Search

Multi-action dialog policy, which generates multiple atomic dialog actions per turn, has been widely applied in task-oriented dialog systems to provide expressive and efficient system responses. Existing policy models usually imitate action combinations from the labeled multi-action dialog examples. Due to data limitations, they generalize poorly toward unseen dialog flows. While reinforcement learning-based methods are proposed to incorporate the service ratings from real users and user simulators as external supervision signals, they suffer from sparse and less credible dialog-level rewards. To cope with this problem, we explore to improve multi-action dialog policy learning with explicit and implicit turn-level user feedback received for historical predictions (i.e., logged user feedback) that are cost-efficient to collect and faithful to real-world scenarios. The task is challenging since the logged user feedback provides only partial label feedback limited to the particular historical dialog actions predicted by the agent. To fully exploit such feedback information, we propose BanditMatch, which addresses the task from a feedback-enhanced semi-supervised learning perspective with a hybrid objective of semi-supervised learning and bandit learning. BanditMatch integrates pseudo-labeling methods to better explore the action space through constructing full label feedback. Extensive experiments show that our BanditMatch outperforms the state-of-the-art methods by generating more concise and informative responses. The source code and the appendix of this paper can be obtained from https://github.com/ShuoZhangXJTU/BanditMatch., Comment: AAAI 2023

Recent years have seen increasing concerns about the unsafe response generation of large-scale dialogue systems, where agents will learn offensive or biased behaviors from the real-world corpus. Some methods are proposed to address the above issue by detecting and replacing unsafe training examples in a pipeline style. Though effective, they suffer from a high annotation cost and adapt poorly to unseen scenarios as well as adversarial attacks. Besides, the neglect of providing safe responses (e.g. simply replacing with templates) will cause the information-missing problem of dialogues. To address these issues, we propose an unsupervised pseudo-label sampling method, TEMP, that can automatically assign potential safe responses. Specifically, our TEMP method groups responses into several clusters and samples multiple labels with an adaptively sharpened sampling strategy, inspired by the observation that unsafe samples in the clusters are usually few and distribute in the tail. Extensive experiments in chitchat and task-oriented dialogues show that our TEMP outperforms state-of-the-art models with weak supervision signals and obtains comparable results under unsupervised learning settings.

The drastic increase in language models' parameters has led to a new trend of deploying models in cloud servers, raising growing concerns about private inference for Transformer-based models. Existing two-party privacy-preserving techniques, however, only take into account natural language understanding (NLU) scenarios. Private inference in natural language generation (NLG), crucial for applications like translation and code completion, remains underexplored.In addition, previous privacy-preserving techniques suffer from convergence issues during model training and exhibit poor inference speed when used with NLG models due to the neglect of time-consuming operations in auto-regressive generations. To address these issues, we propose a fast private text generation framework for Transformer-based language models, namely MERGE.MERGE reuses the output hidden state as the word embedding to bypass the embedding computation and reorganize the linear operations in the Transformer module to accelerate the forward procedure. Extensive experiments show that MERGE achieves a 26.5x speedup to the vanilla encrypted model under the sequence length 512, and reduces 80\% communication cost, with an up to 10x speedup to state-of-the-art approximated models., Comment: Accepted by AAAI 2024

Recent years have witnessed the success of recurrent neural network (RNN) models in time series classification (TSC). However, neural networks (NNs) are vulnerable to adversarial samples, which cause real-life adversarial attacks that undermine the robustness of AI models. To date, most existing attacks target at feed-forward NNs and image recognition tasks, but they cannot perform well on RNN-based TSC. This is due to the cyclical computation of RNN, which prevents direct model differentiation. In addition, the high visual sensitivity of time series to perturbations also poses challenges to local objective optimization of adversarial samples. In this paper, we propose an efficient method called TSFool to craft highly-imperceptible adversarial time series for RNN-based TSC. The core idea is a new global optimization objective known as "Camouflage Coefficient" that captures the imperceptibility of adversarial samples from the class distribution. Based on this, we reduce the adversarial attack problem to a multi-objective optimization problem that enhances the perturbation quality. Furthermore, to speed up the optimization process, we propose to use a representation model for RNN to capture deeply embedded vulnerable samples whose features deviate from the latent manifold. Experiments on 11 UCR and UEA datasets showcase that TSFool significantly outperforms six white-box and three black-box benchmark attacks in terms of effectiveness, efficiency and imperceptibility from various perspectives including standard measure, human study and real-world defense., Comment: 22 pages, 16 figures

