Recent advances in neural Granger causal methods have shown promise in modeling temporal nonlinear dependencies. However, existing approaches remain confined to raw time-series data, inherently lacking contextual semantics and tending to overfit, which undermines their real-world applicability. To address these challenges, we propose LLM-GC, a novel LLM-empowered multimodal Granger causality discovery framework that enriches unimodal temporal dynamics with semantic priors and world knowledge distilled from large language models (LLMs). LLM-GC leverages dual-modality encoding to capture and align temporal and contextual dynamics by Cross-Modal Dual Retrieval while avoiding causal entanglement across modalities. To extract multimodal causal features, we introduce a causality-aware self-attention mechanism by simply inverting the conventional self-attention structure, enabling a shared causality augmenter to effectively highlight consistent causal patterns across modalities. LLM-GC is the first to bridge LLMs and Granger causality, and experiments on synthetic and real-world benchmark datasets demonstrate that LLM-GC outperforms existing state-of-the-art methods in Granger causal discovery.

Discovering Granger causality from time series data is fundamental to understanding dynamic systems, yet most existing methods struggle with unknown intervention targets or causal structures in real-world scenarios. In this paper, we propose DiffuGC, a novel diffusion-based framework that unifies observational and interventional causal discovery through a generative denoising process. By introducing diffusive interventions, which apply progressive interventions without any prior knowledge, DiffuGC amplifies causal signals while preserving structural information. Furthermore, we introduce a denoising NoiFormer with adaptive attention to both short- and long-term causal dependencies, which disentangles trend and seasonal components to enable accurate reconstruction of causal structures from interventional data. To the best of our knowledge, we are the first to integrate diffusion models with interventional Granger causal discovery. Extensive experiments on synthetic, quasi-real, and real-world benchmarks demonstrate that DiffuGC consistently outperforms state-of-the-art baselines in both observational and interventional data. Moreover, we introduce an intriguing notion, Causality Acceleration, characterized by the early emergence of informative causal patterns within the diffusion path, which may open up promising directions for future research on efficient and adaptive causal discovery.

Electrocardiogram (ECG) is essential for the clinical diagnosis of arrhythmias and other heart diseases, but deep learning methods based on ECG often face limitations due to the need for high-quality annotations. Although previous ECG self-supervised learning (eSSL) methods have made significant progress in representation learning from unannotated ECG data, they typically treat ECG signals as ordinary time-series data, segmenting the signals using fixed-size and fixed-step time windows, which often ignore the form and rhythm characteristics and latent semantic relationships in ECG signals. In this work, we introduce a novel perspective on ECG signals, treating heartbeats as words and rhythms as sentences. Based on this perspective, we first designed the QRS-Tokenizer, which generates semantically meaningful ECG sentences from the raw ECG signals. Building on these, we then propose HeartLang, a novel self-supervised learning framework for ECG language processing, learning general representations at form and rhythm levels. Additionally, we construct the largest heartbeat-based ECG vocabulary to date, which will further advance the development of ECG language processing. We evaluated HeartLang across six public ECG datasets, where it demonstrated robust competitiveness against other eSSL methods. Our data and code are publicly available at https://github.com/PKUDigitalHealth/HeartLang.

Imbalanced data distributions are prevalent in real-world scenarios, presenting significant challenges in both classification and regression tasks. This imbalance often causes deep learning models to overfit in regions with abundant data (manyshot regions) while underperforming in regions with sparse data (few-shot regions). Such characteristics limit the applicability of deep learning models across various domains, notably in healthcare, where rare cases often carry greater clinical significance. While recent studies have highlighted the benefits of incorporating distributional information in imbalanced classification tasks, similar strategies have been largely unexplored in imbalanced regression. To address this gap, we propose Dist Loss, a novel loss function that integrates distributional information into model training by jointly optimizing the distribution distance between model predictions and target labels, alongside sample-wise prediction errors. This dual-objective approach encourages the model to balance its predictions across different label regions, leading to significant improvements in accuracy in fewshot regions. We conduct extensive experiments across three datasets spanning computer vision and healthcare: IMDB-WIKI-DIR, AgeDB-DIR, and ECG-KDIR. The results demonstrate that Dist Loss effectively mitigates the impact of imbalanced data distributions, achieving state-of-the-art performance in few-shot regions. Furthermore, Dist Loss is easy to integrate and complements existing methods. To facilitate further research, we provide our implementation at https://github.com/Ngk03/DIR-Dist-Loss.

