895 Hits in 1.8 sec

Towards Interrogating Discriminative Machine Learning Models [article]

Wenbo Guo, Kaixuan Zhang, Lin Lin, Sui Huang, Xinyu Xing
2017 arXiv   pre-print
It is oftentimes impossible to understand how machine learning models reach a decision. While recent research has proposed various technical approaches to provide some clues as to how a learning model makes individual decisions, they cannot provide users with ability to inspect a learning model as a complete entity. In this work, we propose a new technical approach that augments a Bayesian regression mixture model with multiple elastic nets. Using the enhanced mixture model, we extract
more » ... ons for a target model through global approximation. To demonstrate the utility of our approach, we evaluate it on different learning models covering the tasks of text mining and image recognition. Our results indicate that the proposed approach not only outperforms the state-of-the-art technique in explaining individual decisions but also provides users with an ability to discover the vulnerabilities of a learning model.
arXiv:1705.08564v1 fatcat:5amwesebevcdxa25vso6whv3qm

Are Shortest Rationales the Best Explanations for Human Understanding? [article]

Hua Shen, Tongshuang Wu, Wenbo Guo, Ting-Hao 'Kenneth' Huang
2022 arXiv   pre-print
Existing self-explaining models typically favor extracting the shortest possible rationales - snippets of an input text "responsible for" corresponding output - to explain the model prediction, with the assumption that shorter rationales are more intuitive to humans. However, this assumption has yet to be validated. Is the shortest rationale indeed the most human-understandable? To answer this question, we design a self-explaining model, LimitedInk, which allows users to extract rationales at
more » ... y target length. Compared to existing baselines, LimitedInk achieves compatible end-task performance and human-annotated rationale agreement, making it a suitable representation of the recent class of self-explaining models. We use LimitedInk to conduct a user study on the impact of rationale length, where we ask human judges to predict the sentiment label of documents based only on LimitedInk-generated rationales with different lengths. We show rationales that are too short do not help humans predict labels better than randomly masked text, suggesting the need for more careful design of the best human rationales.
arXiv:2203.08788v1 fatcat:d7mtoqzg6rgazjwb434y2qckc4

MuCAN: Multi-Correspondence Aggregation Network for Video Super-Resolution [article]

Wenbo Li, Xin Tao, Taian Guo, Lu Qi, Jiangbo Lu, Jiaya Jia
2020 arXiv   pre-print
Video super-resolution (VSR) aims to utilize multiple low-resolution frames to generate a high-resolution prediction for each frame. In this process, inter- and intra-frames are the key sources for exploiting temporal and spatial information. However, there are a couple of limitations for existing VSR methods. First, optical flow is often used to establish temporal correspondence. But flow estimation itself is error-prone and affects recovery results. Second, similar patterns existing in
more » ... images are rarely exploited for the VSR task. Motivated by these findings, we propose a temporal multi-correspondence aggregation strategy to leverage similar patches across frames, and a cross-scale nonlocal-correspondence aggregation scheme to explore self-similarity of images across scales. Based on these two new modules, we build an effective multi-correspondence aggregation network (MuCAN) for VSR. Our method achieves state-of-the-art results on multiple benchmark datasets. Extensive experiments justify the effectiveness of our method.
arXiv:2007.11803v1 fatcat:uv73kpxvqzbb3hnv6kmd7funra

Visualizing and Understanding Deep Neural Networks in CTR Prediction [article]

Lin Guo, Hui Ye, Wenbo Su, Henhuan Liu, Kai Sun, Hang Xiang
2018 arXiv   pre-print
Although deep learning techniques have been successfully applied to many tasks, interpreting deep neural network models is still a big challenge to us. Recently, many works have been done on visualizing and analyzing the mechanism of deep neural networks in the areas of image processing and natural language processing. In this paper, we present our approaches to visualize and understand deep neural networks for a very important commercial task--CTR (Click-through rate) prediction. We conduct
more » ... eriments on the productive data from our online advertising system with daily varying distribution. To understand the mechanism and the performance of the model, we inspect the model's inner status at neuron level. Also, a probe approach is implemented to measure the layer-wise performance of the model. Moreover, to measure the influence from the input features, we calculate saliency scores based on the back-propagated gradients. Practical applications are also discussed, for example, in understanding, monitoring, diagnosing and refining models and algorithms.
arXiv:1806.08541v1 fatcat:gixzvp7irzbqrbeogctmtmbili


Wenbo Guo, Lang Lin, Hongzhi Zhao, Youxi Tang
2021 Scientia Sinica Informationis  
doi:10.1360/ssi-2021-0160 fatcat:4ourppsdpbce7pkcd2sk5o3r4i

