Training Difficulty Research Articles

RF sensing enabled tracking of human facial expressions using machine learning algorithms.

Automatic analysis of facial expressions has emerged as a prominent research area in the past decade. Facial expressions serve as crucial indicators for understanding human behavior, enabling the identification and assessment of positive and negative emotions. Moreover, facial expressions provide insights into various aspects of mental activities, social connections, and physiological information. Currently, most facial expression detection systems rely on cameras and wearable devices. However, these methods have drawbacks, including privacy concerns, issues with poor lighting and line of sight blockage, difficulties in training with longer video sequences, computational complexities, and disruptions to daily routines. To address these challenges, this study proposes a novel and privacy-preserving human behavior recognition system that utilizes Frequency Modulated Continuous Wave (FMCW) radar combined with Machine Learning (ML) techniques for classifying facial expressions. Specifically, the study focuses on five common facial expressions: Happy, Sad, Fear, Surprise, and Neutral. The recorded data is obtained in the form of a Micro-Doppler signal, and state-of-the-art ML models such as Super Learner, Linear Discriminant Analysis, Random Forest, K-Nearest Neighbor, Long Short-Term Memory, and Logistic Regression are employed to extract relevant features. These extracted features from the radar data are then fed into ML models for classification. The results show a highly promising classification accuracy of 91%. The future applications of the proposed work will lead to advancements in technology, healthcare, security, and communication, thereby improving overall human well-being and societal functioning.

Perceived discrepancies in neurosonology training and certification across Europe: a RRFS/EAN survey.

Neurosonology is a vital paraclinical investigation in modern neurology. However, access to education and certification in neurosonology for neurology residents and young specialists in Europe is challenging, and comprehensive data regarding this topic are scarce. Information regarding difficulties in neurosonology training across Europe may help bring this topic under the spotlight and act as a call for the harmonization of curricula across the continent. We performed an online survey targeting European neurology residents and young specialists, focusing on neurosonology training and certification. The survey was conducted between May and September 2023 and received responses from 282 participants representing 37 European countries. There were disparities in neurosonology training during residency, with 6 (16.2%) out of 37 countries reporting a dedicated curriculum. The respondents expressed an overall lack of satisfaction with theoretical knowledge (rating their experience as very poor 28.0%, poor 20.2%, neutral 25.9%, good 19.3%, and very good 6.6%) and practical skills gained during their training (rating their experience as very poor 30.9%, poor 18.9%, neutral 22.6%, good 18.1%, and very good 9.5%). A total of 282 respondents (5.7%), 16 held a national certification in neurosonology, claiming obstacles such as high costs of certification and a limited number of certifying centers. This survey reveals significant variations in neurosonology training across Europe, indicating difficulties in obtaining certification. Despite the increasing importance of neurosonology, many neurologists feel inadequately prepared and lack practical training during residency, emphasizing the need for better and more standardized access. The survey underscores challenges and disparities in neurosonology training and certification in Europe. Standardization of curricula and increased awareness about available certifications are crucial to address these issues. The interest in European Certification suggests a potential solution for enhancing neurosonology training at the international level.

Application Research of Power Measurement Data Exchange Based on Uninterruptible Power Supply

With the maturity of the smart meter embedded system integration mode, the power consumption information can be fed back to the management center by the intelligent terminal through the power Internet of Things. The smart terminal adopts the narrowband data communication mode. However, the narrowband channel interval is hindered by interference signals, resulting in serious data packet loss. Therefore, a stable and efficient acquisition scheme is needed to ensure its smooth interaction. Based on the application of smart meter data communication, we analyze the power consumption optimization and carrier transmission optimization of the power Internet of Things and propose a data analysis method based on self-learning and external characteristics, which is used to calculate the offline execution strategy of smart meters when they encounter factors such as power outage or intrusion. A production task-driving rule is designed. The external characteristics of deep learning data of the power Internet of Things are collected. The accuracy of data classification is judged according to the parallel Internet of Things of a convolutional neural network and a long-term and short-term memory deep neural network. When dealing with high-dimensional original data, the data network is graded to reduce the work of feature selection and the difficulty of training. Moreover, we improve the autonomous execution ability of an embedded terminal device in an offline state. We also strengthen the ability of abnormal data detection and judgment learning to meet the requirements of efficient and stable operation of the power grid. Simulation results verify that compared with the current mainstream schemes, the performance of the proposed scheme is improved by more than 17%. Besides, the fault tolerance rate is enhanced by 29%, and the power consumption after optimization is reduced by 26.3% and 42.8%. These effects highlight the role of data processing in embedded systems. It solves the problem of abnormal data and energy loss affecting the accuracy of error estimation of electric energy meters, opens up a new way for data acquisition and processing of large-scale smart meters, and further improves the service level of information management of power systems.