While recent multimodal large language models (MLLMs) have advanced automated ECG interpretation, they still face two key limitations: (1) insufficient multimodal synergy between ECG time series and ECG images, and (2) limited explainability in linking diagnoses to granular waveform evidence. We introduce GEM, the first MLLM unifying ECG time series, 12-lead ECG images and text for grounded and clinician-aligned ECG interpretation. GEM enables feature-grounded analysis, evidence-driven reasoning, and a clinician-like diagnostic process through three core innovations: a dual-encoder framework extracting complementary time series and image features, cross-modal alignment for effective multimodal understanding, and knowledge-guided instruction data generation for generating high-granularity grounding data (ECG-Grounding) linking diagnoses to measurable parameters ($e.g.$, QRS/PR Intervals). Additionally, we propose the Grounded ECG Understanding task, a clinically motivated benchmark designed to comprehensively assess the MLLM's capability in grounded ECG understanding. Experimental results on both existing and our proposed benchmarks show GEM significantly improves predictive performance (CSN $7.4\%$ $\uparrow$), explainability ($22.7\%$ $\uparrow$), and grounding ($25.3\%$ $\uparrow$), making it a promising approach for real-world clinical applications. Codes, model, and data are available at https://github.com/lanxiang1017/GEM.

Diffusion-based generative models have exhibited considerable success in conditional video synthesis and editing. Nevertheless, prevailing video diffusion models primarily rely on conditioning with specific input modalities, predominantly text, restricting their adaptability to alternative modalities without necessitating retraining of modality-specific components. In this work, we present EnergyViD, a universal spatio-temporal Energy-guided Video Diffusion model designed for zero-shot video synthesis and editing across diverse conditions. Specifically, we leverage off-the-shelf pre-trained networks to construct generic energy functions, guiding the generation process under specific conditions without the need for retraining. To precisely capture temporal dynamics related to motion conditions (e.g., pose sequences), we introduce a novel kernel Maximum Mean Discrepancy (MMD)-based energy function, which minimizes the global distribution discrepancy between the conditioning input and the generated video. Our extensive qualitative and quantitative experiments demonstrate that our algorithm consistently produces high-quality results across a wide range of motion and non-motion conditions, including text, face ID, style, poses, depths, sketches, canny edges, and segmentation maps, in the context of zero-shot video synthesis and editing. We will release source code upon acceptance of the paper.

*Not all positive pairs are beneficial to time series contrastive learning*. In this paper, we study two types of bad positive pairs that can impair the quality of time series representation learned through contrastive learning: the noisy positive pair and the faulty positive pair. We observe that, with the presence of noisy positive pairs, the model tends to simply learn the pattern of noise (Noisy Alignment). Meanwhile, when faulty positive pairs arise, the model wastes considerable amount of effort aligning non-representative patterns (Faulty Alignment). To address this problem, we propose a Dynamic Bad Pair Mining (DBPM) algorithm, which reliably identifies and suppresses bad positive pairs in time series contrastive learning. Specifically, DBPM utilizes a memory module to dynamically track the training behavior of each positive pair along training process. This allows us to identify potential bad positive pairs at each epoch based on their historical training behaviors. The identified bad pairs are subsequently down-weighted through a transformation module, thereby mitigating their negative impact on the representation learning process. DBPM is a simple algorithm designed as a lightweight **plug-in** without learnable parameters to enhance the performance of existing state-of-the-art methods. Through extensive experiments conducted on four large-scale, real-world time series datasets, we demonstrate DBPM's efficacy in mitigating the adverse effects of bad positive pairs.

GNN-to-MLP distillation aims to utilize knowledge distillation (KD) to learn computationally-efficient multi-layer perceptron (student MLP) on graph data by mimicking the output representations of teacher GNN. Existing methods mainly make the MLP to mimic the GNN predictions over a few class labels. However, the class space may not be expressive enough for covering numerous diverse local graph structures, thus limiting the performance of knowledge transfer from GNN to MLP. To address this issue, we propose to learn a new powerful graph representation space by directly labeling nodes' diverse local structures for GNN-to-MLP distillation. Specifically, we propose a variant of VQ-VAE to learn a structure-aware tokenizer on graph data that can encode each node's local substructure as a discrete code. The discrete codes constitute a codebook as a new graph representation space that is able to identify different local graph structures of nodes with the corresponding code indices. Then, based on the learned codebook, we propose a new distillation target, namely soft code assignments, to directly transfer the structural knowledge of each node from GNN to MLP. The resulting framework VQGraph achieves new state-of-the-art performance on GNN-to-MLP distillation in both transductive and inductive settings across seven graph datasets. We show that VQGraph with better performance infers faster than GNNs by 828×, and also achieves accuracy improvement over GNNs and stand-alone MLPs by 3.90% and 28.05% on average, respectively. Our code is available at https://github.com/YangLing0818/VQGraph

