scholarly journals PCONV: The Missing but Desirable Sparsity in DNN Weight Pruning for Real-Time Execution on Mobile Devices

2020 ◽  
Vol 34 (04) ◽  
pp. 5117-5124 ◽  
Author(s):  
Xiaolong Ma ◽  
Fu-Ming Guo ◽  
Wei Niu ◽  
Xue Lin ◽  
Jian Tang ◽  
...  
Keyword(s):  
Neural Network ◽  
Real Time ◽  
Mobile Devices ◽  
Large Scale ◽  
Coarse Grained ◽  
Convolution Kernel ◽  
Fine Grained ◽  
Model Compression ◽  
Weight Pruning ◽  
Balanced Workload

Model compression techniques on Deep Neural Network (DNN) have been widely acknowledged as an effective way to achieve acceleration on a variety of platforms, and DNN weight pruning is a straightforward and effective method. There are currently two mainstreams of pruning methods representing two extremes of pruning regularity: non-structured, fine-grained pruning can achieve high sparsity and accuracy, but is not hardware friendly; structured, coarse-grained pruning exploits hardware-efficient structures in pruning, but suffers from accuracy drop when the pruning rate is high. In this paper, we introduce PCONV, comprising a new sparsity dimension, – fine-grained pruning patterns inside the coarse-grained structures. PCONV comprises two types of sparsities, Sparse Convolution Patterns (SCP) which is generated from intra-convolution kernel pruning and connectivity sparsity generated from inter-convolution kernel pruning. Essentially, SCP enhances accuracy due to its special vision properties, and connectivity sparsity increases pruning rate while maintaining balanced workload on filter computation. To deploy PCONV, we develop a novel compiler-assisted DNN inference framework and execute PCONV models in real-time without accuracy compromise, which cannot be achieved in prior work. Our experimental results show that, PCONV outperforms three state-of-art end-to-end DNN frameworks, TensorFlow-Lite, TVM, and Alibaba Mobile Neural Network with speedup up to 39.2 ×, 11.4 ×, and 6.3 ×, respectively, with no accuracy loss. Mobile devices can achieve real-time inference on large-scale DNNs.

scholarly journals Attentive User-Engaged Adversarial Neural Network for Community Question Answering

2020 ◽  
Vol 34 (05) ◽  
pp. 9322-9329
Author(s):  
Yuexiang Xie ◽  
Ying Shen ◽  
Yaliang Li ◽  
Min Yang ◽  
Kai Lei
Keyword(s):  
Neural Network ◽  
Large Scale ◽  
Question Answering ◽  
Selection Process ◽  
Coarse Grained ◽  
User Engagement ◽  
Fine Grained ◽  
Community Question Answering ◽  
Questions And Answers ◽  
Real World Datasets

We study the community question answering (CQA) problem that emerges with the advent of numerous community forums in the recent past. The task of finding appropriate answers to questions from informative but noisy crowdsourced answers is important yet challenging in practice. We present an Attentive User-engaged Adversarial Neural Network (AUANN), which interactively learns the context information of questions and answers, and enhances user engagement with the CQA task. A novel attentive mechanism is incorporated to model the semantic internal and external relations among questions, answers and user contexts. To handle the noise issue caused by introducing user context, we design a two-step denoise mechanism, including a coarse-grained selection process by similarity measurement, and a fine-grained selection process by applying an adversarial training module. We evaluate the proposed method on large-scale real-world datasets SemEval-2016 and SemEval-2017. Experimental results verify the benefits of incorporating user information, and show that our proposed model significantly outperforms the state-of-the-art methods.

scholarly journals ShadingNet: Image Intrinsics by Fine-Grained Shading Decomposition

2021 ◽  
Author(s):  
Anil S. Baslamisli ◽  
Partha Das ◽  
Hoang-An Le ◽  
Sezer Karaoglu ◽  
Theo Gevers
Keyword(s):  
Neural Network ◽  
Large Scale ◽  
State Of The Art ◽  
Image Decomposition ◽  
Natural Environments ◽  
Decomposition Algorithms ◽  
Ambient Light ◽  
Fine Grained ◽  
Large Scale Dataset ◽  
Direct Illumination

AbstractIn general, intrinsic image decomposition algorithms interpret shading as one unified component including all photometric effects. As shading transitions are generally smoother than reflectance (albedo) changes, these methods may fail in distinguishing strong photometric effects from reflectance variations. Therefore, in this paper, we propose to decompose the shading component into direct (illumination) and indirect shading (ambient light and shadows) subcomponents. The aim is to distinguish strong photometric effects from reflectance variations. An end-to-end deep convolutional neural network (ShadingNet) is proposed that operates in a fine-to-coarse manner with a specialized fusion and refinement unit exploiting the fine-grained shading model. It is designed to learn specific reflectance cues separated from specific photometric effects to analyze the disentanglement capability. A large-scale dataset of scene-level synthetic images of outdoor natural environments is provided with fine-grained intrinsic image ground-truths. Large scale experiments show that our approach using fine-grained shading decompositions outperforms state-of-the-art algorithms utilizing unified shading on NED, MPI Sintel, GTA V, IIW, MIT Intrinsic Images, 3DRMS and SRD datasets.

