Episodic Training for Domain Generalization Li, Da; Zhang, Jianshu; Yang, Yongxin ...
2019 IEEE/CVF International Conference on Computer Vision (ICCV),
2019-Oct.
Conference Proceeding
Odprti dostop
Domain generalization (DG) is the challenging and topical problem of learning models that generalize to novel testing domains with different statistics than a set of known training domains. The ...simple approach of aggregating data from all source domains and training a single deep neural network end-to-end on all the data provides a surprisingly strong baseline that surpasses many prior published methods. In this paper we build on this strong baseline by designing an episodic training procedure that trains a single deep network in a way that exposes it to the domain shift that characterises a novel domain at runtime. Specifically, we decompose a deep network into feature extractor and classifier components, and then train each component by simulating it interacting with a partner who is badly tuned for the current domain. This makes both components more robust, ultimately leading to our networks producing state-of-the-art performance on three DG benchmarks. Furthermore, we consider the pervasive workflow of using an ImageNet trained CNN as a fixed feature extractor for downstream recognition tasks. Using the Visual Decathlon benchmark, we demonstrate that our episodic-DG training improves the performance of such a general purpose feature extractor by explicitly training a feature for robustness to novel problems. This shows that DG training can benefit standard practice in computer vision.
Current methods for skeleton-based human action recognition usually work with completely observed skeletons. However, in real scenarios, it is prone to capture incomplete and noisy skeletons, which ...will deteriorate the performance of traditional models. To enhance the robustness of action recognition models to incomplete skeletons, we propose a multi-stream graph convolutional network (GCN) for exploring sufficient discriminative features distributed over all skeleton joints. Here, each stream of the network is only responsible for learning features from currently unactivated joints, which are distinguished by the class activation maps (CAM) obtained by preceding streams, so that the activated joints of the proposed method are obviously more than traditional methods. Thus, the proposed method is termed richly activated GCN (RA-GCN), where the richly discovered features will improve the robustness of the model. Compared to the state-of-the-art methods, the RA-GCN achieves comparable performance on the NTU RGB+D dataset. Moreover, on a synthetic occlusion dataset, the performance deterioration can be alleviated by the RA-GCN significantly.
Display omitted
•An evaporation-induced self-assembly method was used to obtain FePc coated AC.•Loading of FePc on AC influenced the properties of Fe-N/AC.•Fe-N/AC could catalyze ORR via indirect 4e- ...pathway with high efficiency.•MFC using Fe–N/AC as cathode showed comparable performance to that using Pt/C.•Fe–N/AC will be promising alternative cathode catalyst of MFCs in the future.
To improve the power generation of microbial fuel cells (MFCs), iron–nitrogen/activated carbon (Fe–N/AC), an excellent oxygen reduction reaction (ORR) electrocatalyst, was prepared by pyrolyzing iron(II) phthalocyanine (FePc)-coated AC, which was formed via an evaporation-induced self-assembly method. Given the high content of nitrogen and iron elements, crystalline structure, high surface area, and appropriate composition of micropores and mesopores, Fe–N/AC synthesized with 50.0 wt% of FePc loading on AC exhibits enhanced ORR electrocatalytic activity. The Fe–N/AC can catalyze ORR via an indirect four-electron pathway in neutral medium with onset potential of 0.883 V versus reversible hydrogen electrode and electron transfer number of 3.91. In comparison with AC, the charge transfer resistance and exchange current density of Fe–N/AC decreases by 62% and increases by a factor of three, respectively. The maximum power density of MFC using Fe–N/AC reaches 1092 mW m−2, which is 63.23% higher than that of AC and equal to that of Pt/C. This study proposes a new approach for the design of alternative non-precious metal ORR electrocatalysts in neutral pH medium, which may have potential application in practical MFCs as cathode catalysts in the future.