This work summarizes two ways to accomplish Time-Series (TS) tasks in today's Large Language Model (LLM) context: LLM-for-TS (model-centric) designs and trains a fundamental large model, or fine-tunes a pre-trained LLM for TS data; TS-for-LLM (data-centric) converts TS into a model-friendly representation to enable the pre-trained LLM to handle TS data. Given the lack of data, limited resources, semantic context requirements, and so on, this work focuses on TS-for-LLM, where we aim to activate LLM's ability for TS data by designing a TS embedding method suitable for LLM. The proposed method is named TEST. It first tokenizes TS, builds an encoder to embed TS via instance-wise, feature-wise, and text-prototype-aligned contrast, where the TS embedding space is aligned to LLM’s embedding layer space, then creates soft prompts to make LLM more open to that embeddings, and finally implements TS tasks using the frozen LLM. We also demonstrate the feasibility of TS-for-LLM through theory and experiments. Experiments are carried out on TS classification, forecasting, and representation tasks using eight frozen LLMs with various structures and sizes. The results show that the pre-trained LLM with TEST strategy can achieve better or comparable performance than today's SOTA TS models, and offers benefits for few-shot and generalization. By treating LLM as the pattern machine, TEST can endow LLM's ability to process TS data without compromising language ability. We hope that this study will serve as a foundation for future work to support TS+LLM progress.

KDD AIDSH 2024 aims to foster discussions and developments that push the boundaries of Artificial Intelligence (AI) and Data Science (DS) in healthcare, enhance diagnostic accuracy and promote human-centric approaches to healthcare, thus stimulating future interdisciplinary collaborations. This year's symposium will focus on expanding the application of AI/DS in healthcare/medicine and bridging existing gaps. The workshop invites submissions of full papers as well as work-in-progress on the application of AI/DS in healthcare. The workshop will feature three invited talks from eminent speakers, spanning academia, industry, and clinical researchers. In addition, selected papers will be invited to publish in Health Data Science, a Science Partner Journal. This summary provides a brief description of the half-day workshop to be held on August 26th, 2024. The webpage for the workshop can be found at https://aimel.ai/kdd2024aidsh.

While time series diffusion models have received considerable focus from many recent works, the performance of existing models remains highly unstable. Factors limiting time series diffusion models include insufficient time series datasets and the absence of guidance. To address these limitations, we propose a Retrieval-Augmented Time series Diffusion model (RATD). The framework of RATD consists of two parts: an embedding-based retrieval process and a reference-guided diffusion model. In the first part, RATD retrieves the time series that are most relevant to historical time series from the database as references. The references are utilized to guide the denoising process in the second part. Our approach allows leveraging meaningful samples within the database to aid in sampling, thus maximizing the utilization of datasets. Meanwhile, this reference-guided mechanism also compensates for the deficiencies of existing time series diffusion models in terms of guidance. Experiments and visualizations on multiple datasets demonstrate the effectiveness of our approach, particularly in complicated prediction tasks. Our code is available at https://github.com/stanliu96/RATD

Graph generation is a fundamental task in machine learning with broad impacts on numerous real-world applications such as biomedical discovery and social science. Most recently, generative models, especially diffusion models (DMs), have shown great promise in synthesizing realistic graphs. However, existing DMs methods typically conduct diffusion processes directly in complex graph space (i.e., node feature, adjacency matrix, or both), resulting in high modeling complexity and poor multimodal distribution coverage. In this paper, we propose Graphusion, a novel and unified latent-based graph generative framework to address the problems. Specifically, Graphusion is composed of a variational graph autoencoder mapping raw graphs with high-dimensional discrete space to low-dimensional topology-injected latent space, and latent DMs running there, producing a smoother, faster, and more expressive graph generation procedure. Thanks to the latest space modeling, we further develop principled latent self-guidance to sufficiently cover the whole semantical distribution of the unlabeled graph set. Experiments show that our Graphusion framework can consistently outperform previous graph generation baselines on both generic and molecular graph datasets, demonstrating the generality and extensibility along with further analytical justifications.