scholarly journals Innovative Deep Neural Network Modeling for Fine-Grained Chinese Entity Recognition

Electronics ◽  
2020 ◽  
Vol 9 (6) ◽  
pp. 1001 ◽  
Author(s):  
Jingang Liu ◽  
Chunhe Xia ◽  
Haihua Yan ◽  
Wenjing Xu
Keyword(s):  
Neural Network ◽  
Language Processing ◽  
Short Term Memory ◽  
Named Entity Recognition ◽  
Training Model ◽  
Entity Recognition ◽  
Coarse Grained ◽  
Neural Network Modeling ◽  
Fine Grained ◽  
Named Entity

Named entity recognition (NER) is a basic but crucial task in the field of natural language processing (NLP) and big data analysis. The recognition of named entities based on Chinese is more complicated and difficult than English, which makes the task of NER in Chinese more challenging. In particular, fine-grained named entity recognition is more challenging than traditional named entity recognition tasks, mainly because fine-grained tasks have higher requirements for the ability of automatic feature extraction and information representation of deep neural models. In this paper, we propose an innovative neural network model named En2BiLSTM-CRF to improve the effect of fine-grained Chinese entity recognition tasks. This proposed model including the initial encoding layer, the enhanced encoding layer, and the decoding layer combines the advantages of pre-training model encoding, dual bidirectional long short-term memory (BiLSTM) networks, and a residual connection mechanism. Hence, it can encode information multiple times and extract contextual features hierarchically. We conducted sufficient experiments on two representative datasets using multiple important metrics and compared them with other advanced baselines. We present promising results showing that our proposed En2BiLSTM-CRF has better performance as well as better generalization ability in both fine-grained and coarse-grained Chinese entity recognition tasks.

Microstructural Evolution of Synthetic Forsterite Aggregates Deformed to High Strain

2004 ◽  
Vol 467-470 ◽  
pp. 579-584 ◽  
Author(s):  
A. Kellermann Slotemaker ◽  
J.H.P. de Bresser ◽  
C.J. Spiers ◽  
M.R. Drury
Keyword(s):  
Grain Size ◽  
Dynamic Recrystallization ◽  
Grain Growth ◽  
Microstructural Evolution ◽  
Large Scale ◽  
High Strain ◽  
Confining Pressure ◽  
Size Reduction ◽  
Coarse Grained ◽  
Fine Grained

Microstructures provide the crucial link between solid state flow of rock materials in the laboratory and large-scale tectonic processes in nature. In this context, microstructural evolution of olivine aggregates is of particular importance, since this material controls the flow of the Earth’s upper mantle and affects the dynamics of the outer Earth. From previous work it has become apparent that if olivine rocks are plastically deformed to high strain, substantial weakening may occur before steady state mechanical behaviour is approached. This weakening appears directly related to progressive modification of the grain size distribution through competing effects of dynamic recrystallization and syn-deformational grain growth. However, most of our understanding of these processes in olivine comes from tests on coarse-grained materials that show grain size reduction through dynamic recrystallization. In the present study we focused on fine-grained (~1 µm) olivine aggregates (i.e., forsterite/Mg2SiO4), containing ~0.5 wt% water and 10 vol% enstatite (MgSiO3), Samples were axially compressed to varying strains up to a maximum of ~45%, at 600 MPa confining pressure and a temperature of 950°C. Microstructures were characterized by analyzing full grain size distributions and textures using SEM/EBSD. We observed syndeformational grain growth rather than grain size reduction, and relate this to strain hardening seen in the stress-strain curves.

scholarly journals Who Likes What? — SplitLBI in Exploring Preferential Diversity of Ratings

2020 ◽  
Vol 34 (01) ◽  
pp. 262-269
Author(s):  
Qianqian Xu ◽  
Jiechao Xiong ◽  
Zhiyong Yang ◽  
Xiaochun Cao ◽  
Qingming Huang ◽  
...  
Keyword(s):  
Large Scale ◽  
Social Preference ◽  
Learning To Rank ◽  
User Preferences ◽  
Coarse Grained ◽  
Fast Analysis ◽  
Fine Grained ◽  
Level Model ◽  
The Common ◽  
Multi Level

In recent years, learning user preferences has received significant attention. A shortcoming of existing learning to rank work lies in that they do not take into account the multi-level hierarchies from social choice to individuals. In this paper, we propose a multi-level model which learns both the common preference or utility function over the population based on features of alternatives to-be-compared, and preferential diversity functions conditioning on user categories. Such a multi-level model, enables us to simultaneously learn a coarse-grained social preference function together with a fine-grained personalized diversity. It provides us prediction power for the choices of new users on new alternatives. The key algorithm in this paper is based on Split Linearized Bregman Iteration (SplitLBI) algorithm which generates a dynamic path from the common utility to personalized preferential diversity, at different levels of sparsity on personalization. A synchronized parallel version of SplitLBI is proposed to meet the needs of fast analysis of large-scale data. The validity of the methodology are supported by experiments with both simulated and real-world datasets such as movie and dining restaurant ratings which provides us a coarse-to-fine grained preference learning.