Among the main bacteria implicated in the pathology of periodontal disease, Aggregatibacter actinomycetemcomitans (Aa) is well known for causing loss of periodontal attachment and systemic disease. ...Recent studies have suggested that secreted extracellular RNAs (exRNAs) from several bacteria may be important in periodontitis, although their role is unclear. Emerging evidence indicates that exRNAs circulate in nanosized bilayered and membranous extracellular vesicles (EVs) known as outer membrane vesicles (OMVs) in gram‐negative bacteria. In this study, we analyzed the small RNA expression profiles in activated human macrophage‐like cells (U937) infected with OMVs from Aa and investigated whether these cells can harbor exRNAs of bacterial origin that have been loaded into the host RNA‐induced silencing complex, thus regulating host target transcripts. Our results provide evidence for the cytoplasmic delivery and activity of microbial EV‐derived small exRNAs in host gene regulation. The production of TNF‐α was promoted by exRNAs via the TLR‐8 and NF‐κB signaling pathways. Numerous studies have linked periodontal disease to neuroinflammatory diseases but without elucidating specific mechanisms for the connection. We show here that intracardiac injection of Aa OMVs in mice showed successful delivery to the brain after crossing the blood‐brain barrier, the exRNA cargos increasing expression of TNF‐α in the mouse brain. The current study indicates that host gene regulation by microRNAs originating from OMVs of the periodontal pathogen Aa is a novel mechanism for host gene regulation and that the transfer of OMV exRNAs to the brain may cause neuroinflammatory diseases like Alzheimer's.—Han, E.‐C., Choi, S.‐Y., Lee, Y., Park, J.‐W., Hong S.‐H., Lee, H.‐J. Extracellular RNAs in periodontopathogenic outer membrane vesicles promote TNF‐α production in human macrophages and cross the blood‐brain barrier in mice. FASEB J. 33, 13412–13422 (2019). www.fasebj.org
NiFe-based (oxy)hydroxides are the benchmark catalysts for the oxygen evolution reaction (OER) in alkaline medium, however, it is still challenging to control their structures and compositions. ...Herein, molybdates (NiFe(MoO
)
) are applied as unique precursors to synthesize ultrafine Mo modified NiFeO
H
(oxy)hydroxide nanosheet arrays. The electrochemical activation process enables the molybdate ions (MoO
) in the precursors gradually dissolve, and at the same time, hydroxide ions (OH
) in the electrolyte diffuse into the precursor and react with Ni
and Fe
ions in confined space to produce ultrafine NiFeO
H
(oxy)hydroxides nanosheets (<10 nm), which are densely arranged into microporous arrays and maintain the rod-like morphology of the precursor. Such dense ultrafine nanosheet arrays produce rich edge planes on the surface of NiFeO
H
(oxy)hydroxides to expose more active sites. More importantly, the capillary phenomenon of microporous structures and hydrophilic hydroxyl groups induce the superhydrophilicity and the rough surface produces the superaerophobic characteristic for bubbles. With these advantages, the optimized catalyst exhibits excellent performance for OER, with a small overpotential of 182 mV at 10 mA cm
and long-term stability (200 h) at 200 mA cm
. Theoretical calculations show that the modification of Mo enhances the electron delocalization and optimizes the adsorption of intermediates.
As one of the serious public health issues, vaccination refusal has been attracting more and more attention, especially for newly approved human papillomavirus (HPV) vaccines. Understanding public ...opinion towards HPV vaccines, especially concerns on social media, is of significant importance for HPV vaccination promotion.
In this study, we leveraged a hierarchical machine learning based sentiment analysis system to extract public opinions towards HPV vaccines from Twitter. English tweets containing HPV vaccines-related keywords were collected from November 2, 2015 to March 28, 2016. Manual annotation was done to evaluate the performance of the system on the unannotated tweets corpus. Followed time series analysis was applied to this corpus to track the trends of machine-deduced sentiments and their associations with different days of the week.
The evaluation of the unannotated tweets corpus showed that the micro-averaging F scores have reached 0.786. The learning system deduced the sentiment labels for 184,214 tweets in the collected unannotated tweets corpus. Time series analysis identified a coincidence between mainstream outcome and Twitter contents. A weak trend was found for "Negative" tweets that decreased firstly and began to increase later; an opposite trend was identified for "Positive" tweets. Tweets that contain the worries on efficacy for HPV vaccines showed a relative significant decreasing trend. Strong associations were found between some sentiments ("Positive", "Negative", "Negative-Safety" and "Negative-Others") with different days of the week.
Our efforts on sentiment analysis for newly approved HPV vaccines provide us an automatic and instant way to extract public opinion and understand the concerns on Twitter. Our approaches can provide a feedback to public health professionals to monitor online public response, examine the effectiveness of their HPV vaccination promotion strategies and adjust their promotion plans.