Out-of-distribution (OOD) graph generalization are critical for many real-world applications. Existing methods neglect to discard spurious or noisy features of inputs, which are irrelevant to the label. Besides, they mainly conduct instance-level class-invariant graph learning and fail to utilize the structural class relationships between graph instances. In this work, we endeavor to address these issues in a unified framework, dubbedIndividual andStructuralGraphInformationBottlenecks (IS-GIB). To remove class spurious feature caused by distribution shifts, we propose Individual Graph Information Bottleneck (I-GIB) which discards irrelevant information by minimizing the mutual information between the input graph and its embeddings. To leverage the structural intra- and inter-domain correlations, we propose Structural Graph Information Bottleneck (S-GIB). Specifically for a batch of graphs with multiple domains, S-GIB first computes the pair-wise input-input, embedding-embedding, and label-label correlations. Then it minimizes the mutual information between input graph and embedding pairs while maximizing the mutual information between embedding and label pairs. The critical insight of S-GIB is to simultaneously discard spurious features and learn invariant features from a high-order perspective by maintaining class relationships under multiple distributional shifts. Notably, we unify the proposed I-GIB and S-GIB to form our complementary framework IS-GIB. Extensive experiments conducted on both node- and graph-level tasks consistently demonstrate the superior generalization ability of IS-GIB. The code is available athttps://github.com/YangLing0818/GraphOOD.

Early classification tasks aim to classify time series before observing full data. It is critical in time-sensitive applications such as early sepsis diagnosis in the intensive care unit (ICU). Early diagnosis can provide more opportunities for doctors to rescue lives. However, there are two conflicting goals in the early classification task-accuracy and earliness. Most existing methods try to find a balance between them by weighing one goal against the other. But we argue that a powerful early classifier should always make highly accurate predictions at any moment. The main obstacle is that the key features suitable for classification are not obvious in the early stage, resulting in the excessive overlap of time series distributions in different time stages. The indistinguishable distributions make it difficult for classifiers to recognize. To solve this problem, this article proposes a novel ranking-based cross-entropy (RCE) loss to jointly learn the feature of classes and the order of earliness from time series data. In this way, RCE can help classifier to generate probability distributions of time series in different stages with more distinguishable boundary. Thus, the classification accuracy at each time step is finally improved. Besides, for the applicability of the method, we also accelerate the training process by focusing the learning process on high-ranking samples. Experiments on three real-world datasets show that our method can perform classification more accurately than all baselines at all moments.

Since labeled samples are typically scarce in real-world scenarios, self-supervised representation learning in time series is critical. Existing approaches mainly employ the contrastive learning framework, which automatically learns to understand similar and dissimilar data pairs. However, they are constrained by the request for cumbersome sampling policies and prior knowledge of constructing pairs. Also, few works have focused on effectively modeling temporal-spectral correlations to improve the capacity of representations. In this article, we propose the cross reconstruction transformer (CRT) to solve the aforementioned issues. CRT achieves time series representation learning through a cross-domain dropping-reconstruction task. Specifically, we obtain the frequency domain of the time series via the fast Fourier transform (FFT) and randomly drop certain patches in both time and frequency domains. Dropping is employed to maximally preserve the global context while masking leads to the distribution shift. Then a Transformer architecture is utilized to adequately discover the cross-domain correlations between temporal and spectral information through reconstructing data in both domains, which is called Dropped Temporal-Spectral Modeling. To discriminate the representations in global latent space, we propose instance discrimination constraint (IDC) to reduce the mutual information between different time series samples and sharpen the decision boundaries. Additionally, a specified curriculum learning (CL) strategy is employed to improve the robustness during the pretraining phase, which progressively increases the dropping ratio in the training process. We conduct extensive experiments to evaluate the effectiveness of the proposed method on multiple real-world datasets. Results show that CRT consistently achieves the best performance over existing methods by 2%-9%. The code is publicly available at https://github.com/BobZwr/Cross-Reconstruction-Transformer.

Diffusion models are a new class of generative models, and have dramatically promoted image generation with unprecedented quality and diversity. Existing diffusion models mainly try to reconstruct input image from a corrupted one with a pixel-wise or feature-wise constraint along spatial axes. However, such point-based reconstruction may fail to make each predicted pixel/feature fully preserve its neighborhood context, impairing diffusion-based image synthesis. As a powerful source of automatic supervisory signal, context has been well studied for learning representations. Inspired by this, we for the first time propose ConPreDiff to improve diffusion-based image synthesis with context prediction. We explicitly reinforce each point to predict its neighborhood context (i.e., multi-stride pixels/features) with a context decoder at the end of diffusion denoising blocks in training stage, and remove the decoder for inference. In this way, each point can better reconstruct itself by preserving its semantic connections with neighborhood context. This new paradigm of ConPreDiff can generalize to arbitrary discrete and continuous diffusion backbones without introducing extra parameters in sampling procedure. Extensive experiments are conducted on unconditional image generation, text-to-image generation and image inpainting tasks. Our ConPreDiff consistently outperforms previous methods and achieves new SOTA text-to-image generation results on MS-COCO, with a zero-shot FID score of 6.21.

