“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

When building a unified vision system or gradually adding new capabilities to a system, the usual assumption is that training data for all tasks is always available. However, as the number of tasks grows, storing and retraining on such data becomes infeasible. A new problem arises where we add new capabilities to a Convolutional Neural Network (CNN), but the training data for its existing capabilities are unavailable. We propose our Learning without Forgetting method, which uses only new task data to train the network while preserving the original capabilities. Our method performs favorably compared to commonly used feature extraction and fine-tuning adaption techniques and performs similarly to multitask learning that uses original task data we assume unavailable. A more surprising observation is that Learning without Forgetting may be able to replace fine-tuning with similar old and new task datasets for improved new task performance.

Learning without Forgetting

The combined impact of new computing resources and techniques with an increasing avalanche of large datasets, is transforming many research areas and may lead to technological breakthroughs that can be used by billions of people. In the recent years, Machine Learning and especially its subfield Deep Learning have seen impressive advances. Techniques developed within these two fields are now able to analyze and learn from huge amounts of real world examples in a disparate formats. While the number of Machine Learning algorithms is extensive and growing, their implementations through frameworks and libraries is also extensive and growing too. The software development in this field is fast paced with a large number of open-source software coming from the academy, industry, start-ups or wider open-source communities. This survey presents a recent time-slide comprehensive overview with comparisons as well as trends in development and usage of cutting-edge Artificial Intelligence software. It also provides an overview of massive parallelism support that is capable of scaling computation effectively and efficiently in the era of Big Data.

https://link.springer.com/content/pdf/10.1007%2Fs10462-018-09679-z.pdf

Machine Learning and Deep Learning frameworks and libraries for large-scale data mining: a survey

Blood pressure (BP) is a direct indicator of hypertension, a dangerous and potentially deadly condition. Regular monitoring of BP is thus important, but many people have aversion towards cuff-based devices, and their limitation is that they can only be used at rest. Using just a photoplethysmogram (PPG) to estimate BP is a potential solution investigated in our study. We analyzed the MIMIC III database for high-quality PPG and arterial BP waveforms, resulting in over 700 h of signals after preprocessing, belonging to 510 subjects. We then used the PPG alongside its first and second derivative as inputs into a novel spectro-temporal deep neural network with residual connections. We have shown in a leave-one-subject-out experiment that the network is able to model the dependency between PPG and BP, achieving mean absolute errors of 9.43 for systolic and 6.88 for diastolic BP. Additionally we have shown that personalization of models is important and substantially improves the results, while deriving a good general predictive model is difficult. We have made crucial parts of our study, especially the list of used subjects and our neural network code, publicly available, in an effort to provide a solid baseline and simplify potential comparison between future studies on an explicit MIMIC III subset.

https://www.mdpi.com/1424-8220/19/15/3420/pdf

Blood Pressure Estimation from Photoplethysmogram Using a Spectro-Temporal Deep Neural Network.

Coronavirus disease 2019 (COVID-19) has emerged as a pandemic with serious clinical manifestations including death. A pandemic at the large-scale like COVID-19 places extraordinary demands on the world's health systems, dramatically devastates vulnerable populations, and critically threatens the global communities in an unprecedented way. While tremendous efforts at the frontline are placed on detecting the virus, providing treatments and developing vaccines, it is also critically important to examine the technologies and systems for tackling disease emergence, arresting its spread and especially the strategy for diseases prevention. The objective of this article is to review enabling technologies and systems with various application scenarios for handling the COVID-19 crisis. The article will focus specifically on 1) wearable devices suitable for monitoring the populations at risk and those in quarantine, both for evaluating the health status of caregivers and management personnel, and for facilitating triage processes for admission to hospitals; 2) unobtrusive sensing systems for detecting the disease and for monitoring patients with relatively mild symptoms whose clinical situation could suddenly worsen in improvised hospitals; and 3) telehealth technologies for the remote monitoring and diagnosis of COVID-19 and related diseases. Finally, further challenges and opportunities for future directions of development are highlighted.

/pdf/wearable-sensing-and-telehealth-technology-with-potential-2p6h85sx22.pdf

Wearable Sensing and Telehealth Technology with Potential Applications in the Coronavirus Pandemic

Existing methods for arterial blood pressure (BP) estimation directly map the input physiological signals to output BP values without explicitly modeling the underlying temporal dependencies in BP dynamics. As a result, these models suffer from accuracy decay over a long time and thus require frequent calibration. In this work, we address this issue by formulating BP estimation as a sequence prediction problem in which both the input and target are temporal sequences. We propose a novel deep recurrent neural network (RNN) consisting of multilayered Long Short-Term Memory (LSTM) networks, which are incorporated with (1) a bidirectional structure to access larger-scale context information of input sequence, and (2) residual connections to allow gradients in deep RNN to propagate more effectively. The proposed deep RNN model was tested on a static BP dataset, and it achieved root mean square error (RMSE) of 3.90 and 2.66 mmHg for systolic BP (SBP) and diastolic BP (DBP) prediction respectively, surpassing the accuracy of traditional BP prediction models. On a multi-day BP dataset, the deep RNN achieved RMSE of 3.84, 5.25, 5.80 and 5.81 mmHg for the 1st day, 2nd day, 4th day and 6th month after the 1st day SBP prediction, and 1.80, 4.78, 5.0, 5.21 mmHg for corresponding DBP prediction, respectively, which outperforms all previous models with notable improvement. The experimental results suggest that modeling the temporal dependencies in BP dynamics significantly improves the long-term BP prediction accuracy.

