Self-Supervised contrastive learning has emerged as a powerful method for obtaining high-quality representations from unlabeled data. However, feature suppression has recently been identified in standard contrastive learning ($e.g.$, SimCLR, CLIP): in a single end-to-end training stage, the contrastive model captures only parts of the shared information across contrasting views, while ignore the other potentially useful information. With feature suppression, contrastive models often fail to learn sufficient representations capable for various downstream tasks. To mitigate the feature suppression problem and ensure the contrastive model to learn comprehensive representations, we develop a novel Multistage Contrastive Learning (MCL) framework. Unlike standard contrastive learning that often result in feature suppression, MCL progressively learn new features that have not been explored in the previous stage, while maintaining the well-learned features. Extensive experiments conducted on various publicly available benchmarks validate the effectiveness of our proposed framework. In addition, we demonstrate that the proposed MCL can be adapted to a variety of popular contrastive learning backbones and boost their performance by learning features that could not be gained from standard contrastive learning procedures.
This paper summarizes our team's efforts in both tracks of the ICMC-ASR Challenge for in-car multi-channel automatic speech recognition. Our submitted systems for ICMC-ASR Challenge include the multi-channel front-end enhancement and diarization, training data augmentation, speech recognition modeling with multi-channel branches. Tested on the offical Eval1 and Eval2 set, our best system achieves a relative 34.3% improvement in CER and 56.5% improvement in cpCER, compared to the offical baseline system.
Self-supervised pre-trained speech models were shown effective for various downstream speech processing tasks. Since they are mainly pre-trained to map input speech to pseudo-labels, the resulting representations are only effective for the type of pre-train data used, either clean or mixture speech. With the idea of selective auditory attention, we propose a novel pre-training solution called Selective-HuBERT, or SHuBERT, which learns the selective extraction of target speech representations from either clean or mixture speech. Specifically, SHuBERT is trained to predict pseudo labels of a target speaker, conditioned on an enrolled speech from the target speaker. By doing so, SHuBERT is expected to selectively attend to the target speaker in a complex acoustic environment, thus benefiting various downstream tasks. We further introduce a dual-path training strategy and use the cross-correlation constraint between the two branches to encourage the model to generate noise-invariant representation. Experiments on SUPERB benchmark and LibriMix dataset demonstrate the universality and noise-robustness of SHuBERT. Furthermore, we find that our high-quality representation can be easily integrated with conventional supervised learning methods to achieve significant performance, even under extremely low-resource labeled data.
The utilization of large language models (LLMs) in the Healthcare domain has generated both excitement and concern due to their ability to effectively respond to freetext queries with certain professional knowledge. This survey outlines the capabilities of the currently developed LLMs for Healthcare and explicates their development process, with the aim of providing an overview of the development roadmap from traditional Pretrained Language Models (PLMs) to LLMs. Specifically, we first explore the potential of LLMs to enhance the efficiency and effectiveness of various Healthcare applications highlighting both the strengths and limitations. Secondly, we conduct a comparison between the previous PLMs and the latest LLMs, as well as comparing various LLMs with each other. Then we summarize related Healthcare training data, training methods, optimization strategies, and usage. Finally, the unique concerns associated with deploying LLMs in Healthcare settings are investigated, particularly regarding fairness, accountability, transparency and ethics. Our survey provide a comprehensive investigation from perspectives of both computer science and Healthcare specialty. Besides the discussion about Healthcare concerns, we supports the computer science community by compiling a collection of open source resources, such as accessible datasets, the latest methodologies, code implementations, and evaluation benchmarks in the Github. Summarily, we contend that a significant paradigm shift is underway, transitioning from PLMs to LLMs. This shift encompasses a move from discriminative AI approaches to generative AI approaches, as well as a shift from model-centered methodologies to datacentered methodologies.