Styled and characteristic Peking opera facial makeup synthesis with co-training and transfer conditional styleGAN2

Against the backdrop of the deep integration of culture and technology, research and practice in digitization of intangible cultural heritage has continued to deepen. However, due to the lack of data and training, it is still very difficult to apply artificial intelligence to the field of cultural heritage protection. This article integrates image generation technology into the digital protection of Peking opera facial makeup, using a self-built Peking opera facial makeup dataset. Based on the StyleGAN2 network, we propose a style generative cooperative training network Co-StyleGAN2, which integrates the adaptive data augmentation (ADA) to alleviate the problem of discriminator overfitting and introduces the idea of cooperative training to stabilize the training process. We design a Peking opera facial makeup image transform conditional generation network TC-StyleGAN2 which is transferred from unconditional generation network. The weights of the unconditional pre-training model are fixed, and an adaptive filtering modulation module is added to modulate the category parameters to complete the conversion from unconditional to conditional StyleGAN2 to deal with the training difficulty of conditional GANs on limited data, which suffer from severe mode collapse. The experimental results show that the proposed training strategy is better than the comparison algorithms, and the image generation quality and diversity have been improved.

A multi-class fundus disease classification system based on an adaptive scale discriminator and hybrid loss

Fundus images are crucial in the observation and detection of ophthalmic diseases. However, detecting multiple ophthalmic diseases from fundus images using deep learning techniques is a complex and challenging task One challenge is the complexity of fundus disease structures, which leads to low detection accuracy. Another challenge is the class imbalance problem common in multi-label image classification, which increases the difficulty of algorithm training and evaluation. To address these issues, this study leverages deep learning to propose an ophthalmic disease classification system. We first employ ResNet50 as the backbone network to extract image features, and then use our designed multi-dimensional attention module and adaptive scale discriminator to enhance the network's ability to detect disease features. During training, we innovatively propose a hybrid loss function method to improve the detection capability on imbalanced data. Finally, we conducted experiments on the ODRI-5K dataset with the proposed classification system. In the test set, our method achieved an AUC of 98.53 and an F1-score of 89.73. This result fully demonstrates the excellent disease classification capability of our method. In summary, the multi-label fundus image disease classification system we proposed exhibits outstanding recognition capability, providing an effective solution for the diagnosis of multi-label fundus image diseases.

A novel hybrid model based on evolving multi-quantile long and short-term memory neural network for ultra-short-term probabilistic forecasting of photovoltaic power

Probabilistic forecasting is extremely crucial in eliminating uncertainty in photovoltaic (PV) power generation. Quantile regression long and short-term memory neural network (QRLSTM) is widely recognized as promising methods for PV power probabilistic forecasting due to their strong generalization ability. However, these models train the model for each quantile individually, which lacks consideration of the correlation and monotonicity between quantiles, and multiple training leads to excessive computational complexity. Furthermore, the non-differentiable pinball loss function generated by QR places significant demands on the optimization algorithms. To address these issues, this paper proposes an evolutive distributed chaotic particle swarm optimization (EDCPSO)-optimized multi-quantile LSTM (MQLSTM) to achieve high-quality probabilistic PV power prediction. MQLSTM is a multi-output network structure that simultaneously outputs all quantile estimates and adopts a loss function with all quantile scores and non-crossing constraints to guide the training of the model. This approach not only improves the quality and reasonableness of quantile estimations, but also reduces computational difficulty. Then, from the perspective of evolutionary computation, considering the weight parameters of each connection layer in MQLSTM as decision variables, we convert the probabilistic forecasting into an optimization problem and propose a EDCPSO to solve the training difficulty. It implements a targeted distributed chaos strategy based on the evolutionary state to improve convergence speed and search capability. The proposed model is tested to be superior in real cases.