Long-term blood pressure prediction with deep recurrent neural networks

Objective: To fight the “silent killer” hypertension, continuous blood pressure (BP) monitoring has been one of the most desired functions in wearable electronics. However, current BP measuring principles and protocols either involve a vessel occlusion process with a cuff or require multiple sensing nodes on the body, which makes it difficult to implement them in compact wearable electronics like smartwatches and wristbands with long-term wearability. Methods: In this work, we proposed a highly compact multi-wavelength photoplethysmography (MWPPG) module and a depth-resolved MWPPG approach for continuous monitoring of BP and systemic vascular resistance (SVR). By associating the wavelength-dependent light penetration depth in the skin with skin vasculatures, our method exploited the pulse transit time (PTT) on skin arterioles for tracking SVR ( n = 20). Then, we developed an arteriolar PTT-based method for beat-to-beat BP measurement. The BP estimation accuracy of the proposed arteriolar PTT method was validated against Finometer ( n = 20) and the arterial line ( n = 4). Results: The correlation between arteriolar PTT and SVR was theoretically deduced and experimentally validated on 20 human subjects performing various maneuvers. The proposed arteriolar PTT-based method outperformed the traditional arterial PTT-based method with better BP estimation accuracy and simpler measurement setup, i.e., with a single sensing node. Conclusion: The proposed depth-resolved MWPPG method can provide accurate measurements of SVR and BP, which are traditionally difficult to measure in a noninvasive or continuous fashion. Significance: This MWPPG work provides the wearable healthcare electronics of compact size with a low-cost and physiology-based solution for continuous measurement of BP and SVR.

Multi-Wavelength Photoplethysmography Enabling Continuous Blood Pressure Measurement With Compact Wearable Electronics

Real-world object detectors are often challenged by the domain gaps between different datasets. In this work, we present the Conditional Domain Normalization (CDN) to bridge the domain gap. CDN is designed to encode different domain inputs into a shared latent space, where the features from different domains carry the same domain attribute. To achieve this, we first disentangle the domain-specific attribute out of the semantic features from one domain via a domain embedding module, which learns a domain-vector to characterize the corresponding domain attribute information. Then this domain-vector is used to encode the features from another domain through a conditional normalization, resulting in different domains' features carrying the same domain attribute. We incorporate CDN into various convolution stages of an object detector to adaptively address the domain shifts of different level's representation. In contrast to existing adaptation works that conduct domain confusion learning on semantic features to remove domain-specific factors, CDN aligns different domain distributions by modulating the semantic features of one domain conditioned on the learned domain-vector of another domain. Extensive experiments show that CDN outperforms existing methods remarkably on both real-to-real and synthetic-to-real adaptation benchmarks, including 2D image detection and 3D point cloud detection.

Adapting Object Detectors with Conditional Domain Normalization

Human beings can quickly adapt to environmental changes by leveraging learning experience. However, the poor ability of adapting to dynamic environments remains a major challenge for AI models. To better understand this issue, we study the problem of continual domain adaptation, where the model is presented with a labeled source domain and a sequence of unlabeled target domains. There are two major obstacles in this problem: domain shifts and catastrophic forgetting. In this work, we propose Gradient Regularized Contrastive Learning to solve the above obstacles. At the core of our method, gradient regularization plays two key roles: (1) enforces the gradient of contrastive loss not to increase the supervised training loss on the source domain, which maintains the discriminative power of learned features; (2) regularizes the gradient update on the new domain not to increase the classification loss on the old target domains, which enables the model to adapt to an in-coming target domain while preserving the performance of previously observed domains. Hence our method can jointly learn both semantically discriminative and domain-invariant features with labeled source domain and unlabeled target domains. The experiments on Digits, DomainNet and Office-Caltech benchmarks demonstrate the strong performance of our approach when compared to the state-of-the-art.

Peng Su

Papers

Long-term blood pressure prediction with deep recurrent neural networks

Multi-Wavelength Photoplethysmography Enabling Continuous Blood Pressure Measurement With Compact Wearable Electronics

Adapting Object Detectors with Conditional Domain Normalization

Gradient Regularized Contrastive Learning for Continual Domain Adaptation

Adapting Object Detectors with Conditional Domain Normalization