Learning information-rich and generalizable representations effectively from unlabeled multivariate cardiac signals to identify abnormal heart rhythms (cardiac arrhythmias) is valuable in real-world clinical settings but often challenging due to its complex temporal dynamics. Cardiac arrhythmias can vary significantly in temporal patterns even for the same patient (i.e., intra subject difference). Meanwhile, the same type of cardiac arrhythmia can show different temporal patterns among different patients due to different cardiac structures (i.e., inter subject difference). In this paper, we address the challenges by proposing an Intra-Inter Subject Self-Supervised Learning (ISL) model that is customized for multivariate cardiac signals. Our proposed ISL model integrates medical knowledge into self-supervision to effectively learn from intra-inter subject differences. In intra subject self-supervision, ISL model first extracts heartbeat-level features from each subject using a channel-wise attentional CNN-RNN encoder. Then a stationarity test module is employed to capture the temporal dependencies between heartbeats. In inter subject self-supervision, we design a set of data augmentations according to the clinical characteristics of cardiac signals and perform contrastive learning among subjects to learn distinctive representations for various types of patients. Extensive experiments on three real-world datasets were conducted. In a semi-supervised transfer learning scenario, our pre-trained ISL model leads about 10% improvement over supervised training when only 1% labeled data is available, suggesting strong generalizability and robustness of the model.

Length of Stay (LoS) estimation is important for efficient healthcare resource management. Since the distribution of LoS is highly skewed, some previous works frame the LoS estimation as a multi-class classification problem by dividing the range of LoS into buckets. However, they ignore the ordinal relationship between labels. The distribution of bucketed LoS, with a heavy head and a heavy tail, is still imbalanced since the long tail is grouped into the last bucket. This paper proposes a Deep Ordinal neural network for Length of stay Estimation in the intensive care units (DOSE). DOSE can exploit the ordinal relationship and mitigate the skewness. The ordinal classification problem is decomposed into a series of binary classification sub-problems by using multiple binary classifiers. To maintain consistency among binary classifiers, the monotonicity constraint penalty is proposed. The number of samples whose labels are higher or lower than a given threshold is at the same level due to the heavy head and tail of the distribution. Therefore, the training data of each binary classifier are balanced. Experiments are conducted on the real-world healthcare dataset. DOSE outperforms all baseline methods in all metrics. The distribution of the prediction of DOSE is more aligned with the ground truth.

In the real world, the class of a time series is usually labeled at the final time, but many applications require to classify time series at every time point. e.g. the outcome of a critical patient is only determined at the end, but he should be diagnosed at all times for timely treatment. Thus, we propose a new concept: Continuous Classification of Time Series (CCTS). It requires the model to learn data in different time stages. But the time series evolves dynamically, leading to different data distributions. When a model learns multi-distribution, it always forgets or overfits. We suggest that meaningful learning scheduling is potential due to an interesting observation: Measured by confidence, the process of model learning multiple distributions is similar to the process of human learning multiple knowledge. Thus, we propose a novel Confidence-guided method for CCTS (C3TS). It can imitate the alternating human confidence described by the Dunning-Kruger Effect. We define the objective-confidence to arrange data, and the self-confidence to control the learning duration. Experiments on four real-world datasets show that C3TS is more accurate than all baselines for CCTS.

Unsupervised/self-supervised graph representation learning is critical for downstream node- and graph-level classification tasks. Global structure of graphs helps discriminating representations and existing methods mainly utilize the global structure by imposing additional supervisions. However, their global semantics are usually invariant for all nodes/graphs and they fail to explicitly embed the global semantics to enrich the representations. In this paper, we propose Omni-Granular Ego-Semantic Propagation for Self-Supervised Graph Representation Learning (OEPG). Specifically, we introduce instance-adaptive global-aware ego-semantic descriptors, leveraging the first- and second-order feature differences between each node/graph and hierarchical global clusters of the entire graph dataset. The descriptors can be explicitly integrated into local graph convolution as new neighbor nodes. Besides, we design an omni-granular normalization on the whole scales and hierarchies of the ego-semantic to assign attentional weight to each descriptor from an omni-granular perspective. Specialized pretext tasks and cross-iteration momentum update are further developed for local-global mutual adaptation. In downstream tasks, OEPG consistently achieves the best performance with a 2%~6% accuracy gain on multiple datasets cross scales and domains. Notably, OEPG also generalizes to quantity- and topology-imbalance scenarios.

