Data-driven motion priors that can guide agents toward producing naturalistic behaviors play a pivotal role in creating life-like virtual characters. Adversarial imitation learning has been a highly effective method for learning motion priors from reference motion data. However, adversarial priors, with few exceptions, need to be retrained for each new controller, thereby limiting their reusability and necessitating the retention of the reference motion data when applied to downstream tasks. In this work, we present Score-Matching Motion Priors (SMP), which leverages pre-trained motion diffusion models and score distillation sampling (SDS) to create reusable task-agnostic motion priors. SMPs can be pre-trained on a motion dataset, independent of any control policy or task. Once trained, SMPs can be kept frozen and reused as general-purpose reward functions to train new policies to produce naturalistic behaviors for downstream tasks. We show that a general motion prior trained on large-scale datasets can be repurposed into a variety of style-specific priors. Furthermore, SMP can compose different styles to synthesize new styles not present in the original dataset. Our method can create reusable and modular motion priors that produce high-quality motions comparable to state-of-the-art adversarial imitation learning methods. In our experiments, we demonstrate the effectiveness of SMP across a diverse suite of control tasks with physically simulated humanoid characters. Video available at youtu.be/jBA2tWk6vzU.

Motion capture (mocap) data often exhibits visually jarring artifacts due to inaccurate sensors and post-processing. Cleaning this corrupted data can require substantial manual effort from human experts, which can be a costly and time-consuming process. Previous data-driven motion cleanup methods offer the promise of automating this cleanup process, but often require in-domain paired corrupted-to-clean training data. Constructing such paired datasets requires access to high-quality, relatively artifact-free motion clips, which often necessitates laborious manual cleanup. In this work, we present StableMotion, a simple yet effective method for training motion cleanup models directly from unpaired corrupted datasets that need cleanup. The core component of our method is the introduction of motion quality indicators, which can be easily annotated— through manual labeling or heuristic algorithms—and enable training of quality-aware motion generation models on raw motion data with mixed quality. At test time, the model can be prompted to generate high-quality motions using the quality indicators. Our method can be implemented through a simple diffusion-based framework, leading to a unified motion generate-discriminate model, which can be used to both identify and fix corrupted frames. We demonstrate that our proposed method is effective for training motion cleanup models on raw mocap data in production scenarios by applying StableMotion to SoccerMocap, a 245-hour soccer mocap dataset containing real-world motion artifacts. The trained model effectively corrects a wide range of motion artifacts, reducing motion pops and frozen frames by 68% and 81%, respectively. On our benchmark dataset, we further show that cleanup models trained with our method on unpaired corrupted data outperform state-of-the-art methods trained on clean or paired data, while also achieving comparable performance in preserving the content of the original motion clips.

The brain's biological age, or brain age, can be inferred using electroencephalography (EEG) and may reflect various factors, including early-life influences or the gradual accumulation of pathological changes. We aimed to investigate the relationship between brain age and age-related changes by evaluating the predictive accuracy of brain age models across different stages of adulthood. We hypothesized that if brain age predominantly reflects the accumulation of stochastic brain damage, the models would better predict the age of younger adults than older adults. We analyzed routine clinical EEG data from public hospitals, categorizing outpatients into young, middle-aged, and older age groups. Source activity was reconstructed from the EEG recordings. We focused on spectral power within six brain lobes defined by the Destrieux brain atlas. Random forest models were trained using EEG features from middle-aged adults to predict brain age. We then evaluated the models' accuracy in young and older adults by assessing the R-squared values of the predictions. Contrary to our hypothesis, the predictive performance, on average, did not decline with increasing age, suggesting that brain age may not primarily reflect a gradual accumulation of damage. These findings imply that individual differences in brain rhythms may remain relatively stable across adulthood, indicating that other factors contribute to brain aging beyond accumulated pathological changes.

The global COVID-19 pandemic increased the interest in automatic analysis and classification of cough sounds. However, developing such a system requires a large dataset of expert-labelled cough sounds, which remains elusive. Data augmentation techniques are often employed to train machine learning models with limited data, while ensuring model robustness to real-world variations. We have recently proposed the use of “natural” spectral data augmentation methods, including noise and reverberation [1]. In this paper, we further investigate the method by studying the alignment between training and testing environments. We augment training sound data with varying levels of reverberation and Gaussian noise, and evaluate augmented convolutional neural network models across a range of test environments on two audio classification tasks: speech command recognition and cough sound classification. Results demonstrate the broad robustness of the data-augmented deep learning model across various test environments. Furthermore, the augmented model's performance on cough classification is compared to 4 expert annotations on a large cough dataset and is seen to be near-human-level accuracy. This augmented model is recommended for classifying cough sounds in natural settings beyond laboratory conditions.