Explaining Deep Learning Models - A Bayesian Non-parametric Approach [article]

Wenbo Guo and Sui Huang and Yunzhe Tao and Xinyu Xing and Lin Lin
2018 arXiv   pre-print
Understanding and interpreting how machine learning (ML) models make decisions have been a big challenge. While recent research has proposed various technical approaches to provide some clues as to how an ML model makes individual predictions, they cannot provide users with an ability to inspect a model as a complete entity. In this work, we propose a novel technical approach that augments a Bayesian non-parametric regression mixture model with multiple elastic nets. Using the enhanced mixture
more » ... odel, we can extract generalizable insights for a target model through a global approximation. To demonstrate the utility of our approach, we evaluate it on different ML models in the context of image recognition. The empirical results indicate that our proposed approach not only outperforms the state-of-the-art techniques in explaining individual decisions but also provides users with an ability to discover the vulnerabilities of the target ML models.
arXiv:1811.03422v1 fatcat:d57foj5lubeuxmnqqwg7w4l6ku

Advances in the Chemical Looping Ammonia Synthesis

Sheng Feng, Wenbo Gao, Hujun Cao, Jianping Guo, Ping Chen
2020 Huaxue xuebao  
Ammonia is not only the main raw material of nitrogen fertilizer, but also a promising energy carrier for the storage and utilization of renewable energy. The fossil fuel-based Haber-Bosch ammonia synthesis industry is an energy-consuming and high CO 2 -emission process. For the sustainable growth of human society, it is critically important to develop "green" ammonia synthesis processes driven by renewable energies. This scenario motivates growing interests on ammonia synthesis via
more » ... s catalysis, electro-chemical and photo-chemical routes as well as chemical looping process. Chemical looping ammonia synthesis (CLAS) process involves a series of individual reactions which produce ammonia in a distinctly different manner to the catalytic process. The CLAS could be operated under ambient pressure, and the switching on/off operation is flexible. Therefore, CLAS may be more amenable to variable and intermittent operation compared to the conventional catalytic process. More importantly, the competitive adsorption of N 2 and H 2 or H 2 O in the catalytic process can be circumvented to a great extent, which opens new opportunities for the design and development of nitrogen carriers especially for low-temperature ammonia production. Because of these unique features, the application of chemical looping technology for ammonia synthesis has been received increasing attention in recent years. The development of high-efficiency nitrogen carriers is the key component for the implementation of CLAS. A wide range of materials including metal nitrides, metal imides, nitride-hydrides and oxynitrides have been evaluated as nitrogen carriers for CLAS. The knowledge accumulated during the past decade will no doubt beneficial for the further optimization and development of nitrogen carriers. This article reviews the research progress in the field of chemical looping ammonia synthesis in recent years, with the focuses on the materials development of nitrogen carriers in CLAS. Furthermore, the challenges and future directions of CLAS are also discussed. With the development of nitrogen carriers and process design, CLAS would potentially play an important role in the green ammonia synthesis as well as the future energy system.
doi:10.6023/a20060207 fatcat:cwhzqdingjhd7ku4ru6jj7yeme


Wenbo Guo, Xueqin Liu, Long Li, Dongsheng Deng
2009 Acta Crystallographica Section E  
D-HÁ Á ÁA D -H HÁ Á ÁA D Á Á ÁA D -HÁ Á ÁA C14-H14BÁ Á ÁO2 i 0. 2-Iodo-3-methoxy-6-methylpyridine Wenbo Guo, Xueqin Liu, Long Li and Dongsheng Deng S1.  ... 
doi:10.1107/s1600536809050739 pmid:21578963 pmcid:PMC2971865 fatcat:opzrfo4o7vbhvorrtwqsdvwypy

RSS-based Multiple Sources Localization with Unknown Log-normal Shadow Fading [article]

