Search

The deployment of large-scale generative models is often restricted by their potential risk of causing harm to users in unpredictable ways. We focus on the problem of black-box red teaming, where a red team generates test cases and interacts with the victim model to discover a diverse set of failures with limited query access. Existing red teaming methods construct test cases based on human supervision or language model (LM) and query all test cases in a brute-force manner without incorporating any information from past evaluations, resulting in a prohibitively large number of queries. To this end, we propose Bayesian red teaming (BRT), novel query-efficient black-box red teaming methods based on Bayesian optimization, which iteratively identify diverse positive test cases leading to model failures by utilizing the pre-defined user input pool and the past evaluations. Experimental results on various user input pools demonstrate that our method consistently finds a significantly larger number of diverse positive test cases under the limited query budget than the baseline methods. The source code is available at https://github.com/snu-mllab/Bayesian-Red-Teaming., ACL 2023 Long Paper - Main Conference

Objectives Exosomes are extracellular nano-sized vesicles which transport proteins, lipids, nucleic acids and polysaccharides. Panax ginseng C.A. Meyer is known to have various pharmacological activities including anti-inflammatory and antioxidant effects. Furthermore, ginseng derived exosome-like nanoparticles (GDEs) may be considered to contribute to inflammation mechanisms. The main aim of this study elucidated that GDEs exert anti-inflammatory activities, which can molecularly modulate an inflammation mechanism in mammalian cells.Methods GDEs were isolated by the method using sucrose cushioning ultracentrifugation and the morphology of GDEs was shown through transmission electron microscope (TEM). The number of particles and size distribution of the GDEs were evaluated using nanoparticle tracking analysis (NTA). To investigate the anti-inflammatory activity of GDEs, the expression levels of inflammatory mediators and cytokines were analyzed in lipopoly-saccarides (LPS)-induced RAW 264.7, murine macrophages. ELISA, qPCR, WB and nitric oxide assay were carried out for analyzing anti-inflammatory effects of the GDEs in mammalian cells.Results GDEs showed a spherical shape with an average diameter of 176.2 nm and a substantial number of GDEs was counted as 1.09 x 1012 ± 1.19 x 1011 particles/ml. The pretreated GDEs in LPS-induced RAW264.7 suppressed the protein expression levels of pro-inflammatory cytokines including interlukin-1beta (IL-1β), interlukin-6 (IL-6) and tumor necrosis factor-alpha (TNF-α). In addition, protein expression levels of inflammation-related enzymes, cyclooxygenase-2 (COX-2) and inducible nitric oxide synthase (iNOS) as well as production of nitric oxide (NO) were dramatically decreased by the GDEs in LPS-induced RAW264.7. Correspondingly, the GDEs treatment clearly reduced LPS-induced mRNA expression levels of the iNOS and COX-2. In addition, expression patterns of the IL-1β, IL-6, TNF-α were gradually down-regulated by the GDEs in a dosage dependent manner.Conclusion These results demonstrated that Ginseng-derived exosome-like nanoparticles inhibited the expression of the inflammation mediators, suggesting that the GDEs possess anti-inflammatory activities and molecularly regulate the inflammatory mechanism in mammalian cells. Collectively, the GDEs has high potential as a novel therapeutic agent to treat chronic inflammatory diseases.

Glucosinolates (GSLs) are secondary metabolites providing defense against pathogens and herbivores in plants, and anti-carcinogenic activity against human cancer cells. Profiles of GSLs vary greatly among members of genus Brassica. In this study, we found that a reference line of Chinese cabbage (B. rapa ssp. pekinensis), ‘Chiifu’ contains significantly lower amounts of total GSLs than the oilseed-type B. rapa (B. rapa ssp. trilocularis) line ‘LP08’. This study aimed to identify the key regulators of the high accumulation of GSLs in Brassica rapa plants using transcriptomic and linkage mapping approaches. Comparative transcriptome analysis showed that, in total, 8,276 and 9,878 genes were differentially expressed between ‘Chiifu’ and ‘LP08’ under light and dark conditions, respectively. Among 162 B. rapa GSL pathway genes, 79 were related to GSL metabolism under light conditions. We also performed QTL analysis using a single nucleotide polymorphism-based linkage map constructed using 151 F5 individuals derived from a cross between the ‘Chiifu’ and ‘LP08’ inbred lines. Two major QTL peaks were successfully identified on chromosome 3 using high-performance liquid chromatography to obtain GSL profiles from 97 F5 recombinant inbred lines. The MYB-domain transcription factor gene BrMYB28.1 (Bra012961) was found in the highest QTL peak region. The second highest peak was located near the 2-oxoacid-dependent dioxygenase gene BrGSL-OH.1 (Bra022920). This study identified major genes responsible for differing profiles of GSLs between ‘Chiifu’ and ‘LP08’. Thus, our study provides molecular insights into differences in GSL profiles between vegetative- and oilseed-type B. rapa plants.