Routine clinical EEG is a standard test used for the neurological evaluation of patients. A trained specialist interprets EEG recordings and classifies them into clinical categories. Given time demands and high inter-reader variability, there is an opportunity to facilitate the evaluation process by providing decision support tools that can classify EEG recordings automatically. Classifying clinical EEG is associated with several challenges: classification models are expected to be interpretable; EEGs vary in duration and EEGs are recorded by multiple technicians operating various devices. Our study aimed to test and validate a framework for EEG classification which satisfies these requirements by transforming EEG into unstructured text. We considered a highly heterogeneous and extensive sample of routine clinical EEGs (n = 5785), with a wide range of participants aged between 15 and 99 years. EEG scans were recorded at a public hospital, according to 10/20 electrode positioning with 20 electrodes. The proposed framework was based on symbolizing EEG signals and adapting a previously proposed method from natural language processing (NLP) to break symbols into words. Specifically, we symbolized the multichannel EEG time series and applied a byte-pair encoding (BPE) algorithm to extract a dictionary of the most frequent patterns (tokens) reflecting the variability of EEG waveforms. To demonstrate the performance of our framework, we used newly-reconstructed EEG features to predict patients' biological age with a Random Forest regression model. This age prediction model achieved a mean absolute error of 15.7 years. We also correlated tokens' occurrence frequencies with age. The highest correlations between the frequencies of tokens and age were observed at frontal and occipital EEG channels. Our findings demonstrated the feasibility of applying an NLP-based approach to classifying routine clinical EEG. Notably, the proposed algorithm could be instrumental in classifying clinical EEG with minimal preprocessing and identifying clinically-relevant short events, such as epileptic spikes.

Cough diagnosis is important for the elderly population, since cough is a key symptom of many respiratory illnesses and conditions. This paper introduces a Transformer-based feature learning approach for the analysis of cough recordings. A Transformer network leveraging feature learning on a big data set is investigated from a feature engineering perspective, in order to find dedicated classification models that can improve overall performance. The latter was achieved through adopting AutoML post-processing techniques on different data sets, driven by the feature engineering process based on both feature selection and feature generation via nonlinear methods. It was found that this approach led to substantial improvements (in the order of 17% from 0.818 to 0.956 of accuracy) on practically all metrics of classification performance, with respect t o t hose obtained with standalone Transformers. Moreover, AutoML models using reduced number of features, either selected or generated, resulted in higher quality models. In particular, a model working only with 1.2 % of the features (nonlinearly generated from the 768 produced by the Transformer), outperformed the model using all of them. These results highlight that big data-derived machine learning models, when post-processed, can play an important role in adapting to small-data scenarios.

Supervised capsule networks are theoretically advantageous over convolutional neural networks, because they aim to model a range of transformations of local physical or abstract objects and part-whole relationships among them. However, it remains unclear how to use the concept of capsules in deep generative models. In this study, to address this challenge, we present a statistical modelling of capsules in deep generative models where distributions are formulated in the exponential family. The major contribution of this unsupervised method is that parse trees as representations of part-whole relationships can be dynamically learned from the data.

There is increasing interest in understanding the cognitive and physiological state of operators in safety critical situations (e.g. pilots), specifically as it relates to task difficulty and mental workload. Herein, we evaluate the potential of electrocardiography (ECG) and electroencephalography (EEG) for detecting in-flight changes in helicopter pilot workload. Two National Research Council Canada test pilots performed a series of flight maneuvers in an NRC Bell 205 helicopter which involved a target tracking task with three levels of difficulty. Subjective ratings of pilot workload were collected using the Cooper-Harper handling quality ratings scale and pilot control activity was quantified based on cyclic control movements. ECG derived measures of heart rate and heart rate variability, as well as EEG derived measures of power in three frequency bands (theta 4-8Hz; alpha 8-13Hz; beta 13-22Hz), were computed and compared across task difficulty levels. A set of support vector machine (SVM) regressors were trained and tested to differentiate the three difficulty levels from ECG and EEG features. Differences in subjective ratings and control activity metrics confirmed the task difficulty manipulations (pECG= 0.17) performing better than the EEG-based regressor (minimum cross-validation MSEEEG= 0.29). This study provides an initial application demonstration of physiological and cognitive metrics and machine learning approaches for detecting differences in task difficulty during helicopter flight. This is the necessary first step for further development of passive brain computer interfaces for real-time in-flight monitoring of helicopter pilot workload.

Analysis on a dataset of 3D scanned surfaces have presented problems because of incompleteness on the surfaces and because of variances in shape, size and pose. In this paper, a high-resolution generic model is aligned to data in the Civilian American and European Surface Anthropometry Resources (CAESAR) database in order to obtain a consistent parameterization. A Radial Basis Function (RBF) network is built for rough deformation by using landmark information from the generic model, anatomical landmarks provided by CAESAR dataset and virtual landmarks created automatically for geometric deformation. Fine mapping then successfully applies a weighted sum of errors on both surface data and the smoothness of deformation. Compared with previous methods, our approach makes robust alignment in a higher efficiency. This consistent parameterization also makes it possible for Principal Components Analysis (PCA) on the whole body as well as human body segments. Our analysis on segmented bodies displays a richer variation than that of the whole body. This analysis indicates that a wider application of human body reconstruction with segments is possible in computer animation.

We present a data-driven approach to build a human body model from a single photograph by performing Principal Component Analysis (PCA) on a database of body segments. We segment a collection of human bodies to compile the required database prior to performing the analysis. Our approach then builds a single PCA for each body segment - head, left and right arms, torso and left and right legs - yielding six PCAs in total. This strategy improves on the flexibility of conventional data-driven approaches in 3D modeling and allows our approach to take variations in ethnicity, age and body posture into account. We demonstrate our approach in practice by constructing models of a Caucasian male, an Asian male and a toddler from corresponding photographs and a Caucasian adult oriented database. We also discuss rapid consistent parameterization based on Radial Basis Functions (RBF) and non-optimization based learning systems to reduce execution time.