‐Free Highly Stable Perovskite Solar Cell: Numerical Modeling for Estimating Performance Ceiling

CsSn0.5Ge0.5I3 perovskite is reportedly highly stable in ambient open air, lead free, and has excellent optoelectrical properties. An inverted p–i–n solar cell device based on this mixed SnGe perovskite utilizing the reported optical and electrical characteristics of the CsSn0.5Ge0.5I3 is simulated. This theoretical device under various recombination regimes to explore the performance ceiling of CsSn0.5Ge0.5I3 is put. An optimized configuration of CsSn0.5Ge0.5I3‐based perovskite solar cell shows an efficiency of 29% under the impact of only intrinsic recombination losses such as radiative (with radiative recombination coefficient of 10−11) and Auger recombination (recombination coefficient of 10−27). When extrinsic factors are considered, such as resistance losses (series resistance as high as 2 Ω cm2 and shunt resistance as low as 1000 Ω cm2), efficiency decreases to 27.5%. The efficiency is 20% when trap‐assisted Shockley–Read–Hall recombination is considered with voltage loss (V Loss) of 0.5 V. Similarly, V Loss = 0.6 V in V OC restricts device efficiency to 15%. Finally, an efficiency waterfall chart summarizes the CsSn0.5Ge0.5I3, efficiency under different extrinsic losses, and the performance loss analysis, providing an optimal design. The results summarized here are expected to be helpful and prompt experimentalists to fabricate this stable lead‐free perovskite solar cell.

Performance optimization design of direct injection turbocharged hydrogen internal combustion engine

The best energy source for the twenty-first century has been praised as hydrogen energy. The hydrogen internal combustion engine (H2ICE) has captured the interest of several researchers as the primary use of hydrogen energy. The performance computation model of the direct-injection (DI) turbocharged H2ICE was developed using a mix of one-dimensional (1D) performance modeling tools and the H2ICE bench test. The testing results were used to verify and examine the model. The performance ceiling was established after taking backfire and knock into account for H2ICE. The performance boundary was investigated through simulated supercharging, and the optimized turbocharger was selected. Then, in order to achieve the best overall performance of the H2ICE, optimization of the valve phase, hydrogen injection duration angle, compression ratio, and turbocharging matching were performed. Finally, a peak power of 133 kW @ 4500 rpm and a maximum torque of 339 N·m @ 2000 rpm could be reached, and the highest effective thermal efficiency 43.06% was achieved. The performance development tests of various kinds of H2ICEs could be guided by the design methodology.

Adaptive energy harvesting from low-level ambient vibrations

Harvesting energy from ambient mechanical vibrations is an enabling technology with profound societal impacts. However, existing methods suffer from two major technical deficiencies: the weak coupling between the harvester and the vibrating source and the low energy transduction efficiency. Here we show an adaptive approach can overcome these challenges, leading to improvements by orders of magnitude. Using a piezoelectric device, we demonstrate that a conservative actuation force can block energy from flowing back to the vibrating source, thereby increasing the coupling strength and increasing the transduction efficiency significantly. Using low actuation forces, the adaptive approach can extract almost two orders of magnitude more power than the traditional approach from a multi-frequency vibration. The improvement can increase to three orders of magnitude when moderate actuation force is used. If materials that can withstand a high actuation voltage are available, the adaptive approach has the potential to surpass the theoretical performance ceiling of the traditional approach, improving the upper-bound performance by over 100 times and creating a significant opportunity for the technology to advance.

Improving power and performance of on-chip network through virtual channel sharing and power gating

Networks-on-Chip (NoCs) are becoming the communication backbone in multicore chips. The router buffer holds a critical significance for the communication performance of NoC. However, the bursty traffic means that the buffers are idle for many periods and generate a significant amount of static power. Moreover, due to the uneven traffic distribution, some buffers within the router stand idle even at network saturation. Both the performance and power budget of NoC heavily depends on the router’s buffer budget. Therefore, we desperately demand a mechanism to maximize the usage of idle buffers to balance the load and improve the performance ceiling of NoC. Dynamically based on network traffic, early switching on or off of buffers to save power consumption. We propose a modified microarchitecture of the virtual channel sharing router, MVCR. The MVCR innovatively applies aggressive fine-grained power gating mechanisms to the idle buffers of each directional input port (east, west, south, and north) to achieve static power savings. Moreover, the MVCR further extends the sleeping VC period through an efficient packet-transmission adjustment mechanism, thus achieving maximum static power savings and improved network performance. Synthetic applications and real benchmarks have been used to validate MCVR. Compared to traditional router scheme, MVCR increases throughput by 30%, reduces power consumption by 47%, and has a latency overhead of about 6%. The key performance metrics are better than other solutions, and the added area cost is moderate.

