Z StupidTanaka, Zhou Jofan and other 439 people agree Pointer Networks (https://arxiv.org/pdf/1506.03134.pdf NIPS15 's paper, through the improvement of RNN to solve some of the traditional CS problems, such as convex hull, triangulation, even TSP, the most magical place is that the effect is still good, and even better than some approximate algorithm effect. OI/ACM of the party, I mean the three views have been ruined ...
Synthesizing Dynamic textures and Sounds by spatial-temporal generative convnet (http//arxiv.org/pdf/1606.0097 2v1.pdf), seems to be cast NIPS16, from Ying Nian Wu and Songchun Zhu Group, using CNN to synthesize video textures and sounds, the effects explode. Maybe one day the people who do the graphics can retire home for dinner or 233.
Neural programmer-interpreters (https://arxiv.org/pdf/1511.06279.pdf), Best of ICLR16 Paper, using LSTM to achieve task-specific automation programming. This application ICLR15, but this article is more effective than the previous practice:)
Learning to learn by gradient descent by gradient descent (https://arxiv.org/pdf/1606.04474v1.pdf DeepMind's magical work, the light to see the name is very magical, not =,=. Learn from lstm optimization algorithm ... DL can bootstrap ... Perhaps a few decades later, people will think, whether it is optimization first out or DL algorithm first out, Ancient said: Optimize the depth of life, depth optimization problems.
Edited on 2016-06-20 20 reviews thanks for sharing collection• No help • Report • Author retention rights Agree objection, will not show your nameWang Yun Maigo, brother also play value:https://www. ZHIHU.COM/ZHI/PEOPLE/723836120999161856 67 people agreeLatent predictor Networks for Code Generation
Using the bidirectional lstm neural Network + attention mechanism, generate the code according to the instructions of the Furnace Stone card.
Although the effect is not directly available to the extent, but this idea itself is very strange. Posted on 2016-06-18 7 Reviews Thank you for sharing your collection• No help • Report • Author retention rights 104 agreed objection, will not show your nameZhang, research dick Wire104 people agreeThe previous respondents introduced neural art, using the characteristics of the study to make other pictures with the style of the Van Gogh painting, by the way, in fact, similar to a variety of styles fused together to form new content generative model recently seems to be a lot, to study
Image Synthesis
different styles of image synthesis
Similar to neural art, such as this article:
Combining Markov Random fields and convolutional neural Networks for Image Synthesis
Http://arxiv.org/pdf/1601.04589v1.pdf
The style of the classic car into the modern car, the design of the new type, very interesting
The composition of the architectural style is the same, although it seems a bit awkward
Deep Dream
let the machine dream!
In a human dream, a random noise signal is generated in the brain, and the random signal that enters the dream can be seen as a construction, integration process, in which the brain awakens the imagery in memory to match the information produced by the brainstem activation, trying to understand this random signal.
Google's work, published last year, is given an image (an example of an image of the sky and an image of space) that greatly activates the activation of a neuron. For low-level feature, which is shallow neurons, the resulting image maximizes the activation of the corresponding filter, suppresses the activation of other filters, and regenerates the image as follows:
Activating high-level neurons, there's a lot of high-level feature on the image.
It looks like a ghost animal.
Here's an online website where you can play http://deepdreamgenerator.com/
At the end of February, Google created an exhibition of these abstract paintings in the United States, which also sold almost $100,000 to finance the young artists and make them creative.
Sound Synthesis
Visually indicated Sounds
The project homepage is here:http/vis.csail.mit.edu/
The effect is based on the video within the beat, friction and other actions, to synthesize the corresponding sound, such as with a stick to the cement ground when the display of the sound of the bared
http/vis.csail.mit.edu/site-videos/header-vid3.mp4
Video Prediction
Deep predictive Coding Networks for Video prediction and unsupervised learning
Using the information in the previous frames in the video to predict the next few frames, the effect is very good, which can be understood when we play table tennis, obviously can not calculate the ball's landing point but can accurately hit the mechanism of
There's a lot of work to be done here, and there seems to be a job to do the light flow prediction.
Image restoration/inpainting
See CVPR2016 in this article:
Context Encoders:feature Learning by inpainting
Http://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Pathak_Context_Encoders_Feature_CVPR_2016_paper.pdf
dig out a piece of the picture and use the rest of the information to recover the damaged area .