Ambiguous questions persist in open-domain question answering, because formulating a precise question with a unique answer is often challenging. Previously, Min et al. (2020) have tackled this issue by generating disambiguated questions for all possible interpretations of the ambiguous question. This can be effective, but not ideal for providing an answer to the user. Instead, we propose to ask a clarification question, where the user's response will help identify the interpretation that best aligns with the user's intention. We first present CAMBIGNQ, a dataset consisting of 5,654 ambiguous questions, each with relevant passages, possible answers, and a clarification question. The clarification questions were efficiently created by generating them using InstructGPT and manually revising them as necessary. We then define a pipeline of tasks and design appropriate evaluation metrics. Lastly, we achieve 61.3 F1 on ambiguity detection and 40.5 F1 on clarification-based QA, providing strong baselines for future work., Comment: 15 pages, 4 figures

Vernalization is the process by which long-term cold like winter triggers transition to flowering in plants. Many biennial and perennial plants including Brassicaceae family plants require vernalization for floral transition. Not only floral transition, but dynamic physiological and metabolic changes might also take place during vernalization. However, vernalization-mediated metabolic change is merely investigated so far. One of secondary metabolites found in Brassiceceae family plants is glucosinolates (GSLs). GSLs provides defense against pathogens and herbivores attack in plants and also exhibits inhibitory activity against human cancer cell. Profiles of GSLs are highly modulated by different environmental stresses in Brassciaceae family plants. To grasp the effect of vernalization on GSLs metabolic dynamics in radish (Raphanus sativus L.), we performed transcriptomic and metabolic analysis during vernalization in radish. Through transcriptome analysis, we found many GSLs metabolic genes were significantly down-regulated by vernalization in radish plants. Ultra-High Performance Liquid Chromatography analysis also revealed that GSLs compounds were substantially reduced in vernalized radish samples compared to non-vernalized radish samples. Furthermore, we found that repressive histone modification (i.e. H3K27me3) is involved in the modulation of GSLs metabolism via epigenetic suppression of Glucoraphasatin Synthase 1 (GRS1) during vernalization in radish. This study revealed that GSLs metabolism is modulated by vernalization, suggestive of a newly identified target of vernalization in radish.

Through in-context learning (ICL), large-scale language models are effective few-shot learners without additional model fine-tuning. However, the ICL performance does not scale well with the number of available training samples as it is limited by the inherent input length constraint of the underlying language model. Meanwhile, many studies have revealed that language models are also powerful feature extractors, allowing them to be utilized in a black-box manner and enabling the linear probing paradigm, where lightweight discriminators are trained on top of the pre-extracted input representations. This paper proposes prompt-augmented linear probing (PALP), a hybrid of linear probing and ICL, which leverages the best of both worlds. PALP inherits the scalability of linear probing and the capability of enforcing language models to derive more meaningful representations via tailoring input into a more conceivable form. Throughout in-depth investigations on various datasets, we verified that PALP significantly enhances the input representations closing the gap between ICL in the data-hungry scenario and fine-tuning in the data-abundant scenario with little training overhead, potentially making PALP a strong alternative in a black-box scenario., AAAI 2023

While Transformers have had significant success in paragraph generation, they treat sentences as linear sequences of tokens and often neglect their hierarchical information. Prior work has shown that decomposing the levels of granularity~(e.g., word, phrase, or sentence) for input tokens has produced substantial improvements, suggesting the possibility of enhancing Transformers via more fine-grained modeling of granularity. In this work, we propose a continuous decomposition of granularity for neural paraphrase generation (C-DNPG). In order to efficiently incorporate granularity into sentence encoding, C-DNPG introduces a granularity-aware attention (GA-Attention) mechanism which extends the multi-head self-attention with: 1) a granularity head that automatically infers the hierarchical structure of a sentence by neurally estimating the granularity level of each input token; and 2) two novel attention masks, namely, granularity resonance and granularity scope, to efficiently encode granularity into attention. Experiments on two benchmarks, including Quora question pairs and Twitter URLs have shown that C-DNPG outperforms baseline models by a remarkable margin and achieves state-of-the-art results in terms of many metrics. Qualitative analysis reveals that C-DNPG indeed captures fine-grained levels of granularity with effectiveness., Accepted to be published in COLING 2022