Yueyan Chu, Wenbin Guo, Kangyong You, Lei Zhao, Tao Peng, Wenbo Wang
2021 arXiv   pre-print
Multi-source localization based on received signal strength (RSS) has drawn great interest in wireless sensor networks. However, the shadow fading term caused by obstacles cannot be separated from the received signal, which leads to severe error in location estimate. In this paper, we approximate the log-normal sum distribution through Fenton-Wilkinson method to formulate a non-convex maximum likelihood (ML) estimator with unknown shadow fading factor. In order to overcome the difficulty in
more » ... ing the non-convex problem, we propose a novel algorithm to estimate the locations of sources. Specifically, the region is divided into N grids firstly, and the multi-source localization is converted into a sparse recovery problem so that we can obtain the sparse solution. Then we utilize the K-means clustering method to obtain the rough locations of the off-grid sources as the initial feasible point of the ML estimator. Finally, an iterative refinement of the estimated locations is proposed by dynamic updating of the localization dictionary. The proposed algorithm can efficiently approach a superior local optimal solution of the ML estimator. It is shown from the simulation results that the proposed method has a promising localization performance and improves the robustness for multi-source localization in unknown shadow fading environments. Moreover, the proposed method provides a better computational complexity from O(K^3N^3) to O(N^3).
arXiv:2110.10435v1 fatcat:jywqgycgqnbnzlgc7fyblejfbq


Zongyan Zhao, Wenbo Wang, Zhenhua Gao, Mingruo Guo
2011 BioResources  
Whey protein is a renewable and environmentally safe biomaterial, a by-product of cheese production. It can be utilized for non-food applications for value-added products. The substances glyoxal (GO), glutaraldehyde (GA), polymeric methylene biphenyl diisocyanate (p-MDI), urea-formaldehyde (UF) resin, and phenol-formaldehyde oligomer (PFO) that contain reactive groups were applied together with whey protein as modifier in order to increase crosslinking density and molecular weight for improving
more » ... the bond strength and water resistance of whey protein. A water-resistant and environmentally safe whey protein-based wood adhesive for plywood was developed by evaluating the effects of these modifiers on the bond strength, bond durability, and free formaldehyde emission of the resulting plywood panels. Results of FTIR and SEM analyses and bond evaluation indicated that GO, GA, and p-MDI were not suitable to modify whey proteins due to their high reactivity with whey proteins, causing phase separation. UF resin was not a good modifier for whey proteins because of either its poor water-resistance or higher emission of hazardous formaldehyde. Whey protein adhesives modified with PFO had a dry shear bond strength of 1.98 MPa and a 28h-boiling-dry-boiling wet shear strength of 1.73 MPa, which were both much higher than the required values for structural use according to standard JIS K6806-2003, while its formaldehyde emission was 0.067mg/L, much lower than the required value for green plywood according to standard JIS A5908.
doaj:b84e9c0757dd48ada889c2e0c51affd7 fatcat:nqetkvednfgyzhu433piesgiru

Identification of biomarkers of venous thromboembolism by bioinformatics analyses

Guiming Wang, Wenbo Zhao, Yudong Yang, Gaochao Yang, Zhigang Wei, Jiansheng Guo
2018 Medicine  
Venous thromboembolism (VTE) is a common vascular disease and a major cause of mortality. This study intended to explore the biomarkers associated with VTE by bioinformatics analyses. Based on Gene Expression Omnibus (GEO) database, the GSE19151 expression profile data were downloaded. The differentially expressed genes (DEGs) between single VTE (sVTE)/recurrent VTE (rVTE) and control were identified. Then, pathway enrichment analysis of DEGs were performed, followed by protein-protein
more » ... on (PPI) network construction. Total 433 upregulated and 222 downregulated DEGs were obtained between sVTE and control samples. For rVTE versus control, 625 upregulated and 302 downregulated DEGs were identified. The overlap DEGs were mainly enriched in the pathways related to ribosome, cancer, and immune disease. The DEGs specific to rVTE were enriched in several pathways, such as nod-like receptor signaling pathway. In the PPI network, 2 clusters of VTE genes, including ribosomal protein family genes and NADH family-ubiquinolcytochrome genes, were identified, such as ribosomal protein L9 (RPL9), RPL5, RPS20, RPL23, and tumor protein p53 (TP53). The nod-like receptor signaling pathway, ribosomal protein family genes, such as RPL9, RPL5, RPS20, and RPL23, and DEG of TP53 may have the potential to be used as targets for diagnosis and treatment of VTE. Abbreviations: DEG = differentially expressed gene, GEO = Gene Expression Omnibus, IL = interleukin, PPI = protein-protein interaction, RPL9 = ribosomal protein L9, rVTE = recurrent venous thromboembolism, sVTE = single venous thromboembolism, TP53 = tumor protein p53, VTE = venous thromboembolism. Editor: Ovidiu Constantin Baltatu. GW and WZ are first coauthors. The authors have no funding and conflicts of interest to disclose.
doi:10.1097/md.0000000000010152 pmid:29620629 pmcid:PMC5902267 fatcat:cs7qclkrsjae3o6tugdm2q2xma