Unsupervised/self-supervised time series representation learning is a challenging problem because of its complex dynamics and sparse annotations. Existing works mainly adopt the framework of contrastive learning with the time-based augmentation techniques to sample positives and negatives for contrastive training. Nevertheless, they mostly use segment-level augmentation derived from time slicing, which may bring about sampling bias and incorrect optimization with false negatives due to the loss of global context. Besides, they all pay no attention to incorporate the spectral information in feature representation. In this paper, we propose a unified framework, namely Bilinear Temporal-Spectral Fusion (BTSF). Specifically, we firstly utilize the instance-level augmentation with a simple dropout on the entire time series for maximally capturing long-term dependencies. We devise a novel iterative bilinear temporal-spectral fusion to explicitly encode the affinities of abundant time-frequency pairs, and iteratively refines representations in a fusion-and-squeeze manner with Spectrum-to-Time (S2T) and Time-to-Spectrum (T2S) Aggregation modules. We firstly conducts downstream evaluations on three major tasks for time series including classification, forecasting and anomaly detection. Experimental results shows that our BTSF consistently significantly outperforms the state-of-the-art methods.

Hypergraphs are natural and expressive modeling tools to encode high-order relationships among entities. Several variations of Hypergraph Neural Networks (HGNNs) are proposed to learn the node representations and complex relationships in the hypergraphs. Most current approaches assume that the input hypergraph structure accurately depicts the relations in the hypergraphs. However, the input hypergraph structure inevitably contains noise, task-irrelevant information, or false-negative connections. Treating the input hypergraph structure as ground-truth information unavoidably leads to sub-optimal performance. In this paper, we propose a Hypergraph Structure Learning (HSL) framework, which optimizes the hypergraph structure and the HGNNs simultaneously in an end-to-end way. HSL learns an informative and concise hypergraph structure that is optimized for downstream tasks. To efficiently learn the hypergraph structure, HSL adopts a two-stage sampling process: hyperedge sampling for pruning redundant hyperedges and incident node sampling for pruning irrelevant incident nodes and discovering potential implicit connections. The consistency between the optimized structure and the original structure is maintained by the intra-hyperedge contrastive learning module. The sampling processes are jointly optimized with HGNNs towards the objective of the downstream tasks. Experiments conducted on 7 datasets show shat HSL outperforms the state-of-the-art baselines while adaptively sparsifying hypergraph structures.

Electronic Health Records (EHR) is the repository of patients' involved medical codes in the hospital, including diagnosis codes, medication codes, procedure codes, lab codes, and so on. EHR inherently contains various kinds of relationships such as the code-code, the patient-patient, and the patient-code relationship. Recent research shows that graph representation learning can be an effective tool for capturing complex relationships. However, none of the existing methods considered high-order interactions between patients and medical codes or considered the three relationships together. In this paper, we propose Hypergraph Contrastive Learning (HCL), to jointly learn patient embeddings and code embeddings from the combination of the above three relationships. HCL first constructs a hypergraph from the EHR data. Then, the medical code graph and the patient graph are constructed based on the hypergraph. Empowered with hypergraph attention network, Transformer, and graph attention network, HCL learns representations from three graphs respectively. Next, contrastive learning is applied to aggregate information from these graphs. Finally, the learned representations can support downstream tasks in supervised learning settings and self-supervised learning settings. Experiments are conducted on eICU and MIMIC-III datasets with mortality prediction and readmission prediction tasks. Results show that our method outperforms almost all compared methods on all evaluation metrics and HCL can learn patient representations from medical codes even without labeled data.

Since data is presented long-tailed in reality, it is challenging for Federated Learning (FL) to train across decentralized clients as practical applications. We present Global-Regularized Personalization (GRP-FED) to tackle the data imbalanced issue by considering a single global model and multiple local models for each client. With adaptive aggregation, the global model treats multiple clients fairly and mitigates the global long-tailed issue. Each local model is learned from the local data and aligns with its distribution for customization. To prevent the local model from just overfitting, GRP-FED applies an adversarial discriminator to regularize between the learned global-local features. Extensive results show that our GRP-FED improves under both global and local scenarios on real-world MIT-BIH and synthesis CIFAR-10 datasets, achieving comparable performance and addressing client imbalance.