The comprehensive integration of machine learning healthcare models within clinical practice remains suboptimal, notwithstanding the proliferation of high-performing solutions reported in the literature. A predominant factor hindering widespread adoption pertains to an insufficiency of evidence affirming the reliability of the aforementioned models. Recently, uncertainty quantification methods have been proposed as a potential solution to quantify the reliability of machine learning models and thus increase the interpretability and acceptability of the result. In this review, we offer a comprehensive overview of prevailing methods proposed to quantify uncertainty inherent in machine learning models developed for various medical image tasks. Contrary to earlier reviews that exclusively focused on probabilistic methods, this review also explores non-probabilistic approaches, thereby furnishing a more holistic survey of research pertaining to uncertainty quantification for machine learning models. Analysis of medical images with the summary and discussion on medical applications and the corresponding uncertainty evaluation protocols are presented, which focus on the specific challenges of uncertainty in medical image analysis. We also highlight some potential future research work at the end. Generally, this review aims to allow researchers from both clinical and technical backgrounds to gain a quick and yet in-depth understanding of the research in uncertainty quantification for medical image analysis machine learning models.
In recent years, estimating the duration of medical intervention based on electronic health records (EHRs) has gained significant attention in the filed of clinical decision support. However, current models largely focus on structured data, leaving out information from the unstructured clinical free-text data. To address this, we present a novel language-enhanced transformer-based framework, which projects all relevant clinical data modalities (continuous, categorical, binary, and free-text features) into a harmonized language latent space using a pre-trained sentence encoder with the help of medical prompts. The proposed method enables the integration of information from different modalities within the cell transformer encoder and leads to more accurate duration estimation for medical intervention. Our experimental results on both US-based (length of stay in ICU estimation) and Asian (surgical duration prediction) medical datasets demonstrate the effectiveness of our proposed framework, which outperforms tailored baseline approaches and exhibits robustness to data corruption in EHRs.
Not all positive pairs are beneficial to time series contrastive learning. In this paper, we study two types of bad positive pairs that impair the quality of time series representation learned through contrastive learning ($i.e.$, noisy positive pair and faulty positive pair). We show 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 spends considerable efforts 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. DBPM utilizes a memory module to 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 then down-weighted using a transformation module. Our experimental results show that DBPM effectively mitigates the negative impacts of bad pairs, and can be easily used as a plug-in to boost performance of state-of-the-art methods. Codes will be made publicly available.
The encoder-decoder model is a commonly used Deep Neural Network (DNN) model for medical image segmentation. Conventional encoder-decoder models make pixel-wise predictions focusing heavily on local patterns around the pixel. This makes it challenging to give segmentation that preserves the object's shape and topology, which often requires an understanding of the global context of the object. In this work, we propose a Fourier Coefficient Segmentation Network~(FCSN) -- a novel DNN-based model that segments an object by learning the complex Fourier coefficients of the object's masks. The Fourier coefficients are calculated by integrating over the whole contour. Therefore, for our model to make a precise estimation of the coefficients, the model is motivated to incorporate the global context of the object, leading to a more accurate segmentation of the object's shape. This global context awareness also makes our model robust to unseen local perturbations during inference, such as additive noise or motion blur that are prevalent in medical images. When FCSN is compared with other state-of-the-art models (UNet+, DeepLabV3+, UNETR) on 3 medical image segmentation tasks (ISIC\_2018, RIM\_CUP, RIM\_DISC), FCSN attains significantly lower Hausdorff scores of 19.14 (6\%), 17.42 (6\%), and 9.16 (14\%) on the 3 tasks, respectively. Moreover, FCSN is lightweight by discarding the decoder module, which incurs significant computational overhead. FCSN only requires 22.2M parameters, 82M and 10M fewer parameters than UNETR and DeepLabV3+. FCSN attains inference and training speeds of 1.6ms/img and 6.3ms/img, that is 8$\times$ and 3$\times$ faster than UNet and UNETR.
We suggested a unified system with core components of data augmentation, ImageNet-pretrained ResNet-50, cost-sensitive loss, deep ensemble learning, and uncertainty estimation to quickly and consistently detect COVID-19 using acoustic evidence. To increase the model's capacity to identify a minority class, data augmentation and cost-sensitive loss are incorporated (infected samples). In the COVID-19 detection challenge, ImageNet-pretrained ResNet-50 has been found to be effective. The unified framework also integrates deep ensemble learning and uncertainty estimation to integrate predictions from various base classifiers for generalisation and reliability. We ran a series of tests using the DiCOVA2021 challenge dataset to assess the efficacy of our proposed method, and the results show that our method has an AUC-ROC of 85.43 percent, making it a promising method for COVID-19 detection. The unified framework also demonstrates that audio may be used to quickly diagnose different respiratory disorders.
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.