Computer vision-based cascade detection of peeling and cracking in masonry house walls

Self-built masonry structures are commonly used as houses in rural areas of the central plains of China. These structures pose significant safety risks due to the low level of self-building and non-standard designs. The primary type of damage to the walls of masonry houses is cracks, while the decorative layers of the walls may also experience peeling or cracking. The shapes, sizes, and surface characteristics of these damages are complex and unique to each type. To achieve fast identification of surface damage on masonry house walls, this paper proposes a cascade detection model that combines a multilevel cascade classifier with a parallel sub-segmentation network. The VGG16 neural network is used as the backbone for a multi-level series classifier. The model is trained by using a damage image set of the wall decorative layers in rural masonry houses. The classification of background and damage, as well as the classification of cracks and peels, are sequentially completed. Then, the parallel sub-segmentation model uses EfficientNet-B7 as the encoder and combines it with the U-Net framework skeleton to perform pixel-level segmentation of peeling and cracks. Finally, the output of parallel sub-segmentation networks is superimposed and fused to generate a complete image containing segmentation information of peeling and cracks. The cascade form network structure adopted in this model significantly reduces the training difficulty and enhances the model accuracy. Compared with Segformer and Deeplabv3+ network, the average IoU of the proposed method for on-site masonry wall images can be increased by 0.29 and 0.08, respectively.

An early warning method of pipeline leakage monitoring with limited leakage samples

Aiming at the problems of low computational efficiency of the real-time transient model (RTTM), and training difficulty of data-driven models with small leakage samples in existing pipeline leakage monitoring methods, this paper proposes a novel pipeline leakage detection model (Multiple Regression Modeling of Real-time Transients, MR-RTTM) based on the fusion of real-time transient mechanism model and data-driven inference. Firstly, the real-time transient mechanism model of pipelines is derived based on the Bernoulli equation, and the required eigenvalues and label values of the data-driven model are determined to achieve the integration of the mechanism and the data-driven model. Secondly, MR-RTTM is constructed by applying the data of pipelines in normal working conditions and combining with the multiple regression algorithm to fit the real-time transient mechanism model of pipelines. An optimal threshold interval selection method of the model is also proposed based on the probability of detection (POD) and the probability of false call (POFC). Experiments show that MR-RTTM can detect pipeline leakage by arranging common pressure and flow rate sensors in the pipeline inlets and outlets, and the minimum detectable leakage is 1 % with an acceptable interval of 80 %∼100 % for POD and 0 ∼ 10 % for POFC, and the leakage discrimination rate is up to 100 %.

Integrating machine learning with dynamic multi-objective optimization for real-time decision-making

Real-time decision-making in dynamic multi-objective optimization problems (DMOPs) is challenging due to constantly changing objectives and constraints. This paper integrates machine learning with Non-dominated Sorting Genetic Algorithm II (NSGA-II) to solve DMOPs and make real-time decisions. Learning-based methods have gained popularity for predicting solutions in new environments and capturing changing patterns in optimal solutions. However, existing approaches often struggle with training difficulty and reduced prediction accuracy due to irrelevant or redundant variables. Therefore, we introduce a new interdependent prediction (IDP) technique to identify correlations between variables and prediction targets and select significant variables for a predictive model. In this way, a better initial population is predicted. The IDP strategy is integrated within the dynamic NSGA-II, introducing a new algorithm called IDP-DNSGA-II. This integration facilitates rapid convergence, finding optimal or near-optimal solutions. The proposed method is evaluated against standard benchmarks, demonstrating superior performance in convergence speed and solution diversity with the changes in the problem environment. The IDP-DNSGA-II is validated through real-world optimization challenges in sustainable automobile production distribution in order-to-delivery systems to enhance environmental sustainability and operational efficiency. This study identifies the minimum frequency of change required in real-world problems to adequately track the optimal decision in real-time.