Dan He,1 Wan Dan,1 Qing Du,1 Bing-Bing Shen,1 Lin Chen,1 Liang-zi Fang,1 Jian-Jun Kuang,1 Chun-yu Tang,2 Ping Cai,1 Rong Yu,1,3 Shui-han Zhang,1 Jian-hua Huang1,3 1Hunan Academy of Chinese Medicine, Hunan University of Chinese Medicine, Changsha, Hunan, 410013, People’s Republic of China; 2Hunan Times Sunshine Pharmaceutical Co., Ltd., Changsha, Hunan, 425007, People’s Republic of China; 3Hunan Key Laboratory of TCM Prescription and Syndromes Translational Medicine Hunan, Changsha, Hunan, 410208, People’s Republic of ChinaCorrespondence: Shui-han Zhang; Jian-hua Huang, Hunan Academy of Chinese Medicine, Hunan University of Chinese Medicine, Changsha, Hunan, 410013, People’s Republic of China, Tel +86 13637400650 ; +86 18692265317, Email zhangshuihan0220@126.com; huangjianhua1985@163.comPurpose: This study aimed to reveal the multicomponent synergy mechanisms of SWP based on network pharmacology and metabolomics for exploring the relationships of active ingredients, biological targets, and crucial metabolic pathways.Materials: Network pharmacology, including TRRUST, GO, and KEGG, enrichment was used to discover the active ingredients and potential regulation mechanisms of SWP. LC-MS and multivariate data analysis method were further applied to analyze serum metabolomics profiling for discovering the potential metabolic mechanisms of SWP on AA induced by Cyclophosphamide (CTX) and 1-Acetyl-2-phenylhydrazine (APH).Results: A total of 27 important bioactive ingredients meeting the ADME (absorption, distribution, metabolism, and excretion) screening criteria from SWP were selected. Interaction networks were constructed and validated based on the 10 associated ingredients with the relevant targets. A total of 125 biomarkers were found by Metabolomics approach, which associated with the development of AA, mainly involved in amino acid metabolism and lipid metabolism. While SWP can reverse the above 12 metabolites changed by AA. Network analysis revealed the synergi

In the context of the coronavirus disease 2019 pandemic, we investigated the epidemiological and clinical characteristics of a young patient infected by avian influenza A (H5N6) virus in Anhui Province, East China, and analyzed genomic features of the pathogen in 2020. Through the cross-sectional investigation of external environment monitoring (December 29–31, 2020), 1909 samples were collected from Fuyang City. It was found that the positive rate of H5N6 was higher than other areas obviously in Tianma poultry market, where the case appeared. In addition, dual coinfections were detected with a 0.057% polymerase chain reaction positive rate the surveillance years. The virus was the clade 2.3.4.4, which was most likely formed by genetic reassortment between H5N6 and H9N2 viruses. This study found that the evolution rates of the hemagglutinin and neuraminidase genes of the virus were higher than those of common seasonal influenza viruses. The virus was still highly pathogenic to poultry and had a preference for avian receptor binding.

Olivine is a ubiquitous mineral in mafic-ultramafic rocks and has been widely used as a mineral marker in various geological processes. However, its development of trace elements is limited. Here we present newly-obtained trace element data 7Li, 27Al, 29Si, 31P, 43Ca, 45Sc, 49Ti, 51V, 53Cr, 55Mn, 59Co, 60Ni, and 66Zn of olivine in typical mantle xenoliths, mantle peridotites in ophiolites, and plutonic rocks from layered and Alaskan-type intrusions to develop trace element proxies for the petrogenesis, mineralization and discrimination of various mafic-ultramafic rocks. Residual olivine grains in mantle xenoliths and ophiolitic peridotites, which represent residues of mantle melting, have higher Ni/Co (>20) and Ni/Mn (>2) ratios than magmatic olivine (Ni/Co < 20, Ni/Mn < 2), which are consistent with the compatibilities of these elements during partial melting and magma differentiation. Lower Ni content, and lower Ni/Co and Ni/Mn ratios at a given Fo content can distinguish olivine in Alaskan-type intrusions from layered intrusions, reflecting the nature of their mantle sources. The V and Sc contents and V/Sc ratios in olivine can distinguish mantle xenoliths (V > 2 ppm, V/Sc > 0.5) from ophiolitic peridotites (V < 2 ppm, V/Sc < 0.5), indicating a more reduced state of continental lithospheric mantle compared to the oceanic lithospheric mantle. As a consequence, the four occurrences of mafic-ultramafic rocks can be distinguished by olivine with (Sc × 10)-(Ti × 2)-Zn and V/Sc-(Co/Ni × 2)-(Zn/Mn × 5) ternary diagrams. In addition, Li, Ti and P contents in olivine are good tracers of melt/fluid metasomatism, whereas Ni/Co, Ni/Mn and Mn/Zn ratios are indicators of chromite mineralization. Therefore, trace elements in olivine can be used as chemical proxies to distinguish the origin of various mafic-ultr