This paper investigates physical layer security (PLS) in cognitive radio inspired non-orthogonal multiple access (CR-NOMA) networks with multiple primary and secondary users. To manage the ...interferences among the users and guarantee the quality of services of primary users, a new secure NOMA transmission strategy is designed, where the primary and secondary users are paired according to their channel gains, respectively, and power-domain NOMA is employed to transmit the signal. Then, the closed-form expressions for connection outage probability, secrecy outage probability, and effective secrecy throughput are derived for the primary users over Nakagami-<inline-formula><tex-math notation="LaTeX">m</tex-math></inline-formula> fading channels when the secondary users are considered as eavesdroppers. Typically, the secrecy performance can be improved by pairing the primary users with best channel gains or reducing the number of secondary users. In addition, we also investigate the performances of secondary users by deriving the closed-form expressions for throughput of secondary network. Furthermore, simulations are conducted to verify our analysis results and provide insights into the impact of the parameters on system performance.
We aim to learn a domain generalizable person re-identification (ReID) model. When such a model is trained on a set of source domains (ReID datasets collected from different camera networks), it can ...be directly applied to any new unseen dataset for effective ReID without any model updating. Despite its practical value in real-world deployments, generalizable ReID has seldom been studied. In this work, a novel deep ReID model termed Domain-Invariant Mapping Network (DIMN) is proposed. DIMN is designed to learn a mapping between a person image and its identity classifier, i.e., it produces a classifier using a single shot. To make the model domain-invariant, we follow a meta-learning pipeline and sample a subset of source domain training tasks during each training episode. However, the model is significantly different from conventional meta-learning methods in that: (1) no model updating is required for the target domain, (2) different training tasks share a memory bank for maintaining both scalability and discrimination ability, and (3) it can be used to match an arbitrary number of identities in a target domain. Extensive experiments on a newly proposed large-scale ReID domain generalization benchmark show that our DIMN significantly outperforms alternative domain generalization or meta-learning methods.
Most existing studies on unsupervised domain adaptation (UDA) assume that each domain's training samples come with domain labels (e.g., painting, photo). Samples from each domain are assumed to ...follow the same distribution and the domain labels are exploited to learn domain-invariant features via feature alignment. However, such an assumption often does not hold true-there often exist numerous finer-grained domains (e.g., dozens of modern painting styles have been developed, each differing dramatically from those of the classic styles). Therefore, forcing feature distribution alignment across each artificially-defined and coarse-grained domain can be ineffective. In this paper, we address both single-source and multi-source UDA from a completely different perspective, which is to view each instance as a fine domain . Feature alignment across domains is thus redundant. Instead, we propose to perform dynamic instance domain adaptation (DIDA). Concretely, a dynamic neural network with adaptive convolutional kernels is developed to generate instance-adaptive residuals to adapt domain-agnostic deep features to each individual instance. This enables a shared classifier to be applied to both source and target domain data without relying on any domain annotation. Further, instead of imposing intricate feature alignment losses, we adopt a simple semi-supervised learning paradigm using only a cross-entropy loss for both labeled source and pseudo labeled target data. Our model, dubbed DIDA-Net, achieves state-of-the-art performance on several commonly used single-source and multi-source UDA datasets including Digits, Office-Home, DomainNet, Digit-Five, and PACS.
In this article, we propose a method for performing the simultaneous localization and mapping (SLAM) in indoor environments using only multiple-input and multiple-output (MIMO) frequency-modulated ...continuous wave (FMCW) radar. The SLAM is a technology that maps the surrounding environment and position of a platform simultaneously. The following are the steps of the overall the SLAM implementation of the proposed method. First, the ego-velocity of the platform is estimated using the relative velocity with respect to stationary objects. In this case, a random sample consensus (RANSAC) algorithm is used in the velocity-angle map generated from the detection results to identify stationary objects. Second, the rotation angle of the platform is estimated using the linear component extracted from the walls in the <inline-formula> <tex-math notation="LaTeX">{xy} </tex-math></inline-formula> plane. Then, we determine the ego-motion of the platform using the estimated ego-velocity and rotation angle. Finally, we map the position of the platform and the indoor environment simultaneously in the absolute coordinate system. We validate the mapping result generated using our proposed method by comparing it with the ground truth-based mapping result.