Time-series forecasting is one of the most active research topics in artificial intelligence. It has the power to bring light to problems in several areas of knowledge, such as epidemiological studies, healthcare inference, and climate change analysis. Applications in real-world time series should consider two factors for achieving reliable predictions: modeling dynamic dependencies among multiple variables and adjusting the model's intrinsic hyperparameters. An open gap in the literature is that statistical and ensemble learning approaches systematically present lower predictive performance than deep learning methods. The existing applications consistently disregard the data sequence aspect entangled with multivariate data represented in more than one time series. Conversely, this work presents a novel neural network architecture for time-series forecasting that combines the power of graph evolution with deep recurrent learning on distinct data distributions, named after Recurrent Graph Evolution Neural Network ( ReGENN ). The idea is to infer multiple multivariate relationships between co-occurring time-series by assuming that the temporal data depends not only on inner variables and intra-temporal relationships (i.e., observations from itself) but also on outer variables and inter-temporal relationships (i.e., observations from other-selves). An extensive set of experiments was conducted comparing ReGENN with tens of ensemble methods and classical statistical ones. The results outperformed both statistical and ensemble-learning approaches, showing an improvement of 64.87 percent over the competing algorithms on the SARS-CoV-2 dataset of the renowned John Hopkins University for 188 countries simultaneously. For further validation, we tested our architecture in two other public datasets of different domains, the PhysioNet Computing in Cardiology Challenge 2012 and Brazilian Weather datasets. We also analyzed the Evolution Weights arising from the hidden layers of ReGENN to describe how the variables of the dataset interact with each other; and, as a result of looking at inter and intra-temporal relationships simultaneously, we concluded that time-series forecasting is majorly improved if paying attention to how multiple multivariate data synchronously evolve.

There is a growing interest in applying deep learning (DL) to healthcare, driven by the availability of data with multiple feature channels inrich-dataenvironments (e.g., intensive care units). However, in many other practical situations, we can only access data with much fewer feature channels in apoor-dataenvironments (e.g., at home), which often results in predictive models with poor performance. How can we boost the performance of models learned from suchpoor-dataenvironment by leveraging knowledge extracted from existing models trained usingrich datain a related environment? To address this question, we develop a knowledge infusion framework namedCHEERthat can succinctly summarize suchrich modelinto transferable representations, which can be incorporated into thepoor modelto improve its performance. The infused model is analyzed theoretically and evaluated empirically on several datasets. Our empirical results showed thatCHEERoutperformed baselines by 5.60 to 46.80 percent in terms of the macro-F1 score on multiple physiological datasets.

Prediction based on Irregularly Sampled Time Series (ISTS) is of wide concern in real-world applications. For more accurate prediction, methods had better grasp more data characteristics. Different from ordinary time series, ISTS is characterized by irregular time intervals of intra-series and different sampling rates of inter-series. However, existing methods have suboptimal predictions due to artificially introducing new dependencies in a time series and biasedly learning relations among time series when modeling these two characteristics. In this work, we propose a novel Time Encoding (TE) mechanism. TE can embed the time information as time vectors in the complex domain. It has the properties of absolute distance and relative distance under different sampling rates, which helps to represent two irregularities. Meanwhile, we create a new model named Time Encoding Echo State Network (TE-ESN). It is the first ESNs-based model that can process ISTS data. Besides, TE-ESN incorporates long short-term memories and series fusion to grasp horizontal and vertical relations. Experiments on one chaos system and three real-world datasets show that TE-ESN performs better than all baselines and has better reservoir property.

This paper presents an active distillation method for a local institution (e.g., hospital) to find the best queries within its given budget to distill an on-server black-box model’s predictive knowledge into a local surrogate with transparent parameterization. This allows local institutions to understand better the predictive reasoning of the black-box model in its own local context or to further customize the distilled knowledge with its private dataset that cannot be centralized and fed into the server model. The proposed method thus addresses several challenges of deploying machine learning (ML) in many industrial settings (e.g., healthcare analytics) with strong proprietary constraints. These include: (1) the opaqueness of the server model’s architecture which prevents local users from understanding its predictive reasoning in their local data contexts; (2) the increasing cost and risk of uploading local data on the cloud for analysis; and (3) the need to customize the server model with private onsite data. We evaluated the proposed method on both benchmark and real-world healthcare data where significant improvements over existing local distillation methods were observed. A theoretical analysis of the proposed method is also presented.

Visualizing latent embeddings is a popular approach to explain classification models, including deep neural networks. However, existing visualization methods such as t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation Projection (UMAP) are often used as a post-processing step which is independent of the classification models. The resulting visualization can be misaligned with the classification models. In this paper, we propose ViVA, a novel method for semi-supervised Visualization via Variational Autoencoders. ViVA learns from both unlabeled and labeled data by jointly optimizing both visualization loss and classification loss. As a parameterized model using neural networks, ViVA can easily project new data to the same embedding space. Experiments show that ViVA can achieve better visualization quality as well as classification accuracy on multiple challenging datasets compared to several visualization baselines, including t-SNE and UMAP.