Purpose: This study evaluated the efficiency of preoperative ultrasound (US)-guided tattooing of the axillary lymph nodes with activated charcoal and the correlation between sonographically suspicious nodes and final histologic results by node-to-node analysis. The concordance rate between the tattooed nodes and sentinel nodes was also determined. Methods : : US-guided tattooing of sonographically suspicious axillary nodes was performed preoperatively by an injection of activated charcoal. The identification of black pigment and the concordance between the sentinel and tattooed nodes was evaluated. Results : : Regarding node-to-node analysis, the false-negative rate of US-fine needle aspiration (FNA) was 43.3%. The sensitivity and negative predictive values were 56.7% and 81.7%, respectively. The specificity and positive predictive values were 100%. The accuracy of US-FNA was 85.2%. In the final pathology, 45/125 patients (36.0%) had positive nodes, including two micrometastases. The false-negative rate of sentinel lymph node biopsy (SLNB) was 4.0%, but there were no skip metastases. The sensitivity and specificity of SLNB were 95.6% and 100%, respectively. The negative predictive value was 97.6%, and the positive predictive value was 100%. The accuracy of SLNB was 98.4%. In 117 of 125 patients (93.6%), there was concordance between the charcoal tattooed axillary lymph nodes and SLNs. Conclusion : : SLNB, in conjunction with US-guided tattooing of sonographically suspicious axillary lymph nodes, is a useful procedure to reduce the false-negative rate of SLNB and improve the accuracy of an intraoperative evaluation of axillary nodes in breast cancer patients. This paper proposes the concept of targeted axillary node biopsy with preoperative US-guided tattooing for the most accurate axillary staging in patients with breast cancer.

Current natural language interaction for self-tracking tools largely depends on bespoke implementation optimized for a specific tracking theme and data format, which is neither generalizable nor scalable to a tremendous design space of self-tracking. However, training machine learning models in the context of self-tracking is challenging due to the wide variety of tracking topics and data formats. In this paper, we propose a novel NLP task for self-tracking that extracts close- and open-ended information from a retrospective activity log described as a plain text, and a domain-agnostic, GPT-3-based NLU framework that performs this task. The framework augments the prompt using synthetic samples to transform the task into 10-shot learning, to address a cold-start problem in bootstrapping a new tracking topic. Our preliminary evaluation suggests that our approach significantly outperforms the baseline QA models. Going further, we discuss future application domains toward which the NLP and HCI researchers can collaborate., Accepted to NAACL '22 2nd Workshop on Bridging Human-Computer Interaction and Natural Language Processing. 10 pages including appendix, 2 figures, and 1 table

Despite recent explosion of interests in in-context learning, the underlying mechanism and the precise impact of the quality of demonstrations remain elusive. Intuitively, ground-truth labels should have as much impact in in-context learning (ICL) as supervised learning, but recent work reported that the input-label correspondence is significantly less important than previously thought. Intrigued by this counter-intuitive observation, we re-examine the importance of ground-truth labels in in-context learning. With the introduction of two novel metrics, namely Label-Correctness Sensitivity and Ground-truth Label Effect Ratio (GLER), we were able to conduct quantifiable analysis on the impact of ground-truth label demonstrations. Through extensive analyses, we find that the correct input-label mappings can have varying impacts on the downstream in-context learning performances, depending on the experimental configuration. Through additional studies, we identify key components, such as the verbosity of prompt templates and the language model size, as the controlling factor to achieve more noise-resilient ICL., Accepted to EMNLP Long. Kang Min Yoo and Junyeob Kim contributed equally. Kang Min Yoo and Taeuk Kim are the corresponding authors

Remembering important information from the past and continuing to talk about it in the present are crucial in long-term conversations. However, previous literature does not deal with cases where the memorized information is outdated, which may cause confusion in later conversations. To address this issue, we present a novel task and a corresponding dataset of memory management in long-term conversations, in which bots keep track of and bring up the latest information about users while conversing through multiple sessions. In order to support more precise and interpretable memory, we represent memory as unstructured text descriptions of key information and propose a new mechanism of memory management that selectively eliminates invalidated or redundant information. Experimental results show that our approach outperforms the baselines that leave the stored memory unchanged in terms of engagingness and humanness, with larger performance gap especially in the later sessions., Comment: Accepted to EMNLP2022 Findings

Task-oriented dialogue (TOD) systems are mainly based on the slot-filling-based TOD (SF-TOD) framework, in which dialogues are broken down into smaller, controllable units (i.e., slots) to fulfill a specific task. A series of approaches based on this framework achieved remarkable success on various TOD benchmarks. However, we argue that the current TOD benchmarks are limited to surrogate real-world scenarios and that the current TOD models are still a long way to cover the scenarios. In this position paper, we first identify current status and limitations of SF-TOD systems. After that, we explore the WebTOD framework, the alternative direction for building a scalable TOD system when a web/mobile interface is available. In WebTOD, the dialogue system learns how to understand the web/mobile interface that the human agent interacts with, powered by a large-scale language model.