Research on a semi-supervised soft sensor modelling method for complex chemical processes based on INGO-VMD-ESN

Abstract The dynamic and non-linear nature of complex chemical processes often leads to low prediction accuracy of key quality variables by traditional soft sensors, thus affecting the overall system control accuracy and operational efficiency. Therefore, this paper proposes a semi-supervised soft sensor modelling method based on improved the northern goshawk optimization (INGO)-variable mode decomposition (VMD)-echo state network (ESN). Firstly, a new semi-supervised fusion method is proposed to address the problem of model training difficulty due to the scarcity of labelled samples and process dynamics, which reconstructs the sample dataset by fusing labelled and unlabelled samples into more representative new samples, improving the model’s generalization ability. Secondly, for the noise interference present in the reconstructed data, the input data is denoised using the VMD method to improve the quality of data. Then, a soft sensor model is built based on ESN. Additionally, the denoising and prediction performance of VMD and ESN is significantly affected by parameters, therefore the paper utilizes the INGO algorithm to achieve parameter rectification for VMD and ESN. Finally, the method is validated based on actual sulphur recovery data from a refinery. The results demonstrate that the method effectively mitigates the impact of dynamics and nonlinearity in the complex chemical process which enhances prediction accuracy.

Statistical Metamodel of Liner Acoustic Impedance Based on Neural Network and Probabilistic Learning for Small Datasets

The main novelty of this paper consists of presenting a statistical artificial neural network (ANN)-based model for a robust prediction of the frequency-dependent aeroacoustic liner impedance using an aeroacoustic computational model (ACM) dataset of small size. The model, focusing on percentage of open area (POA) and sound pressure level (SPL) at a zero Mach number, takes into account uncertainties using a probabilistic formulation. The main difficulty in training an ANN-based model is the small size of the ACM dataset. The probabilistic learning carried out using the probabilistic learning on manifolds (PLoM) algorithm addresses this difficulty as it allows constructing a very large training dataset from learning the probabilistic model from a small dataset. A prior conditional probability model is presented for the PCA-based statistical reduced representation of the frequency-sampled vector of the log-resistance and reactance. It induces some statistical constraints that are not straightforwardly taken into account when training such an ANN-based model by classical optimizations methods under constraints. A second novelty of this paper consists of presenting an alternate solution that involves using conditional statistics estimated with learned realizations from PLoM. A numerical example is presented.

Review of imbalanced fault diagnosis technology based on generative adversarial networks

Abstract In the field of industrial production, machine failures not only negatively affect productivity and product quality, but also lead to safety accidents, so it is crucial to accurately diagnose machine failures in time and take appropriate measures. However, machines cannot operate with faults for extended periods, and the diversity of fault modes results in limited data collection, posing challenges to building accurate fault prediction models. Despite recent advancements, intelligent fault diagnosis methods based on traditional sampling and machine learning have shown notable progress. Nonetheless, these methods heavily rely on human expertise, making it challenging to extract comprehensive feature information. To address these challenges, numerous imbalance fault diagnosis methods based on generative adversarial networks (GANs) have emerged, GANs can generate realistic samples that conform to the distribution of the original data, showing promising results in diagnosing imbalances in critical components such as bearings and gears, despite their great potential, GAN methods also face challenges, including difficulties in training and generating abnormal samples. However, whether it is GAN-based resampling technology or traditional sampling technology, there are fewer reviews on noise-containing imbalance, intra- and inter-class dual imbalance, multi-class imbalance, time series imbalance and other problems in small samples, and there is a lack of a more comprehensive summary of the solutions to the above imbalance problems. Therefore, the purpose of this paper is to deeply explore the imbalance problems under various failure modes, and review and analyze the research methods and results based on GANs on this basis. By suggesting future research directions, this paper aims to provide guidance and reference for research in the field of industrial production maintenance.

