We propose a novel framework to solve risk-sensitive reinforcement learning (RL) problems where the agent optimises time-consistent dynamic spectral risk measures. Based on the notion of conditional elicitability, our methodology constructs (strictly consistent) scoring functions that are used as penalizers in the estimation procedure. Our contribution is threefold: we (i) devise an efficient approach to estimate a class of dynamic spectral risk measures with deep neural networks, (ii) prove that these dynamic spectral risk measures may be approximated to any arbitrary accuracy using deep neural networks, and (iii) develop a risk-sensitive actor-critic algorithm that uses full episodes and does not require any additional nested transitions. We compare our conceptually improved reinforcement learning algorithm with the nested simulation approach and illustrate its performance in two settings: statistical arbitrage and portfolio allocation on both simulated and real data.
Considering that acoustic scenes and sound events are closely related to each other, in some previous papers, a joint analysis of acoustic scenes and sound events utilizing multitask learning (MTL)-based neural networks was proposed. In conventional methods, a strongly supervised scheme is applied to sound event detection in MTL models, which requires strong labels of sound events in model training; however, annotating strong event labels is quite time-consuming. In this paper, we thus propose a method for the joint analysis of acoustic scenes and sound events based on the MTL framework with weak labels of sound events. In particular, in the proposed method, we introduce the multiple-instance learning scheme for weakly supervised training of sound event detection and evaluate four pooling functions, namely, max pooling, average pooling, exponential softmax pooling, and attention pooling. Experimental results obtained using parts of the TUT Acoustic Scenes 2016/2017 and TUT Sound Events 2016/2017 datasets show that the proposed MTL-based method with weak labels outperforms the conventional single-task-based scene classification and event detection models with weak labels in terms of both the scene classification and event detection performances.
Early diagnosis of disease can result in improved health outcomes, such as higher survival rates and lower treatment costs. With the massive amount of information in electronic health records (EHRs), there is great potential to use machine learning (ML) methods to model disease progression aimed at early prediction of disease onset and other outcomes. In this work, we employ recent innovations in neural ODEs to harness the full temporal information of EHRs. We propose ICE-NODE (Integration of Clinical Embeddings with Neural Ordinary Differential Equations), an architecture that temporally integrates embeddings of clinical codes and neural ODEs to learn and predict patient trajectories in EHRs. We apply our method to the publicly available MIMIC-III and MIMIC-IV datasets, reporting improved prediction results compared to state-of-the-art methods, specifically for clinical codes that are not frequently observed in EHRs. We also show that ICE-NODE is more competent at predicting certain medical conditions, like acute renal failure and pulmonary heart disease, and is also able to produce patient risk trajectories over time that can be exploited for further predictions.
Real-world image super-resolution is a practical image restoration problem that aims to obtain high-quality images from in-the-wild input, has recently received considerable attention with regard to its tremendous application potentials. Although deep learning-based methods have achieved promising restoration quality on real-world image super-resolution datasets, they ignore the relationship between L1- and perceptual- minimization and roughly adopt auxiliary large-scale datasets for pre-training. In this paper, we discuss the image types within a corrupted image and the property of perceptual- and Euclidean- based evaluation protocols. Then we propose a method, Real-World image Super-Resolution by Exclusionary Dual-Learning (RWSR-EDL) to address the feature diversity in perceptual- and L1- based cooperative learning. Moreover, a noise-guidance data collection strategy is developed to address the training time consumption in multiple datasets optimization. When an auxiliary dataset is incorporated, RWSR-EDL achieves promising results and repulses any training time increment by adopting the noise-guidance data collection strategy. Extensive experiments show that RWSR-EDL achieves competitive performance over state-of-the-art methods on four in-the-wild image super-resolution datasets.
The excessive runtime of high-fidelity partial differential equation (PDE) solvers makes them unsuitable for time-critical applications. We propose to accelerate PDE solvers using reduced-order modeling (ROM). Whereas prior ROM approaches reduce the dimensionality of discretized vector fields, our continuous reduced-order modeling (CROM) approach builds a smooth, low-dimensional manifold of the continuous vector fields themselves, not their discretization. We represent this reduced manifold using neural fields, relying on their continuous and differentiable nature to efficiently solve the PDEs. CROM may train on any and all available numerical solutions of the continuous system, even when they are obtained using diverse methods or discretizations. After the low-dimensional manifolds are built, solving PDEs requires significantly less computational resources. Since CROM is discretization-agnostic, CROM-based PDE solvers may optimally adapt discretization resolution over time to economize computation. We validate our approach on an extensive range of PDEs with training data from voxel grids, meshes, and point clouds. Large-scale experiments demonstrate that our approach obtains speed, memory, and accuracy advantages over prior ROM approaches while gaining 109$\times$ wall-clock speedup over full-order models on CPUs and 89$\times$ speedup on GPUs.