And ICML2016 has done a similar job in this article:
Pixel Recurrent neural Networks
Http://arxiv.org/pdf/1601.06759v2.pdf
More about generative model can see
generative adversarial Nets
Http://arxiv.org/pdf/1406.2661v1.pdf
Adversarial training methods, this network trained two model, a generative model, a discriminant model, joint optimization, the generation model as far as possible to learn to express and training set of the same distribution of samples, Discriminant model as far as possible to distinguish between real training samples and generated model expression of the sample, the two adversarial optimization, and finally can make the generated model can generate and train a sample with the same distribution, so that the network can predict the distribution of the samples not seen
As well as OpenAI's latest work, generating images from a picture set
Generative models edited on 2016-06-23 6 reviews thanks for sharing collection• No help • Report • Author retention rights 104 agreed objection, will not show your nameCharlotte, the mathematical Department of Data Mining workers http://www. Cnblogs.com/charlotte77/ 104 people agreeHave seen an art style that uses deep learning to change pictures
Github:https://github.com/fzliu/style-transfer
From Weibo: http://m.weibo.cn/1402400261/3982309310926836?moduleID=feed&uicode=10000002&mid=3982316391582790 &luicode=10000198&_status_id=3982309310926836&lfid=1076031837287505_-_weibo_second_profile_weibo &lcardid=1076031837287505_-_weibo_second_profile_weibo_-_3982316391582790
I think is very simple, is not able to use deep learning to add filters to photos =_= edited on 2016-06-18 17 reviews thanks for sharing collection• No help • Report • Author retention rights Agree objection, will not show your nameYang Zhenyu, professional approval Division26 people agree
- Chainerによるディープラーニングでav Women like portrait Eiken cable サービスをつくったノウハウを Gong able
- ディープラーニングで photo から busty かどうかを してみる (うまくいったか subtlety)
Edited on 2016-06-25 8 reviews thanks for sharing collection• No help • Report • Author retention rights 5 Approval objection, will not show your nameChen Liu, drink 8,000 cups of water every day! 5 people agreeUsing deep learning to detect the damage of insulators in high-voltage power grids, you can't imagine?! Article link: Application of deep learning in image recognition and fault detection of Smart grid published on 2016-06-20 add comments Thanks for sharing collection• No help • Report • Author retention rights 2 Approval objection, will not show your nameFledsu, Computer Science,phd2 people agreeRecently learned that a work is using neural networks to estimate the structure of proteins published on 2016-06-20 2 comments Thanks for sharing collection• No help • Report • Author retention rights Agree objection, will not show your nameAnonymous User45 people agreeEveryone seems to be very interested in the appearance of the results (< 1%) to send GitHub.
-----------------------------------------------------------------------------------------------------------
Some time ago and the classmate blew, talked about the DL can do the image enhancement. Then a whim, this thing is not can be used to do AV to code. Edited on 2016-06-19 43 reviews thanks for sharing collection• No help • Report • Author retention rights 3 Approval objection, will not show your nameWang name Yang3 people agreewaifu2x
WAIFU2X-Search Results
Using deep learning to enlarge anime images posted on 2016-06-18 1 reviews thanks for sharing collection• No help • Report • Author retention rights 6 Approval objection, will not show your nameZhou, programmer/find work ing6 people agreeGoogle Magenta to create songs through the Neural Learning Network
http/yuweining.cn/t/song/gooGle-magenta-song.mp3 Edited on 2016-06-19 2 reviews thanks for sharing collection• No help • Report • Author retention rights 5 Approval objection, will not show your nameNrek, one sentence is too short5 people agreeUse neural network to play Super Mario automatically
Videohttps://www. youtube.com/watch? v=qv6uvoq0f44
Codehttp/pastebin.com/zzmsnahx
-paper:http/nn.cs.utexas.edu/downloads/papers/stanley.ec02.pdf Posted on 2016-06-20 Add Comment thanks for sharing collection• No help • Report • Author retention rights 1 Approval objection, will not show your nameThe Fist of Heaven1 People agreeTo be exact, not yet. Posted on 2016-06-26 Add Comment thanks for sharing collection• No help • Report • Author retention rights 0 Approval objection, will not show your nameImplant sleep, in-depth study of tobacco and Wine Hygiene machine learning NLPCNN can achieve sentiment analysis, classification, image word recognition, speech recognition, collaborative translation, etc. from:https://www.zhihu.com/question/47563637#answer-38117824
Deep learning applications in which areas make you feel like "I'm going, that's OK!" ”？