Vision Improvement in Keratoconus Patients Trained With Perceptual Learning: A Randomized Controlled Trial

BackgroundKeratoconus (KC) is a corneal ectasia disease in which the vision of some patients cannot achieve satisfaction by spectacle corrections. However, not everyone can embrace contact lenses to achieve better vision. Perceptual learning (PL) is a potential treatment for vision improvement in such patients. PurposeTo investigate the effectiveness and maintenance of PL on vision improvement in KC patients corrected with spectacles. DesignRandomized, double-blind clinical trial. ParticipantsThirty-five non-progressive KC patients aged 9 years or older with unsatisfied spectacle-corrected vision were enrolled. MethodsNon-progressive KC patients with best spectacle-corrected distance visual acuity (CDVA) of 0 to 1.0 logMAR (Snellen equivalent range 20/20 to 20/200) and contact lenses intolerant were enrolled. Eligible subjects were randomized into PL and control groups to receive PL and placebo training for 3 months, respectively. Spectacle-corrected visual acuity, contrast sensitivity function (CSF), stereoacuity, and visual functioning and quality of life questionnaires were measured at baseline, 3 months, and 6 months of follow-up. Statistics were analyzed following the intention-to-treat (ITT) principle. ResultsAfter 3 months of training, the CDVA of patients in the PL group improved as compared to the placebo group (0.17 ± 0.15 logMAR vs. 0.02 ± 0.06 logMAR; P = 0.0006). Eight out of seventeen (47.06 %) patients in the PL group reached CDVA improvement ≥ 2 lines (P=0.0010). This improvement persisted for at least 6 months (from baseline) as compared to the placebo group (0.17 ± 0.17 logMAR vs. 0.01 ± 0.07 logMAR; P = 0.0011). The increase of CSF in the PL group mainly was found for moderate spatial frequency (0.11 ± 0.17 log units at 3 cpd; 0.12 ± 0.19 log units at 6 cpd). Linear regression indicated that patients with worse initial CDVA achieved better gains in CDVA after PL (P = 0.009). No side effects were observed and no subjects quit because of training difficulties. ConclusionThree-month perceptual learning improved vision in KC patients and the improvement maintained after 3 months of treatment cessation. The results indicate that perceptual learning may be a promising therapy for KC patients with unsatisfied spectacle-corrected visual acuity.

Enhancing Temporal Knowledge Graph Representation with Curriculum Learning

Temporal knowledge graph representation approaches encounter significant challenges in handling the complex dynamic relations among entities, relations, and time. These challenges include the high difficulty of training and poor generalization performance, particularly with large datasets. To address these issues, this paper introduces curriculum learning strategies from machine learning, aiming to improve learning efficiency through effective curriculum planning. The proposed framework constructs a high-dimensional filtering model based on graph-based high-order receptive fields and employs a scoring model that uses a curriculum temperature strategy to evaluate the difficulty of temporal knowledge graph data quadruples at each stage. By progressively expanding the receptive field and dynamically adjusting the difficulty of learning samples, the model can better understand and capture multi-level information within the graph structure, thereby improving its generalization capabilities. Additionally, a temperature factor is introduced during model training to optimize parameter gradients, alongside a gradually increasing training strategy to reduce training difficulty. Experiments on the benchmark datasets ICEWS14 and ICEWS05-15 demonstrate that this framework not only significantly enhances model performance on these datasets but also substantially reduces training convergence time.

Attention-based stackable graph convolutional network for multi-view learning

In multi-view learning, graph-based methods like Graph Convolutional Network (GCN) are extensively researched due to effective graph processing capabilities. However, most GCN-based methods often require complex preliminary operations such as sparsification, which may bring additional computation costs and training difficulties. Additionally, as the number of stacking layers increases in most GCN, over-smoothing problem arises, resulting in ineffective utilization of GCN capabilities. In this paper, we propose an attention-based stackable graph convolutional network that captures consistency across views and combines attention mechanism to exploit the powerful aggregation capability of GCN to effectively mitigate over-smoothing. Specifically, we introduce node self-attention to establish dynamic connections between nodes and generate view-specific representations. To maintain cross-view consistency, a data-driven approach is devised to assign attention weights to views, forming a common representation. Finally, based on residual connectivity, we apply an attention mechanism to the original projection features to generate layer-specific complementarity, which compensates for the information loss during graph convolution. Comprehensive experimental results demonstrate that the proposed method outperforms other state-of-the-art methods in multi-view semi-supervised tasks.