scholarly journals Paleoenvironment and taphonomy of the dinosaur-bearing Ischigualasto Formation (Upper Triassic, Argentina)

1992 ◽  
Vol 6 ◽  
pp. 249-249 ◽  
Author(s):  
Raymond R. Rogers ◽  
Catherine A. Forster ◽  
Cathleen L. May ◽  
Alfredo Monetta ◽  
Paul C. Sereno
Keyword(s):  
Large Scale ◽  
Lower Half ◽  
Coarse Grained ◽  
Fine Grained ◽  
Karoo Basin ◽  
Sand Body ◽  
Faunal Succession ◽  
New Material ◽  
Group A ◽  
Depositional Setting

The oldest-known dinosaurs (Herrerasaurus, Pisanosaurus) occur within the Ischigualasto Formation. Recent work in the formation has brought to light significant new material, including the complete skeleton of a new primitive dinosaur. We sketch below the paleoenvironment and faunal succession during the range of these early dinosaurs, and review some of the taphonomic factors that shaped their fossil record.The Ischigualasto Formation (Carnian?) is included within the Agua de la Peña Group, a series of continental Triassic deposits exposed in the Ischigualasto-Ville Union Basin of northwest Argentina. Ischigualasto sediments rest unconformably upon the carbonaceous fluvial/lacustrine Los Rastros Formation; this contact is characterized locally by marked angular discordance. The upper contact is gradational into red-beds of the Los Colorados Formation. Medium- to coarse-grained conglomeratic sandstones, siltstones, and silty mudstones dominate the section. Sand bodies are characterized by medium- to large-scale trough cross-stratification and broad lenticular/narrow sheet geometries, and are interpreted as deposits of shallow, low-sinuosity streams. Siltstones and mudstones show pervasive evidence of soil development, including root traces, nodular caliche horizons, and pedogenic slickensides. Deposits attributable to lacustrine/paludal sedimentation are scarce, and freshwater vertebrates and invertebrates are extremely rare. These data suggest an upland depositional setting on a low-relief alluvial plain with seasonal climate.The Ischigualasto vertebrate fauna includes archosaurs, rhynchosaurs, traversodontid and carnivorous cynodonts, and temnospondyl amphibians. Rhynchosaurs dominate (relative specimen abundance) in the lower half of the section, but are absent from the upper half. Traversodontid cynodonts occur throughout the formation, but are much more abundant up-section. Archosaurs, carnivorous cynodonts, and particularly temnospondyls are rare throughout, with dinosaurs limited to the lower half. No major stratigraphic or sedimentologic changes occur up-section, and there is no evidence for significant shifts in physical or chemical taphonomic processes. Thus, trends in relative taxon abundance likely record a true biotic signal (e.g., local extinction, immigration) rather than a taphonomically-driven preservational bias.Fossils are preserved as isolated carcasses or disarticulated elements, most often in fine-grained overbank facies. Bone beds and microsites are conspicuously absent. Temnospondyl remains were found within a local carbonaceous lens developed upon a sand body, suggesting autochthonous burial in an abandoned-channel setting. Isolated skulls, particularly those of the traversodontid Exaeretodon, are extremely common. Fifteen isolated crania of this cynodont were mapped in a single stratum with limited areal exposure. Abundant preservation of isolated therapsid crania has also been reported in the Beaufort Series (Permo-Triassic) of the Karoo Basin, South Africa (Smith, 1980). Post-disarticulation hydrodynamic sorting (enhanced by scavenging?) of an areally dispersed mass-mortality assemblage may explain this unusual occurrence.

scholarly journals Hierarchical-Matching-Based Online and Real-Time Multi-Object Tracking with Deep Appearance Features

Algorithms ◽  
2020 ◽  
Vol 13 (4) ◽  
pp. 80 ◽  
Author(s):  
Qingge Ji ◽  
Haoqiang Yu ◽  
Xiao Wu
Keyword(s):  
Neural Network ◽  
Convolutional Neural Network ◽  
Object Tracking ◽  
Real Time ◽  
Large Scale ◽  
Accuracy Rate ◽  
Position Information ◽  
Matching Strategy ◽  
Online Methods ◽  
Different Levels

Based on tracking-by-detection, we propose a hierarchical-matching-based online and real-time multi-object tracking approach with deep appearance features, which can effectively reduce the false positives (FP) in tracking. For the purpose of increasing the accuracy rate of data association, we define the trajectory confidence using its position information, appearance information, and the information of historical relevant detections, after which we can classify the trajectories into different levels. In order to obtain discriminative appearance features, we developed a deep convolutional neural network to extract the appearance features of objects and trained it on a large-scale pedestrian re-identification dataset. Last but not least, we used the proposed diverse and hierarchical matching strategy to associate detection and trajectory sets. Experimental results on the MOT benchmark dataset show that our proposed approach performs well against other online methods, especially for the metrics of FP and frames per second (FPS).

Sign in / Sign up

Export Citation Format

Share Document