Recent studies have proposed unified user modeling frameworks that leverage user behavior data from various applications. Many of them benefit from utilizing users' behavior sequences as plain texts, representing rich information in any domain or system without losing generality. Hence, a question arises: Can language modeling for user history corpus help improve recommender systems? While its versatile usability has been widely investigated in many domains, its applications to recommender systems still remain underexplored. We show that language modeling applied directly to task-specific user histories achieves excellent results on diverse recommendation tasks. Also, leveraging additional task-agnostic user histories delivers significant performance benefits. We further demonstrate that our approach can provide promising transfer learning capabilities for a broad spectrum of real-world recommender systems, even on unseen domains and services., Comment: ACL 2023 main conference

Text-to-image generation and image captioning are recently emerged as a new experimental paradigm to assess machine intelligence. They predict continuous quantity accompanied by their sampling techniques in the generation, making evaluation complicated and intractable to get marginal distributions. Based on a recent trend that multimodal generative evaluations exploit a vison-and-language pre-trained model, we propose the negative Gaussian cross-mutual information using the CLIP features as a unified metric, coined by Mutual Information Divergence (MID). To validate, we extensively compare it with competing metrics using carefully-generated or human-annotated judgments in text-to-image generation and image captioning tasks. The proposed MID significantly outperforms the competitive methods by having consistency across benchmarks, sample parsimony, and robustness toward the exploited CLIP model. We look forward to seeing the underrepresented implications of the Gaussian cross-mutual information in multimodal representation learning and the future works based on this novel proposition.

Many recent studies on large-scale language models have reported successful in-context zero- and few-shot learning ability. However, the in-depth analysis of when in-context learning occurs is still lacking. For example, it is unknown how in-context learning performance changes as the training corpus varies. Here, we investigate the effects of the source and size of the pretraining corpus on in-context learning in HyperCLOVA, a Korean-centric GPT-3 model. From our in-depth investigation, we introduce the following observations: (1) in-context learning performance heavily depends on the corpus domain source, and the size of the pretraining corpus does not necessarily determine the emergence of in-context learning, (2) in-context learning ability can emerge when a language model is trained on a combination of multiple corpora, even when each corpus does not result in in-context learning on its own, (3) pretraining with a corpus related to a downstream task does not always guarantee the competitive in-context learning performance of the downstream task, especially in the few-shot setting, and (4) the relationship between language modeling (measured in perplexity) and in-context learning does not always correlate: e.g., low perplexity does not always imply high in-context few-shot learning performance., Comment: Accepted to NAACL2022 as a long paper. Camera-ready version

Pre-trained language models (PLM) have marked a huge leap in neural dialogue modeling. While PLMs are pre-trained on large-scale text corpora, they are usually fine-tuned on scarce dialogue data with specific domain knowledge and dialogue styles. However, tailoring the language models while fully utilizing prior knowledge in large pre-trained models remains a challenge. In this paper, we present a novel approach for pre-trained dialogue modeling that casts the dialogue generation problem as a prompt-learning task. Instead of fine-tuning on limited dialogue data, our approach, DialogPrompt, learns continuous prompt embeddings optimized for dialogue contexts, which appropriately elicit knowledge from the large pre-trained model. To encourage the model to better utilize the prompt embeddings, the prompt encoders are designed to be dynamically generated based on the dialogue context. Experiments on popular conversation datasets show that our approach significantly outperforms the fine-tuning baseline and the generic prompt-learning methods. Furthermore, human evaluations strongly support the superiority of DialogPrompt in regard to response generation quality.

Metadata attributes (e.g., user and product IDs from reviews) can be incorporated as additional inputs to neural-based NLP models, by modifying the architecture of the models, in order to improve their performance. Recent models however rely on pretrained language models (PLMs), where previously used techniques for attribute injection are either nontrivial or ineffective. In this paper, we propose a lightweight and memory-efficient method to inject attributes to PLMs. We extend adapters, i.e. tiny plug-in feed-forward modules, to include attributes both independently of or jointly with the text. To limit the increase of parameters especially when the attribute vocabulary is large, we use low-rank approximations and hypercomplex multiplications, significantly decreasing the total parameters. We also introduce training mechanisms to handle domains in which attributes can be multi-labeled or sparse. Extensive experiments and analyses on eight datasets from different domains show that our method outperforms previous attribute injection methods and achieves state-of-the-art performance on various datasets.

A hypernomic reaction or an abnormal inflammatory process could cause a series of diseases, such as cardiovascular disease, neurodegeneration, and cancer. Additionally, oxidative stress has been identified to induce severe tissue injury and inflammation. Carpesium cernuum L. (C. cernuum) is a Chinese folk medicine used for its anti-inflammatory, analgesic, and detoxifying properties. However, the underlying molecular mechanism of C. cernuum in inflammatory and oxidative stress conditions remains largely unknown. The aim of this study was to examine the effects of a methanolic extract of C. cernuum (CLME) on lipopolysaccharide- (LPS-) induced RAW 264.7 mouse macrophages and a sepsis mouse model. The data presented in this study indicated that CLME inhibited LPS-induced production of proinflammatory mediators such as nitric oxide (NO) and prostaglandin E2 (PGE2) in RAW 264.7 cells. CLME treatment also reduced reactive oxygen species (ROS) generation and enhanced the expression of heme oxygenase-1 (HO-1) protein in a dose-dependent manner in the LPS-stimulated RAW 264.7 cells. Moreover, CLME treatment abolished the nuclear translocation of nuclear factor-κB (NF-κB), enhanced the activation of nuclear factor-erythroid 2 p45-related factor 2 (Nrf2), and reduced the expression of extracellular signal-related kinase (ERK) and ERK kinase (MEK) phosphorylation in LPS-stimulated RAW 264.7 cells. These outcomes implied that CLME could be a potential antioxidant and anti-inflammatory agent.

