[Thu, 9 Jul 2015 ~ Tue, 14 Jul 2015] Deep Learning in arxiv

这一期的神作论文有蛮多的，都非常有意思。

Feature Representation In ConvolutionalNeural Networks

该论文中论述了在某种CNN结构下，是否有准确率较高的off model的分类方法（这里是指非softmax）能达到更有效的分类结果呢？

论文给出了肯定的答案。

该论文还给出了各层特征重要性的图表，蛮有意思的

该论文还交代了实验中用到的开源代码。

Towards Good Practices for Very DeepTwo-Stream ConvNets

Openmpi for multi-gpu

Code: https://github.com/yjxiong/caffe/tree/action_recog

Two-Stream Convolutional Networks forAction Recognition in Videos

其中单个stream的准确率如下：

Combine的结果如下：

Understanding Intra-Class Knowledge InsideCNN

该论文利用可视化技术表述了cnn是如何进行类内区分的

另外在该文中也有引用到利用可视化表述cnn如何进行类间区分的。

是很好的一篇cnn可视化理解文章

COMPRESSING DEEP CONVOLUTIONAL NETWORKSUSING VECTOR QUANTIZATION

该论文描述了手机级别的模型存储压缩，非常好的一篇论文。

其实模型参数的压缩不仅会起到存储空间压缩（PQ）的作用，还会起到加速模型的作用例如SVD。

Unconstrained Facial Landmark Localizationwith Backbone-Branches Fully-Convolutional Networks

论文提出了Backbone-Branches Fully-Convolutional Neural Network (BB-FCN)的网络结构，非常有趣，论文中还给出了很多商业非商业的方法做比较，是一篇很好的人脸定位切入文章。

其中的网络结构如下：

结果对比表格：

Facial Landmark Detection by DeepMulti-task Learning

多任务学习可以提高某单一任务的精度。

时间： 2024-07-31 11:10:31

[Thu, 9 Jul 2015 ~ Tue, 14 Jul 2015] Deep Learning in arxiv的相关文章

[Fri, 3 Jul 2015 ~ Tue, 7 Jul 2015] Deep Learning in arxiv

Convolutional Color Constancy can this be used for training cnn to narrow the gap between different lighting conditions? Describing Multimedia Content using Attention-based Encoder–Decoder Networks lots of machine translation, speech recognition, ima

[Fri 26 Jun 2015 ~ Thu 2 Jul 2015] Deep Learning in arxiv

Natural Neural Networks Google DeepMind又一神作 Projected Natural Gradient Descent algorithm (PRONG) better than SGD as evidenced by the boost in performance offered by batch normalization (BN) Deep Convolutional Matching DEEP-PLANT: PLANT IDENTIFICATION

[Tue, 21 Jul 2015 ~ Mon, 27 Jul 2015] Deep Learning in arxiv

Compression of Fully-Connected Layer inNeural Network by Kronecker Product 又是一篇压缩网络文章,但没有给出在imagenet上的错误率变化,有待观测. Building a Large-scale Multimodal KnowledgeBase for Visual Question Answering 李飞飞视觉QA KB示意图系统流程图 Bottom-up and top-down reasoning withc

[Tue, 11 Aug 2015 ~ Mon, 17 Aug 2015] Deep Learning in arxiv

Image Representations and New Domains inNeural Image Captioning we find that a state-of-theart neuralcaptioning algorithm is able to produce quality captions even when providedwith surprisingly poor image representations Deep Boosting: Joint Feature

[Fri 19 Jun 2015 ~ Thu 25 Jun 2015] Deep Learning in arxiv

A Neural Network Approach to Context-Sensitive Generation of Conversational Responses Leverage Financial News to Predict Stock Price Movements Using Word Embeddings and Deep Neural Networks MatchNet: Unifying Feature and Metric Learning for Patch-Bas

[12 Jun 2015 ~ 18 Jun 2015] Deep Learning in arxiv

Multi-pathConvolutional Neural Network for Complex Image Classification Suppresshigh frequency components with Bilateral filter in the second path ParseNet:Looking Wider to See Better code:https: //github.com/weiliu89/caffe/tree/fcn SEMANTICIMAGE SEG

[3 Jun 2015 ~ 9 Jun 2015] Deep Learning in arxiv

arXiv is an e-print service in the fields of physics, mathematics, computer science, quantitative biology, quantitative finance and statistics. There'll be lots of papers in advance. Here's some recent papers which is important or interesting. 1. Obj

[10 Jun 2015 ~ 11 Jun 2015] Deep Learning in arxiv

1. similarity nets DeepSimNets 2. multi-task Learning Multiple Tasks with Deep Relationship Networks 3. interactive learning system Constructionof a Large-scale Image Dataset using Deep Learning with Humans in the Loop

Hadoop自测题及参考答案（持续更新中--2015.6.14）

单选题 1.与其他几项不同的是 A. Mesos B. Mongodb C. Corona D. Borg E. YARN 注:其他几项都是资源统一管理系统或者资源统一调度系统,而 Mongodb一种非关系型数据库. 2.[java基础] 以下不属于线程安全的数据结构是 A. HashMap B. HashTable C. CopyOnWriteArrayList D. ConcurrentHashMap 3.hadoop2.x采用什么技术构建源代码 A. ant B. ivy C. maven