This paper establishes a multi-scattering point chest wall motion model by combining the human respiration signal (RS) and HS (HS) measured by radar. An algorithmic process is designed based on the model to accurately separate the human respiration and heartbeat motion. Firstly, a human maximum motion velocity constraint method is proposed to correct human chest wall tracking, determine the radial position of the chest wall relative to the radar, and extract the phase signal corresponding to the chest wall motion. Then an improved time-difference method is proposed to suppress the interference of RS harmonics on HS and the interference of low-frequency noise on RS. Finally, an adaptive Gaussian weighting filter is designed to extract the RS with less distortion from the phase signal. A low-order finite-length unit impulse response (FIR) filter is used to extract the HS with less distortion from the phase signal. To verify the effectiveness of the proposed algorithm, simulating the process of measuring the RS and HS of the chest wall motion model by radar. The simulation results show that, ideally, the radar measurement results of the RS and HS are less distorted relative to the actual values. In addition, we used a millimeter-wave experimental radar system in the 60 GHz band to measure the respiration rate (RR) and HR (HR) of two subjects. The experimental results showed that the measured RR and HR correlated well with the actual values. The quantitative analysis of simulation results and experimental results show that the proposed method can achieve accurate and robust measurement of RS and HS.
Artificial intelligence(AI)-assisted method had received much attention in the risk field such as disease diagnosis. Different from the classification of disease types, it is a fine-grained task to classify the medical images as benign or malignant. However, most research only focuses on improving the diagnostic accuracy and ignores the evaluation of model reliability, which limits its clinical application. For clinical practice, calibration presents major challenges in the low-data regime extremely for over-parametrized models and inherent noises. In particular, we discovered that modeling data-dependent uncertainty is more conducive to confidence calibrations. Compared with test-time augmentation(TTA), we proposed a modified Bootstrapping loss(BS loss) function with Mixup data augmentation strategy that can better calibrate predictive uncertainty and capture data distribution transformation without additional inference time. Our experiments indicated that BS loss with Mixup(BSM) model can halve the Expected Calibration Error(ECE) compared to standard data augmentation, deep ensemble and MC dropout. The correlation between uncertainty and similarity of in-domain data is up to -0.4428 under the BSM model. Additionally, the BSM model is able to perceive the semantic distance of out-of-domain data, demonstrating high potential in real-world clinical practice.
In this technical report, we introduce our solution to human-centric spatio-temporal video grounding task. We propose a concise and effective framework named STVGFormer, which models spatiotemporal visual-linguistic dependencies with a static branch and a dynamic branch. The static branch performs cross-modal understanding in a single frame and learns to localize the target object spatially according to intra-frame visual cues like object appearances. The dynamic branch performs cross-modal understanding across multiple frames. It learns to predict the starting and ending time of the target moment according to dynamic visual cues like motions. Both the static and dynamic branches are designed as cross-modal transformers. We further design a novel static-dynamic interaction block to enable the static and dynamic branches to transfer useful and complementary information from each other, which is shown to be effective to improve the prediction on hard cases. Our proposed method achieved 39.6% vIoU and won the first place in the HC-STVG track of the 4th Person in Context Challenge.
Photorealistic telepresence requires both high-fidelity body modeling and faithful driving to enable dynamically synthesized appearance that is indistinguishable from reality. In this work, we propose an end-to-end framework that addresses two core challenges in modeling and driving full-body avatars of real people. One challenge is driving an avatar while staying faithful to details and dynamics that cannot be captured by a global low-dimensional parameterization such as body pose. Our approach supports driving of clothed avatars with wrinkles and motion that a real driving performer exhibits beyond the training corpus. Unlike existing global state representations or non-parametric screen-space approaches, we introduce texel-aligned features -- a localised representation which can leverage both the structural prior of a skeleton-based parametric model and observed sparse image signals at the same time. Another challenge is modeling a temporally coherent clothed avatar, which typically requires precise surface tracking. To circumvent this, we propose a novel volumetric avatar representation by extending mixtures of volumetric primitives to articulated objects. By explicitly incorporating articulation, our approach naturally generalizes to unseen poses. We also introduce a localized viewpoint conditioning, which leads to a large improvement in generalization of view-dependent appearance. The proposed volumetric representation does not require high-quality mesh tracking as a prerequisite and brings significant quality improvements compared to mesh-based counterparts. In our experiments, we carefully examine our design choices and demonstrate the efficacy of our approach, outperforming the state-of-the-art methods on challenging driving scenarios.
Traffic flow forecasting on graphs has real-world applications in many fields, such as transportation system and computer networks. Traffic forecasting can be highly challenging due to complex spatial-temporal correlations and non-linear traffic patterns. Existing works mostly model such spatial-temporal dependencies by considering spatial correlations and temporal correlations separately and fail to model the direct spatial-temporal correlations. Inspired by the recent success of transformers in the graph domain, in this paper, we propose to directly model the cross-spatial-temporal correlations on the spatial-temporal graph using local multi-head self-attentions. To reduce the time complexity, we set the attention receptive field to the spatially neighboring nodes, and we also introduce an adaptive graph to capture the hidden spatial-temporal dependencies. Based on these attention mechanisms, we propose a novel Adaptive Graph Spatial-Temporal Transformer Network (ASTTN), which stacks multiple spatial-temporal attention layers to apply self-attention on the input graph, followed by linear layers for predictions. Experimental results on public traffic network datasets, METR-LA PEMS-BAY, PeMSD4, and PeMSD7, demonstrate the superior performance of our model.