Deep learning models have achieved expert-level performance in healthcare with an exclusive focus on training accurate models. However, in many clinical environments such as intensive care unit (ICU), real-time model serving is equally if not more important than accuracy, because in ICU patient care is simultaneously more urgent and more expensive. Clinical decisions and their timeliness, therefore, directly affect both the patient outcome and the cost of care. To make timely decisions, we argue the underlying serving system must be latency-aware. To compound the challenge, health analytic applications often require a combination of models instead of a single model, to better specialize individual models for different targets, multi-modal data, different prediction windows, and potentially personalized predictions. To address these challenges, we propose HOLMES---an online model ensemble serving framework for healthcare applications. HOLMES dynamically identifies the best performing set of models to ensemble for highest accuracy, while also satisfying sub-second latency constraints on end-to-end prediction. We demonstrate that HOLMES is able to navigate the accuracy/latency tradeoff efficiently, compose the ensemble, and serve the model ensemble pipeline, scaling to simultaneously streaming data from 100 patients, each producing waveform data at 250~Hz. HOLMES outperforms the conventional offline batch-processed inference for the same clinical task in terms of accuracy and latency (by order of magnitude). HOLMES is tested on risk prediction task on pediatric cardio ICU data with above 95% prediction accuracy and sub-second latency on 64-bed simulation.

In many situations, we need to build and deploy separate models in related environments with different data qualities. For example, an environment with strong observation equipments (e.g., intensive care units) often provides high-quality multi-modal data, which are acquired from multiple sensory devices and have rich-feature representations. On the other hand, an environment with poor observation equipment (e.g., at home) only provides low-quality, uni-modal data with poor-feature representations. To deploy a competitive model in a poor-data environment without requiring direct access to multi-modal data acquired from a rich-data environment, this paper develops and presents a knowledge distillation (KD) method (RDPD) to enhance a predictive model trained on poor data using knowledge distilled from a high-complexity model trained on rich, private data. We evaluated RDPD on three real-world datasets and shown that its distilled model consistently outperformed all baselines across all datasets, especially achieving the greatest performance improvement over a model trained only on low-quality data by 24.56% on PR-AUC and 12.21% on ROC-AUC, and over that of a state-of-the-art KD model by 5.91% on PR-AUC and 4.44% on ROC-AUC.

Electrocardiography (ECG) signals are commonly used to diagnose various cardiac abnormalities. Recently, deep learning models showed initial success on modeling ECG data, however they are mostly black-box, thus lack interpretability needed for clinical usage. In this work, we propose MultIlevel kNowledge-guided Attention networks (MINA) that predict heart diseases from ECG signals with intuitive explanation aligned with medical knowledge. By extracting multilevel (beat-, rhythm- and frequency-level) domain knowledge features separately, MINA combines the medical knowledge and ECG data via a multilevel attention model, making the learned models highly interpretable. Our experiments showed MINA achieved PR-AUC 0.9436 (outperforming the best baseline by 5.51%) in real world ECG dataset. Finally, MINA also demonstrated robust performance and strong interpretability against signal distortion and noise contamination.

Atrial Fibrillation (AF) is an abnormal heart rhythm which can trigger cardiac arrest and sudden death. Nevertheless, its interpretation is mostly done by medical experts due to high error rates of computerized interpretation. One study found that only about 66% of AF were correctly recognized from noisy ECGs. This is in part due to insufficient training data, class skewness, as well as semantical ambiguities caused by noisy segments in an ECG record. In this paper, we propose a K-margin-based Residual-Convolution-Recurrent neural network (K-margin-based RCR-net) for AF detection from noisy ECGs. In detail, a skewness-driven dynamic augmentation method is employed to handle the problems of data inadequacy and class imbalance. A novel RCR-net is proposed to automatically extract both long-term rhythm-level and local heartbeat-level characters. Finally, we present a K-margin-based diagnosis model to automatically focus on the most important parts of an ECG record and handle noise by naturally exploiting expected consistency among the segments associated for each record. The experimental results demonstrate that the proposed method with 0.8125 F1NAOP score outperforms all state-of-the-art deep learning methods for AF detection task by 6.8%.

Predicting medicines for patients with co-morbidity has long been recognized as a hard task due to complex dependencies between diseases and medicines. Efforts have been made recently to build high-order dependency between diseases and medicines by extracting knowledge from electronic health records (EHR). But current works failed to utilize additional knowledge and ignored the data skewness problem which lead to sub-optimal combination of medicines. In this paper, we formulate the medicines prediction task in multi-instance multi-label learning framework considering the multi-diagnoses as input instances and multi-medicines as output labels. We propose a knowledge-guided multi-instance multi-label networks called \mname where two types of additional knowledge are incorporated into a RNN encoder-decoder model. The utilization of structural knowledge like clinical ontology provides a way to learn better representation called tree embedding by utilizing the ancestors’ information. Contextual knowledge is a global summarization of input instances which is informative for personal prediction. Experiments are conducted on a real world clinical dataset which showed the necessity to combine both contextual and structural knowledge and the \mname performs better than baselines up to 4+% in terms of Jaccard similarity score.