Tracking suspected cases of COVID-19 is crucial to suppressing the spread of COVID-19 pandemic. Active monitoring and proactive inspection are indispensable to mitigate COVID-19 spread, though these require considerable social and economic expense. To address this issue, we introduce CareCall, a call-based dialog agent which is deployed for active monitoring in Korea and Japan. We describe our system with a case study with statistics to show how the system works. Finally, we discuss a simple idea which uses CareCall to support proactive inspection.

Smart grids incorporating internet-of-things are emerging solutions to provide a reliable, sustainable and efficient electricity supply, and electric vehicle drivers can access efficient charging services in the smart grid. However, traditional electric vehicle charging systems are vulnerable to distributed denial of service and privileged insider attacks when the central charging server is attacked. The blockchain-based charging systems have been proposed to resolve these problems. In 2018, Huang et al. proposed the electric vehicle charging system using lightning network and smart contract. However, their system has an inefficient charging mechanism and does not guarantee security of key. We propose a secure charging system for electric vehicles based on blockchain to resolve these security flaws. Our charging system ensures the security of key, secure mutual authentication, anonymity, and perfect forward secrecy, and also provides efficient charging. We demonstrate that our proposed system provides secure mutual authentication using Burrows&ndash, Abadi&ndash, Needham logic and prevents replay and man-in-the-middle attacks using automated validation of internet security protocols and applications simulation tool. Furthermore, we compare computation and communication costs with previous schemes. Therefore, the proposed charging system efficiently applies to practical charging systems for electric vehicles.

Here, we sought to monitor bone marrow-derived dendritic cell (BMDC) migration and antitumor effects using a multimodal reporter imaging strategy in living mice. BMDCs were transduced with retroviral vector harboring human sodium iodide symporter (hNIS, nuclear imaging reporter), firefly luc2 (optical imaging reporter), and thy1.1 (surrogate marker of NIS and luc2) genes (BMDC/NF cells). No significant differences in biological functions, including cell proliferation, antigen uptake, phenotype expression, and migration ability, were observed between BMDC and BMDC/NF cells. Combined bioluminescence imaging and I-124 positron emission tomography/computed tomography clearly revealed the migration of BMDC/NF cells to draining popliteal lymph nodes at day 7 postinjection. Interestingly, marked tumor protection was observed in mice immunized with TC-1 lysate-pulsed BMDC/NF cells. Our findings suggested that multimodal reporter gene imaging of NIS and luciferase could provide insights into the biological behaviors of dendritic cells in living organisms and could be a useful tool for the optimization of DC-based immunotherapy protocols.

Eradication of Helicobacter pylori is recommended for the management of various gastric diseases, including peptic ulcers and mucosa-associated lymphoid tissue lymphoma. Because of the increasing prevalence of antibiotic resistance, the eradication rates of antibiotic-based therapies have decreased. Therefore, alternative treatments should be considered. The antibacterial properties of fatty acids (FAs) have been investigated in various organisms, including H. pylori. Some FAs, particularly polyunsaturated FAs, have been shown to have bactericidal activity against H. pylori in vitro; however, their antibacterial effects in vivo remain controversial. Poor solubility and delivery of FAs may be important reasons for this discrepancy. Recently, a series of studies demonstrated the antibacterial effects of a liposomal formulation of linolenic acid against H. pylori, both in vitro and in vivo. Further research is needed to improve the bioavailability of FAs and apply them in clinical use.

Goal-oriented dialog has been given attention due to its numerous applications in artificial intelligence. Goal-oriented dialogue tasks occur when a questioner asks an action-oriented question and an answerer responds with the intent of letting the questioner know a correct action to take. To ask the adequate question, deep learning and reinforcement learning have been recently applied. However, these approaches struggle to find a competent recurrent neural questioner, owing to the complexity of learning a series of sentences. Motivated by theory of mind, we propose "Answerer in Questioner's Mind" (AQM), a novel information theoretic algorithm for goal-oriented dialog. With AQM, a questioner asks and infers based on an approximated probabilistic model of the answerer. The questioner figures out the answerer's intention via selecting a plausible question by explicitly calculating the information gain of the candidate intentions and possible answers to each question. We test our framework on two goal-oriented visual dialog tasks: "MNIST Counting Dialog" and "GuessWhat?!". In our experiments, AQM outperforms comparative algorithms by a large margin., Comment: Selected for a spotlight presentation at NIPS, 2018. Camera ready version