Breaking through clouds: A hierarchical fusion network empowered by dual-domain cross-modality interactive attention for cloud-free image reconstruction

Cloud obscuration undermines the availability of optical images for continuous monitoring in earth observation. Fusing features from synthetic aperture radar (SAR) has been recognized as a feasible strategy to guide the reconstruction of corrupted signals in cloud-contaminated regions. However, due to the different imaging mechanisms and reflection characteristics, the substantial domain gap between SAR images and optical images makes it a challenging problem to effectively perform cross-modality feature fusion. Although several SAR-assisted cloud removal methods have been proposed, most of them are often unable to achieve adequate information interaction between different modalities, which greatly limits the effectiveness and reliability of cross-modality fusion. In this paper, we proposed a novel hierarchical framework for cloud-free multispectral image reconstruction, which effectively integrates SAR and optical data by a dual-domain interactive attention mechanism. The overall encoder–decoder network is a W-shaped asymmetric structure with a two-branch encoder and one decoder. The encoder branches extract features from SAR images and optical images separately, while the decoder exploits multiscale residual block groups to expand the receptive field and multi-output strategy for reducing the training difficulty. The core cross-modality feature fusion module at the bottleneck adopts the dual-domain interactive attention (DDIA) mechanism which enhances the reciprocal infusion of SAR and optical features to encourage the reconstruction of spectral and structural information. Furthermore, features in the spatial and frequency domains are integrated to improve the effectiveness of the fusion process. To echo the overall network structure, the loss function is designed as a multiscale loss in dual domains. The proposed method can realize sufficient information communication and effective cross-modality fusion between SAR features and optical features. Extensive experiments on the SMILE-CR dataset and SEN12MS-CR dataset demonstrated that the proposed method can outperform the seven representative deep-learning comparative methods in terms of visual performance and quantitative accuracy.

A randomized clinical trial to evaluate feasibility, tolerability, and preliminary target engagement for a novel executive working memory training in adolescents with ADHD

ObjectiveWorking memory training for Attention-Deficit/Hyperactivity Disorder (ADHD) has focused on increasing working memory capacity, with inconclusive evidence for its effectiveness. Alternative training targets are executive working memory (EWM) processes that promote flexibility or bolster stability of working memory contents to guide behavior via selective attention. This randomized, placebo-controlled study was designed to assess feasibility, tolerability, and behavioral target engagement of a novel EWM training for ADHD. Method62 ADHD-diagnosed adolescents (12–18 years) were randomized to EWM training or placebo arms for 20 remotely coached sessions conducted over 4–5 weeks. Primary outcome measures were behavioral changes on EWM tasks. Secondary outcomes were intervention tolerability, trial retention, and responsiveness to adaptive training difficulty manipulations. ResultsLinear regression analyses found intervention participants showed medium effect size improvements, many of which were statistically significant, on Shifting and Filtering EWM task accuracy and Shifting and Updating reaction time measures. Intervention participants maintained strong self-rated motivation, mood, and engagement and progressed through the adaptive difficulty measures, which was further reflected in high trial retention. ConclusionsThe results suggest that these EWM processes show promise as training targets for ADHD. The subsequent NIMH R33-funded extension clinical trial will seek to replicate and extend these findings.