Large-scale neural biomedical entity linking with layer overwriting

Motivation: Entity linking is the task of linking entity mentions to the database entries corresponding to the entity mentions. Entity linking enables the treatment of superficially different but semantically identical mentions as the same entity. Since millions of concepts are listed in biomedical databases, selecting the correct database entry for each targeted entity is challenging. Simple string matching between the word and each synonym in biomedical databases is insufficient to handle a wide variety of variants of biomedical entities appearing in the biomedical literature. Recent progress in neural approaches is promising for entity linking. Still, existing neural methods require sufficient data, which is difficult to prepare in biomedical entity linking that deals with millions of biomedical concepts. Therefore, we need to develop a new neural method to train entity-linking models over the sparse training data covering a very limited part of the biomedical concepts. Results:We have devised a pure neural model that classifies biomedical entity mentions into millions of biomedical concepts. The classifier employs (1) the layer overwriting that breaks through the performance ceiling during training, (2) training data augmentation using database entries that compensate for the problem of insufficient training data, and (3) the cosine similarity-based loss function that helps distinguish the millions of biomedical concepts. Our system using the proposed classifier was ranked first in the official run of the National NLP Clinical Challenges (n2c2) 2019 Track 3, which targeted linking medical/clinical entity mentions to 434,056 Concept Unique Identifier (CUI) entries. We also applied our system to the MedMentions dataset, which has 3.2M candidate concepts. Experimental results confirmed the same advantages of our proposed method. We further evaluated our system on the NLM-CHEM corpus with 350K candidate concepts, and our system achieved a new state-of-the-art performance on the corpus.Availability:https://github.com/tti-coin/bio-linkingContact:makoto.miwa@toyota-ti.ac.jp

Extracellular volume is an independent predictor of peak cardiac power in asymptomatic patients with moderate-to-severe aortic stenosis

Abstract Funding Acknowledgements Type of funding sources: Other. Main funding source(s): The PRIMID study was funded by an NIHR fellowship (Post-Doctoral Fellowship supported by the National Institute for Health Research (NIHR-PDF 2011–04–51 Gerald P McCann.) Background Cardiac power (CP) describes the cardiac work done to maintain arterial pressure in the vasculature and peak CP is the ceiling of pumping performance the heart can achieve. Mean pressure gradient adjusted CP (MPG-CP) gives a more accurate representation of afterload in patients with aortic stenosis (AS) but its association with left-ventricular remodelling parameters is unknown. Purpose To establish the determinants of peak MPG-CP in asymptomatic patients with AS. Methods In this prospective, multi-centre observational study, asymptomatic patients with moderate to severe AS underwent echocardiography, cardiovascular magnetic resonance (CMR) imaging and cardiopulmonary exercise testing. Peak MPG-CP was calculated at peak exercise during an incremental symptom-limited exercise test on semi-supine ergometer, using the formula shown (Figure-1). Clinical and imaging univariable and multivariable associations of peak MPG-CP were assessed by linear regression. Results Peak MPG-CP was measured in 86 patients (mean age 66±12 years, 77% male, mean aortic valve area (AVA) 1.08±0.33 cm^2). On univariable analysis, peak MPG-CP was directly correlated with echocardiogram-measured AVA, peak pressure gradient, max velocity, and CMR-measured myocardial perfusion reserve (MPR), LV mass index (LVMI) and LV end-diastolic volume index (LVEDVI), and inversely correlated with extracellular volume (ECV) and NT-proBNP (Table-1). ECV was the only variable independently associated with peak MPG-CP regardless of which measure of AS severity was inputted in the multivariable model. Conclusions ECV, a marker of diffuse interstitial fibrosis, is an independent predictor of peak MPG-CP in asymptomatic patients with moderate-severe AS. Given ECV's role as a prognosticator, CP parameters may be important markers of subclinical disease and serve as surrogates for fibrotic burden and risk-stratification tools in asymptomatic AS.

Analysing the performance ceiling of RbSnGeI3-based lead-free stable perovskite solar cell

Analysing the performance ceiling of RbSnGeI3-based lead-free stable perovskite solar cell

Full-Range Amplitude–Phase Metacells for Sidelobe Suppression of Metalens Antenna Using Prior-Knowledge-Guided Deep- Learning -Enabled Synthesis