Recently, reinforcement learning has been successfully applied to the logical game of Go, various Atari games, and even a 3D game, Labyrinth, though it continues to have problems in sparse reward settings. It is difficult to explore, but also difficult to exploit, a small number of successes when learning policy. To solve this issue, the subgoal and option framework have been proposed. However, discovering subgoals online is too expensive to be used to learn options in large state spaces. We propose Micro-objective learning (MOL) to solve this problem. The main idea is to estimate how important a state is while training and to give an additional reward proportional to its importance. We evaluated our algorithm in two Atari games: Montezuma's Revenge and Seaquest. With three experiments to each game, MOL significantly improved the baseline scores. Especially in Montezuma's Revenge, MOL achieved two times better results than the previous state-of-the-art model.

Catastrophic forgetting is a problem of neural networks that loses the information of the first task after training the second task. Here, we propose a method, i.e. incremental moment matching (IMM), to resolve this problem. IMM incrementally matches the moment of the posterior distribution of the neural network which is trained on the first and the second task, respectively. To make the search space of posterior parameter smooth, the IMM procedure is complemented by various transfer learning techniques including weight transfer, L2-norm of the old and the new parameter, and a variant of dropout with the old parameter. We analyze our approach on a variety of datasets including the MNIST, CIFAR-10, Caltech-UCSD-Birds, and Lifelog datasets. The experimental results show that IMM achieves state-of-the-art performance by balancing the information between an old and a new network., Comment: Accepted for NIPS 2017

Temporal hollowing after lateral facial skeletal exposure via a coronal approach is not uncommon. This is caused by injury to the superficial temporal fat pad. However, to prevent facial nerve injury at the zygomatic arch, incision of the superficial layer of the temporalis fascia (TF) and exposure of the temporal fat pad are accompanied with most coronal approaches. Here we introduce our method of dissection of the suprasuperficial layer of the TF without exposure of the temporal fat pad and report on its safety and efficacy. Temporal dissection just superficial to the superficial layer of the TF was performed to the superior border of the zygomatic arch. After identifying the zygomatic arch, an incision was made at the posterosuperior aspect of the arch, when exposure of the arch was necessary for access to the lateral skeleton. From 1989 to 2008, 522 patients undergoing craniofacial surgery via coronal approach were analyzed retrospectively. No patient developed temporal hollowing. However, 14 patients (2.7%) developed frontal weakness immediately postoperatively, and 2 of them had permanent paresis. The frequency of this motor morbidity was low compared with that of other reports. Preservation of the temporal fat pad in coronal approach is essential for the prevention of temporal hollowing, and for this, we recommend suprasuperficial layer of TF dissection.

Endoscopic variceal band ligation (EVL) is an effective procedure to control and prevent variceal bleeding in patients with liver cirrhosis, but it can be complicated by bleeding from post-EVL ulcers. Several studies have reported that proton pump inhibitors (PPIs) decrease the size of post-EVL ulcers. However, evidence are limited as to whether PPIs actually reduce the risk of bleeding after EVL. This study aimed to analyze the factors associated with bleeding after prophylactic EVL and to assess the effect of PPI therapy. Five hundred and five cirrhotic patients with high risk esophageal varices who received primary prophylactic EVL were included for this retrospective cohort study. Post-EVL bleeding was defined as bleeding after prophylactic EVL within 8 weeks evidenced by the occurrence of melena or hematemesis, or by a decrease of hemoglobin by >2.0 g/dL. If evidence of bleeding from ulceration of the EVL sites was confirmed by endoscopy, we defined it as post-EVL ulcer bleeding. Fourteen patients developed bleeding after prophylactic EVL. Factors associated with post-EVL bleeding included alcohol as etiology, low albumin, high total bilirubin, high Child-Pugh score, high MELD score, coexistence of gastric varices, and not administrating PPI medication by univariate analysis. In multivariate logistic analysis, Co-existing gastric varix (odds ratio [OR] 5.680, P = 0.005] and not administrating PPIs (OR 8.217, P = 0.002) were associated with bleeding after prophylactic EVL. In the subgroup analysis excluding patients whose gastric varices were treated, not administering PPI medication (OR 8.827, P = 0.008) was the sole factor associated with post-EVL bleeding. We suggest that PPI therapy needs to be considered in patients receiving prophylactic EVL to reduce the risk of bleeding after prophylactic EVL.