Experimental data suggests that PM1 is more toxic than PM2.5 although the epidemiologic evidence suggests that the health associations are similar. However, few objective exposure data are available to compare the associations of PM1 and PM2.5 with children lung function. Our objectives are a) to evaluate associations between long-term exposure to PM1, PM2.5 and children's lung function, and b) to compare the associations between PM1 and PM2.5. From 2012 to 2013, we enrolled 6,740 children (7–14 years), randomly recruited from primary and middle schools located in seven cities in northeast China. We measured lung function including forced vital capacity (FVC), forced expiratory volume in 1 s (FEV1), peak expiratory flow (PEF), and maximal mid-expiratory flow (MMEF) utilizing two portable electronic spirometers. We dichotomized continuous lung function measures according the expected values for gender and age. The spatial resolution at which PM1 and PM2.5 estimated were estimated using a machine learning method and the temporal average concentrations were averaged from 2009 to 2012. A multilevel regression model was used to estimate the associations of PM1, PM2.5 exposure and lung function measures, adjusted for confounding factors. Associations with lower lung function were consistently larger for PM1 than for PM2.5. Adjusted odds ratios (OR) per interquartile range greater PM1 ranged from 1.53 for MMEF (95% confidence interval [CI]: 1.20–1.96) to 2.14 for FEV1 (95% CI: 1.66–2.76) and ORs for PM2.5 ranged from 1.36 for MMEF (95%CI: 1.12–1.66) to 1.82 for FEV1 (95%CI: 1.49–2.22), respectively. PM1 and PM2.5 had significant associations with FVC and FEV1 in primary school children, and on PEF and MMEF in middle school children. Long-term PM1 and PM2.5

Zi-Hong Liang,1 Yan-Bo Jia,2 Mei-Ling Wang,1 Zi-Ru Li,1 Min Li,1 Yong-Li Yun,1 Run-Xiu Zhu1 1Department of Neurology, Inner Mongolia Autonomous Region People’s Hospital, Huhhot, Inner Mongolia, People’s Republic of China; 2Department of Orthopaedics, The Second Affiliated Hospital of Inner Mongolia Medical University, Huhhot, Inner Mongolia, People’s Republic of ChinaCorrespondence: Run-Xiu ZhuDepartment of Neurology, Inner Mongolia Autonomous Region People’s Hospital, No. 20 Zhaowuda Road, Huhhot, Inner Mongolia010017, People’s Republic of ChinaTel +86 0 471 328 3999Fax +86 0 471 328 3999Email xcinhah-21@foxmail.comBackground: Post-stroke depression (PSD) is one of the most common psychiatric diseases afflicting stroke survivors. This study was conducted to assess the efficacy of ginkgo biloba extract as augmentation of venlafaxine in treating PSD.Methods: The included PSD patients were randomly assigned into the experiment group (receiving ginkgo biloba extract plus venlafaxine) and control group (receiving venlafaxine alone). The treatment was continued for eight weeks. The Hamilton Depression Rating Scale (HDRS) and the Self-rating Depression Scale (SDS) were used to assess the depressive symptoms. The National Institutes of Health Stroke Scale (NIHSS) was used to assess the neurological defect, and the Activities of Daily Living (ADL) was used to assess recovery of abilities of patients after stroke. Meanwhile, the levels of serum 5-hydroxytryptamine (5-HT) and brain-derived neurotrophic factor (BDNF) were measured before and after treatment. The dose of venlafaxine used and adverse events were also recorded.Results: Each group had 40 PSD patients. After treatment, the depressive symptoms, neurological defect and living function were significantly improved in both groups. But the patients receiving ginkgo biloba extract plus venlafaxine had the signi