A Method for Predicting High-Resolution 3D Variations in Temperature and Salinity Fields Using Multi-Source Ocean Data

High-resolution three-dimensional (3D) variations in ocean temperature and salinity fields are of great significance for ocean environment monitoring. Currently, AI-based 3D temperature and salinity field predictions rely on expensive 3D data, and as the prediction period increases, the stacking of high-resolution 3D data greatly increases the difficulty of model training. This paper transforms the prediction of 3D temperature and salinity into the prediction of sea surface elements and the inversion of subsurface temperature and salinity using sea surface elements, by leveraging the relationship between sea surface factors and subsurface temperature and salinity. This method comprehensively utilizes multi-source ocean data to avoid the issue of data volume caused by stacking high-resolution historical data. Specifically, the model first utilizes 1/4° low-resolution satellite remote sensing data to construct prediction models for sea surface temperature (SST) and sea level anomaly (SLA), and then uses 1/12° high-resolution temperature and salinity data as labels to build an inversion model of subsurface temperature and salinity based on SST and SLA. The prediction model and inversion model are integrated to obtain the final high-resolution 3D temperature and salinity prediction model. Experimental results show that the 20-day prediction results in the two sea areas of the coastal waters of China and the Northwest Pacific show good performance, accurately predicting ocean temperature and salinity in the vast majority of layers, and demonstrate higher resource utilization efficiency.

What is the Total Ankle Arthroplasty Experience of Podiatric Foot and Ankle Surgery Fellows? A National Survey

In 2011, the Council of Podiatric Medical Education, the accrediting body of the American Podiatric Medical Association, approved the conversion of all Podiatric Residencies to 3-year surgical programs. In 2012, there were 12 podiatric fellowships recognized by the American College of Foot and Ankle Surgeons. To date, there are 53 programs listed under the college's website. As podiatric fellowships expand further research is needed to identify advantages and difficulties of fellowship training. One of the most salient topics is Total Ankle Replacement. Our primary aim was to obtain current fellow survey data to formulate improved outcomes of Total Ankle Arthroplasty and enhance our understanding of podiatric reconstructive foot and ankle surgery fellowship training programs. Invitation was administered by email and 73.6% of active reconstructive 2023-24 American College of Foot and Ankle Surgeons post-graduate fellows responded. Fellowship total ankle replacement case volume was significantly greater than residency (p = 0.037). Completion of 0-5 total ankle replacement(s) was 30.8%, and greater than 30 in 17.9% of fellows. Fifty nine percent reported feeling “comfortable” or “very comfortable” with total ankle arthroplasty. Patient specific instrumentation was used in a majority of cases by 66.7% of programs. Over three fourths (79.8%) of fellows stated they planned on performing TAR as an attending surgeon after their fellowship. Despite its limitations, we hope our survey data can aid graduating and previous fellows and add to the body of knowledge for future TAR educational programs and industry involvement. As podiatric fellowships continue to transform, so too must our research efforts to track progress.

Audio2Gestures: Generating Diverse Gestures From Audio.

People may perform diverse gestures affected by various mental and physical factors when speaking the same sentences. This inherent one-to-many relationship makes co-speech gesture generation from audio particularly challenging. Conventional CNNs/RNNs assume one-to-one mapping, and thus tend to predict the average of all possible target motions, easily resulting in plain/boring motions during inference. So we propose to explicitly model the one-to-many audio-to-motion mapping by splitting the cross-modal latent code into shared code and motion-specific code. The shared code is expected to be responsible for the motion component that is more correlated to the audio while the motion-specific code is expected to capture diverse motion information that is more independent of the audio. However, splitting the latent code into two parts poses extra training difficulties. Several crucial training losses/strategies, including relaxed motion loss, bicycle constraint, and diversity loss, are designed to better train the VAE. Experiments on both 3D and 2D motion datasets verify that our method generates more realistic and diverse motions than previous state-of-the-art methods, quantitatively and qualitatively. Besides, our formulation is compatible with discrete cosine transformation (DCT) modeling and other popular backbones (i.e., RNN, Transformer). As for motion losses and quantitative motion evaluation, we find structured losses/metrics (e.g. STFT) that consider temporal and/or spatial context complement the most commonly used point-wise losses (e.g. PCK), resulting in better motion dynamics and more nuanced motion details. Finally, we demonstrate that our method can be readily used to generate motion sequences with user-specified motion clips on the timeline.