Amyloid fibrils have recently received much attention due to not only their important role in disease pathogenesis but also their excellent mechanical properties, which are comparable to those of mechanically strong protein materials such as spider silk. This indicates the necessity of understanding fundamental principles providing insight into how amyloid fibrils exhibit the excellent mechanical properties, which may allow for developing biomimetic materials whose material (e.g., mechanical) properties can be controlled. Here, we describe recent efforts to characterize the nanomechanical properties of amyloid fibrils using computational simulations (e.g., atomistic simulations) and single-molecule experiments (e.g., atomic force microscopy experiments). This paper summarizes theoretical models, which are useful in analyzing the mechanical properties of amyloid fibrils based on simulations and experiments, such as continuum elastic (beam) model, elastic network model, and polymer statistical model. In this paper, we suggest how the nanomechanical properties of amyloid fibrils can be characterized and determined using computational simulations and/or atomic force microscopy experiments coupled with the theoretical models.

Purpose Methods and materials Results Conclusion Personal information References, Purpose: On coronarycomputed tomography angiography (CCTA), we have knowing that mis-registration artifacts are caused by heart rate variation, patient's incomplete breath holding, patient'smotion during scan, ECG signal errors and so on [1-9]. Among those...

Deep neural networks continue to advance the state-of-the-art of image recognition tasks with various methods. However, applications of these methods to multimodality remain limited. We present Multimodal Residual Networks (MRN) for the multimodal residual learning of visual question-answering, which extends the idea of the deep residual learning. Unlike the deep residual learning, MRN effectively learns the joint representation from vision and language information. The main idea is to use element-wise multiplication for the joint residual mappings exploiting the residual learning of the attentional models in recent studies. Various alternative models introduced by multimodality are explored based on our study. We achieve the state-of-the-art results on the Visual QA dataset for both Open-Ended and Multiple-Choice tasks. Moreover, we introduce a novel method to visualize the attention effect of the joint representations for each learning block using back-propagation algorithm, even though the visual features are collapsed without spatial information., Comment: 13 pages, 7 figures, accepted for NIPS 2016

Medical dissolution of gallstone is usually performed on radiolucent gallstones in a functioning gallbladder. However, absence of visible gallstone on plain abdominal x-ray does not always preclude calcification. This study aims to compare the response and dissolution rates between ursodeoxycholic acid (UDCA) alone or in combination with chenodeoxycholic acid (CDCA) according to stone density on computed tomography (CT) scan. A total of 126 patients underwent dissolution therapy with either UDCA alone or combination of CDCA and UDCA (CNU) from December 2010 to March 2014 at Korea University Ansan Hospital. In the end, 81 patients (CNU group = 44, UDCA group = 37) completed dissolution therapy for 6 months. Dissolution rate (percentage reduction in the gallstone volume) and response to therapy (complete dissolution or partial dissolution defined as reduction in stone volume of >50%) were compared between the 2 groups. Dissolution and response rates of sludge was also compared between the 2 groups. The overall response rate was 50.6% (CNU group 43.2% vs UDCA group 59.5%, P = 0.14), and the overall dissolution rate was 48.34% (CNU group 41.5% vs UDCA group 56.5%, P = 0.13). When analyzed according to stone density, response rate was 33.3%, 87.1%, 30.0%, and 6.2% for hypodense, isodense, hyperdense, and calcified stones, respectively. Response rate (85.7% vs 88.2%, P = 0.83) and dissolution rate (81.01% vs 85.38%, P = 0.17) of isodense stones were similar between CNU and UDCA group. When only sludge was considered, the overall response rate was 87.5% (CNU group 71.4% vs UDCA group 94.1%, P = 0.19), and the overall dissolution rate was 85.42% (CNU group 67.9% vs UDCA group 92.7%, P = 0.23). Patients with isodense gallstones and sludge showed much better response to dissolution therapy with CNU and UDCA showing comparable efficacy. Therefore, CT scan should be performed before medication therapy if stone dissolution is intended.

상선승무원들의 승선근무로 인한 질병 발생 실태를 조사하여 승선근무로 인해 발생되는 질병예방 및 건강증진을 위한 기초 자료를 제공하기 위한 목적으로 1049명의 상선승무원들을 대상으로 설문 및 면접을 통해 조사 분석한 결과는 다음과 같다. 최근 12개월 동안 승선 중 당직근무에 지장을 받을 정도의 질병을 경험한 선원은 69.0%였으며, 질병경험분포에 유의성을 나타낸 변수는 연령(p 무좀 6.6%>위궤양 6.4%>외상 5.3% 순이었다. 질병군별 질병발생은 근골격계질환이 17.8%로 가장 많았고 구강계질환 13.6%> 피부계질환 12.4%> 소화계질환 12.1% 순이었으며, 발생 질병의 불편기간은 31일 이상이 35.7%, 입원기간과 치료기간은 각각 7일 이하가 50.2%, 42.8%였고 의료시설 이용은 의원급이 27.9%로 가장 높게 조사되었다. 【The purpose of this study was to find out morbidity rate and pattern of disease and affect of variables related disease and medical management of seafares'. The subjects this study were 1049 seafares' were took education in Korea Institute of Maritime and Fisheries Technology. This questionnaire was focused on finding the basic data for prevention of disease and promotion health for the seafares'. The collected data were analyzed by using descriptive statistics, Chi-square, cross tab, linear regression by SPSS 10.1 package. The result of this study are as follow. 1)The morbidity rate within recent 12 months was 69.0%. 2)there were significant differences qf occurred disease in age(p tinea 6.6%> gastric ulcer 6.4, and musculoskeletal disease group were 20.9%, which revealed the highest rate oral disease 13.6%> skin disease> 12.4%, digestive disease> 12.1%. 4) As refer to medical management, The pain above 31days 35.7%, hospitalization and treatment below 7days were each 50.2%, 42.8%, medical service were doctor's office 27.9, which revealed the highest rate.】