A Spontaneous Driver Emotion Facial Expression (DEFE) Dataset for Intelligent Vehicles [article]

Wenbo Li, Yaodong Cui, Yintao Ma, Xingxin Chen, Guofa Li, Gang Guo, Dongpu Cao
2020 arXiv   pre-print
Wenbo Li received the B.S. and M.Sc. degree in automotive engineering from Chongqing University, Chongqing, China, in 2014, and 2017, respectively.  ... 
arXiv:2005.08626v1 fatcat:mzitvppzobcaddrylijsxopulq

Mechanical Properties and Thermal Conductivity of Ytterbium-Silicate-Mullite Composites

Jie Xiao, Wenbo Chen, Liangliang Wei, Wenting He, Hongbo Guo
2020 Materials  
Various Ytterbium-Silicate-Mullite composites were successfully fabricated by adding Yb2SiO5 into mullite ceramics and then using pressureless sintering at 1550 °C. The influence of Yb2SiO5 addition on the microstructure, mechanical properties, and thermal conductivity of ytterbium-silicate-mullite composites was investigated. Results show that the composites mainly consisted of a mullite matrix and some in situ formed Yb2Si2O7 and Al2O3 phases. By the addition of Yb2SiO5, the Vickers hardness
more » ... f composites reached ~9.0 at an additive concentration of 5 mol %. Fracture toughness increased to ~2.7 MPa·m1/2 at the additive concentration of 15 mol %, owing to the trans-granular fracture and crack deflection of the pinning effect of the Al2O3 phase in the composites. With the increase of the Al2O3 phase in the composite, the thermal conductivity for the 15YbAM reached around 4.0 W/(m·K) at 1200 °C. Although the thermal conductivity increased, it is still acceptable for such composites to be used as environmental barrier coatings.
doi:10.3390/ma13030671 pmid:32028582 pmcid:PMC7040839 fatcat:o4zxusroujdkvmiujgkgvztzby

Dynamical manipulation of electromagnetic polarization using anisotropic meta-mirror

Jianhua Cui, Cheng Huang, Wenbo Pan, Mingbo Pu, Yinghui Guo, Xiangang Luo
2016 Scientific Reports  
Polarization control of electromagnetic wave is very important in many fields. Here, we propose an active meta-mirror to dynamically manipulate electromagnetic polarization state at a broad band. This meta-mirror is composed of a double-layered metallic pattern backed by a metallic flat plate, and the active elements of PIN diodes are integrated into the meta-atom to control the reflection phase difference between two orthogonal polarization modes. Through switching the operating state of the
more » ... N diodes, the meta-mirror is expected to achieve three polarization states which are left-handed, right-handed circular polarizations and linear polarization, respectively. We fabricated this active metamirror and validated its polarization conversion performance by measurement. The linearly polarized incident wave can be dynamically converted to right-handed or left-handed circular polarization in the frequency range between 3.4 and 8.8 GHz with the average loss of 1 dB. Furthermore, it also can keep its initial linear polarization state.
doi:10.1038/srep30771 pmid:27469028 pmcid:PMC4965767 fatcat:6hchvmrkcnhuzjbanraf3i5fqy

DANCE: Enhancing saliency maps using decoys [article]

Yang Lu, Wenbo Guo, Xinyu Xing, William Stafford Noble
2021 arXiv   pre-print
Saliency methods can make deep neural network predictions more interpretable by identifying a set of critical features in an input sample, such as pixels that contribute most strongly to a prediction made by an image classifier. Unfortunately, recent evidence suggests that many saliency methods poorly perform, especially in situations where gradients are saturated, inputs contain adversarial perturbations, or predictions rely upon inter-feature dependence. To address these issues, we propose a
more » ... ramework that improves the robustness of saliency methods by following a two-step procedure. First, we introduce a perturbation mechanism that subtly varies the input sample without changing its intermediate representations. Using this approach, we can gather a corpus of perturbed data samples while ensuring that the perturbed and original input samples follow the same distribution. Second, we compute saliency maps for the perturbed samples and propose a new method to aggregate saliency maps. With this design, we offset the gradient saturation influence upon interpretation. From a theoretical perspective, we show the aggregated saliency map could not only capture inter-feature dependence but, more importantly, robustify interpretation against previously described adversarial perturbation methods. Following our theoretical analysis, we present experimental results suggesting that, both qualitatively and quantitatively, our saliency method outperforms existing methods.
arXiv:2002.00526v3 fatcat:zf3m364hxnhlna55u2g7irzuwm
« Previous Showing results 1 — 15 out of 895 results