A prior-knowledge-guided deep-learning-enabled (PK-DL) synthesis method is proposed to design the metacells with the full-range amplitude and phase control for suppressing the sidelobe levels of a metalens antenna. The PK-DL synthesis method is based on conditional deep convolutional generative adversarial network (cDCGAN). First, the metacells are pixelated to offer the higher degrees of design freedom and expand their transmission responses to full two-dimensional coverage of amplitude and phase responses so that a full-range metacell dataset is built up to simultaneously control the amplitude and phase of a single metacell. Conventionally, by optimizing the dimensions of predefined patterns of a metacell, only phase response of a single metacell is achieved with a specific amplitude response, vice versa. The metalens design based on the synthesized metacells has much higher degrees of freedom so that the metalens antenna has the potential to break the ceiling of performance and to realize more functions than existing metalens designs. As examples, two types of single-lens metalens antennas are designed for sidelobe suppression using the proposed metacell dataset. The first metalens is excited by a dipole antenna with an omnidirectional radiation pattern to demonstrate the simultaneous control of amplitude and phase responses using the synthesized metacells. The second metalens fed by a standard horn antenna is designed to suppress the sidelobe levels in both the E- and H-planes. The horn-fed square metalens has an aperture of 13.2 λ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> × 13.2 λ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</sub> at 12 GHz. Measurement shows that the proposed metalens antenna realizes the gain of 26.9 dBi at 12 GHz and the first sidelobe levels in the E- and H-planes are -33.2 dB and -30.9 dB, respectively. The proposed synthesis method greatly expands the boundary of amplitude and phase responses of a single metacell so that the metacell dataset offers much higher degrees of freedom for not only metalenses but also other metasurfaces.

Sequential self-propelled morphology transitions of nanoscale condensates enable a cascade jumping-droplet condensation

Jumping-droplet condensation, namely the out-of-plane jumping of condensed droplets upon coalescence, has been a promising technical innovation in the fields of energy harvesting, droplet manipulation, thermal management, etc., yet is limited owing to the challenge of enabling a sustainable and programmable control. Here, we characterized the morphological evolutions and dynamic behaviors of nanoscale condensates on different nanopillar surfaces, and found that there exists an unrevealed domino effect throughout the entire droplet lifecycle and the coalescence is not the only mechanism to access the droplet jumping. The vapor nucleation preferentially occurs in structure intervals, thus the formed liquid embryos incubate and grow in a spatially confined mode, which stores an excess surface energy and simultaneously provides a asymmetric Laplace pressure, stimulating the trapped droplets to undergo a dewetting transition or even a self-jumping, which can be facilitated by the tall and dense nanostructures. Subsequently, the adjacent droplets merge mutually and further trigger more multifarious self-propelled behaviors that are affected by underlying surface nanostructure, including dewetting transition, coalescence-induced jumping and jumping relay. Moreover, an improved energy-based model was developed by considering both the surface structure properties and the nano-physical effects, the theoretical prediction not only extends the coalescence-induced jumping to the nanometer-sized droplets but also correlates the surface nanostructure topology to the jumping velocity. Such a domino effect of nucleation-growth-coalescence on the ultimate morphology of droplet may offer a new strategy for designing functional nanostructured surfaces that serve to orientationally manipulate, transport and collect droplets, and motivate surface engineers to realize a stable cascade jumping-droplet condensation and achieve its performance ceiling.

Hierarchical Associative Encoding and Decoding for Bottom-Up Human Pose Estimation

Bottom-up human pose estimation decouples computational complexity from the number of people but requires additional operations to match the detected keypoints to each human instance. Existing approaches treat all keypoints equally while ignoring the relationships among keypoints, which in turn limit the performance ceilings. In this work, we propose a hierarchical associative encoding and decoding framework for bottom-up human pose estimation by introducing additional prior knowledge. Specifically, in addition to keypoint-level and instance-level associations, we further divide keypoints into groups and explore group-level associations. This way, prior knowledge is incorporated to determine the keypoint groups for better associative encoding. To deal with complex poses, we introduce a focal pulling loss to focus more on the hard-to-associate keypoints. Moreover, instead of using a pre-defined order for keypoint grouping, we propose a progressive associative decoding method to dynamically determine the order of keypoints for grouping, which helps reduce isolated keypoints. Experimental results on the MS-COCO, CrowdPose and MPII datasets show superior performance of our proposed associative encoding and decoding algorithms. More importantly, we prove, through validation, that hierarchical associative encoding and decoding can be used as a plug-n-play module for performance improvement regardless of backbone architecture. Our source code and pretrained models are available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/ducongju/HAE</uri> .

Quasi-static magnetic compression of field-reversed configuration plasma: amended scalings and limits from two-dimensional MHD equilibrium

In this work, several key scaling laws of the quasi-static magnetic compression of field reversed configuration (FRC) plasma (Spencer et al 1983 Phys. Fluids 26 1564) are amended from a series of two-dimensional FRC MHD equilibriums numerically obtained using the Grad–Shafranov equation solver NIMEQ. Based on the new scaling for the elongation and the magnetic fields at the separatrix and the wall, the empirically stable limits for the compression ratio, the fusion gain, and the neutron yield are evaluated, which may serve as a more accurate estimate for the upper ceiling of performance from the magnetic compression of FRC plasma as a potential fusion energy as well as neutron source devices.