The purpose of this study was to find out morbidity rate and pattern of occupational disease and affect of variables related disease of Fishing crew The subjects this study were 624 Fishing crew who took education course in Korea Institute of Maritime and Fisheries Technology. This questionnaire was focused on finding the basic data for prevention of occupational disease and promotion health on the Fishing crew The collected data were analyzed by using descriptive statistics, Chi-square, cross tab by SPSS package. The result of this study are as follow. 1)The morbidity rate within recent 12 months was 70.2%. there were significant differences of occurred disease in age, rank, perceived health status, fatigue symptoms, rest time, workload, overtime duration 2) the more aging(p

A poly(1-hexyl-3,4-dimethyl-2,5-pyrrolylene) (PHDP) was prepared and its luminescence in tetrahydrofuran (THF) was studied. When PHDP is excited by UV light, it produces very strong blue luminescence. The quantum yield of PHDP (Q = 36.9%) is much greater than that of the monomer, 1-hexyl-3,4-dimethylpyrrole (HDP) with Q = 0.61%. The principal luminescence of PHDP has a single decay component with ca. 1 ns, whereas the decay of HDP is complicated. The molecular structure and conformational behavior of HDP and the oligomers up to trimer have been also determined by ab initio Hartree-Fock (HF/6-31G**), density functional theory (DFT-B3LYP/6-31G**), and semiempirical (ZINDO) methods. According to the results of calculations, it is proposed that the enhanced quantum yield of the polymer PHDP results mostly from the πconjugation between neighboring pyrrole rings.

Introduction Theory The Pattern of American Dominance in Japan Methodology Government Policies and Media Diffusion Conclusion

The aim of the present study is to establish a bacterial clone capable of secreting an integrin alpha v beta 3 targeting probe with bioluminescent and fluorescent activities, and to verify its specific targeting and optical activities using molecular imaging. A bacterial vector expressing a fusion of secretory Gaussia luciferase (sGluc), mCherry, and RGD (sGluc-mCherry-RGDX3; GCR), and a control vector expressing a fusion of secretory Gaussia luciferase and mCherry (sGluc-mCherry; GC) were constructed. The GCR and GC proteins were expressed in E. coli and secreted into the growth medium, which showed an approximately 10-fold higher luciferase activity than the bacterial lysate. Successful purification of GCR and GC was achieved using the 6X His-tag method. The GCR protein bound with higher affinity to U87MG cells than CHO cells in confocal microscopy and IVIS imaging, and also showed a high affinity for integrin alpha v beta 3 expressing tumor xenografts in an in vivo animal model. An E. coli clone was established to secrete an integrin alpha v beta 3 targeting imaging probe with bioluminescent and fluorescent activities. The probe was produced feasibly and at low cost, and has shown to be useful for the assessment of angiogenesis in vitro and in vivo.

In sagittal split ramus osteotomy (SSRO) surgery, variable methods are used to maintain the proximal segment in the anatomic position during fixation to minimize early built-in relapse. However, there is still a chance of condyle malposition.In this study, a new titanium plate, which has a sliding oval hole in its distal part, was used to enable postoperative mobility of the proximal segment. This plate makes it possible to reposition the condyle during the postoperative intermaxillary fixation period.These plates were used successfully in 23 patients who received SSRO setback surgery. All of the patients were followed up without any complications, including early relapse or temporomandibular joint problems.The authors used a new plate that enables sliding of the proximal segment after SSRO, which can minimize early relapse without any postoperative complications.

IPsec protocol[1] is a set of security extensions developed by the IETF and it provides privacy and authentication services at the IP layer by using modern cryptography. In this paper, we describe both of H/W and S/W architectures of our router system, SRS-10. The system is designed to support high performance routing and IPsec VPN. Especially, we used Cavium-s CN2560 processor to implement IPsec processing in inline-mode., {"references":["\"Security Architecture for the Internet Protocol,\" RFC 2401, 1998","Eric J. Johnson and Aaron R. Kunze, \"IXP2400/2800 Programming,\"\nIntel Press, ch 2","\"Nitrox-II Security Processor Hardware Manual,\" Cavium Networks,\nDoc:CN2xxx-HM","Bill Carlson, \"Intel Internet Exchange Architecture and Applications,\"\nIntel Press, ch 1"]}