Zi-Hong Liang,1 Yan-Bo Jia,2 Zi-Ru Li,1 Min Li,1 Mei-Ling Wang,1 Yong-Li Yun,1 Li-Jun Yu,1 Lei Shi,1 Run-Xiu Zhu11Department of Neurology, Inner Mongolia Autonomous Region People’s Hospital, Huhhot, Inner Mongolia, People’s Republic of China; 2Department of Orthopaedics, The Second Affiliated Hospital of Inner Mongolia Medical University, Huhhot, Inner Mongolia, People’s Republic of ChinaBackground: Depression can seriously affect the quality of life of type 2 diabetes mellitus (T2DM) patients after stroke. However, there were still no objective methods to diagnose T2DM patients with poststroke depression (PSD). Therefore, we conducted this study to deal with this problem.Methods: Gas chromatography-mass spectroscopy (GC-MS)-based metabolomics profiling method was used to profile the urinary metabolites from 83 nondepressed T2DM patients after stroke and 101 T2DM patients with PSD. The orthogonal partial least-squares discriminant analysis was conducted to explore the metabolic differences in T2DM patients with PSD. The logistic regression analysis was performed to identify the optimal and simplified biomarker panel for diagnosing T2DM patients with PSD. The receiver operating characteristic curve analysis was used to assess the diagnostic performance of this biomarker panel.Results: In total, 23 differential metabolites (7 decreased and 16 increased in T2DM patients with PSD) were found. A panel consisting of pseudouridine, malic acid, hypoxanthine, 3,4-dihydroxybutyric acid, fructose and inositol was identified. This panel could effectively separate T2DM patients with PSD from nondepressed T2DM patients after stroke. The area under the curve was 0.965 in the training set and 0.909 in the validation set. Meanwhile, we found that the galactose metabolism was significantly affected in T2DM patients with PSD.Conclusion: Our results could be helpful for future development of an objective method to diagnose T2DM patients with PSD and provid

Polarized growth of pollen tubes is a critical step for successful reproduction in angiosperms and is controlled by ROP GTPases. Spatiotemporal activation of ROP (Rho GTPases of plants) necessitates a complex and sophisticated regulatory system, in which guanine nucleotide exchange factors (RopGEFs) are key components. It was previously shown that a leucine-rich repeat receptor-like kinase, Arabidopsis pollen receptor kinase 2 (AtPRK2), interacted with RopGEF12 for its membrane recruitment. However, the mechanisms underlying AtPRK2-mediated ROP activation in vivo are yet to be defined. It is reported here that over-expression of AtPRK2 induced tube bulging that was accompanied by the ectopic localization of ROP-GTP and the ectopic distribution of actin microfilaments. Tube depolarization was also induced by a potentially kinase-dead mutant, AtPRK2K366R, suggesting that the over-expression effect of AtPRK2 did not require its kinase activity. By contrast, deletions of non-catalytic domains in AtPRK2, i.e. the juxtamembrane (JM) and carboxy-terminal (CT) domains, abolished its ability to affect tube polarization. Notably, AtPRK2K366R retained the ability to interact with RopGEF12, whereas AtPRK2 truncations of these non-catalytic domains did not. Lastly, it has been shown that the JM and CT domains of AtPRK2 were not only critical for its interaction with RopGEF12 but also critical for its distribution at the plasma membrane. These results thus provide further insight into pollen receptor kinase-mediated ROP activation during pollen tube growth.

Polarized growth of pollen tubes is a critical step for successful reproduction in angiosperms and is controlled by ROP GTPases. Spatiotemporal activation of ROP (Rho GTPases of plants) necessitates a complex and sophisticated regulatory system, in which guanine nucleotide exchange factors (RopGEFs) are key components. It was previously shown that a leucine-rich repeat receptor-like kinase, Arabidopsis pollen receptor kinase 2 (AtPRK2), interacted with RopGEF12 for its membrane recruitment. However, the mechanisms underlying AtPRK2-mediated ROP activation in vivo are yet to be defined. It is reported here that over-expression of AtPRK2 induced tube bulging that was accompanied by the ectopic localization of ROP-GTP and the ectopic distribution of actin microfilaments. Tube depolarization was also induced by a potentially kinase-dead mutant, AtPRK2K366R, suggesting that the over-expression effect of AtPRK2 did not require its kinase activity. By contrast, deletions of non-catalytic domains in AtPRK2, i.e. the juxtamembrane (JM) and carboxy-terminal (CT) domains, abolished its ability to affect tube polarization. Notably, AtPRK2K366R retained the ability to interact with RopGEF12, whereas AtPRK2 truncations of these non-catalytic domains did not. Lastly, it has been shown that the JM and CT domains of AtPRK2 were not only critical for its interaction with RopGEF12 but also critical for its distribution at the plasma membrane. These results thus provide further insight into pollen receptor kinase-mediated ROP activation during pollen tube growth.