Depth Comma ai first autopilot based on the road video forecast paper

Lei feng's network: by the end of this article from the compilation.

George Hotz and Comma.ai background:

George Hotz in 2007 and first crack to crack Sony PS3 first person for the iPhone,2010 year. Internships in Google,Facebook, spent 4 months in Space after working in 2015, artificial intelligence startups Vicarious, left in July and in September of the same year founded Comma.ai, alone in the garage study on automatic driving technology, announced that challenge Google,Mobileye, in April this year, the company got a $ 3.1 million investment. On August 6, George Hotz has open source code and thesis research. (Paper and source code download)

Original author Eder Santana,George Hotz. Following is a compiler full text:

Application of artificial intelligence in automatic driving, Comma.ai's strategy is to build up an agent (agent), train cars through the model to predict future traffic events mimic human driving behavior and drive plans. This paper presents our current research methods for driving simulator to study variation auto-encoder (Variational Autoencoder, known as VAE) and generated against network-based (generative adversarial network, known as GAN), cost function for realizing video forecasts (cost function). , We trained a combination based on recurrent neural network (RNN) model (transition model). Fossil iPhone case

The optimized model of cost function does not exist in pixel space, but shows us ways to achieve its forecast for multi-frame realism.

| Profile

The self-driving car [1] is in the study of artificial intelligence, one of the most promising areas in the short term, which makes use of a lot of driving at this stage in the process, with tags and contextual information-rich data. Taking into account its perceived complexity and control, automatic driving technologies, once realized, will also develop many interesting topics, such as action recognition in video and drive plans. At this stage, with the camera as primary sensors, combined with Visual processing and artificial intelligence technology for automatic driving of the advantage in terms of cost.

Due to the depth of recursive neural network for learning, development, virtual reality interaction (interaction) of a more convenient, vision-based control and reinforcement learning in the following reference [7][8][9][10] has been successful. This interactive form allows us to to repeat the test with a different policy scenarios, and can simulate all possible events to train the neural-network-based controller. For example, Alpha Go[9] depth using Convolutional neural networks (CNN) by accumulating and his chess game experience to predict the winning probability of the next. Go game engine to simulate all possible changes in the results of a game, and used as a Markov chain tree (Markov Chain Tree) search. At present, such as Let Go learned to play the game screen Torcs[7] or Atari[8], required hours of training and learning.

Because the learning agent is difficult to implement exhaustive interaction with the reality, which at present is about two solutions, one of which is manually developed a simulator, the other is to train the ability to predict future scenarios. Programme relates to the physical world of the former rule definitions, and stochastic modeling of the real area of expertise, but such expertise has been covering all the information associated with the control, covering the existing Flight Simulator [11], robot, [12] and so on.

We focus on setting human agent (agent) to make its own simulations to predict real-world scenarios, installed on the windscreen of a front-facing camera as input to a video stream. FOSSIL iPhone 5

Early years is based on the State space of a physical agent [13] to controller training simulations, other models of Visual processing alone can only adapt to the low dimension or texture features simple video, such as the game Atari[14][16]. Texture features for complex video, is through the passive video projections (passive video prediction) to identify actions [17].

This paper relevant literature complemented the existing video projections, we let the controller itself to train the model and predict real video scene, lower dimensional compression is calculated and converted into the corresponding action. In the next section, we described to predict live traffic video shot by the DataSet (a DataSet).

| Datasets (DataSet)

Part of our open source used in this paper to automatic drive test data. Data sets of test data with self-driving cars comma.AI test platform used is consistent with the cameras and sensors. Our Acura ILX 2016 one Point Grey cameras installed on the windscreen, and by 20hz image acquisition and frequency on the road. Release the data set contains a total of 7.25 hours of driving data, 11 video, video frame is captured from captured video 160*320 pixel picture. In addition to video, DataSet includes data from several sensors, respectively, measured at different frequencies, which interpolated 100Hz, sample data included vehicle speed, steering angle, such as GPS, gyroscopes, IMU. Specific details of the data sets and measurement devices can be accessed through synchronization site gets.

We record the sensor and the timestamp when the captured video frame and insert the test time and linear synchronous sensor data and video. We have also released original video and sensor data stored in HDF5 format, the format of choice because it is easier to machine learning and control software in use.

This article will focus on video frames, steering angle and vehicle speed. Our original by downsampling the image data is 80*160, and the image is-1 to 1 pixel to fine-tune your actions (renormalizing), this completes the pretreatment. Sample image shown in Figure 1. In the next section we define this article aims to study the problem.

| Problem definition (Problem definition)

T-frame of the XT is a dataset, Xt is the frame length for videos n says:

St is the control signal, and directly related to the image frame:

At the speed and steering angle (steering angle) corresponds to.

When projecting images define a valuation function f:

  Predictions for the next frame as follows:

Note that the defined for higher dimensions and the dimensions are interconnected, similar problems can also occur in the machine learning such as slow convergence or poor fitting of data (underfit) [26].

Studies have shown that [20], using dynamic neural network convolution (Convolutional dynamic network), without appropriate adjustments (regularization), the model is on a single set of data to simulate well but the overall prediction accuracy of the data is low.

Way is through a simple video, artificial [14] direct training valued function f, recent papers [20][17] shows that generate texture complexity can predict video, but did not solve the problem of motion transfer, nor a compact intermediate representation of the data generated. In other words, their models without downsampling or hidden coding for low dimension, but full implementation by convolution transform.  But because of dense space (dense space) [18], the probability that filter (filter) and control the output blurred the definition of (ill-defined), compact (compact intermediate representation) is vital to our work.

As far as we know, this is the first article from the real road scenarios to predict subsequent frames of video essays, in this connection, in this article, we have decided to piecewise function f in order to debug debugging it can block.

 First of all, we learned a Autoencoder frame XT embedding data into Gaussian hidden Zt (Gaussian latent space),

Dimensions of 2048 is the demand by the experiment, variational Bayesian [1] encoding (variational Autoencoding Bayes) enforce the Gaussian assumption. First step is space simplifies the transfer of learning to the pixels in the hidden layer (latent space), but assumed from the encoder Autoencoder can properly learn hidden layer of Gaussian, then just transfer model could promise not to leave the high density area of the embedded space, we can simulate a realistic video. Hypersphere of RADIUS ρ in high density areas, is the embedding dimension, and Gaussian prior variance function. Described in the next section, we will begin a detailed Autoencoder and transfer model.

| Driving Simulator (Driving Simulator)

Taking into account the complexity of the issue, we do not consider the end-to-end (End-to-End), but uses a separate network to learn video projections. Proposed architecture is based on two models: one is to use Autoencoder to dimension, the other is converted using RNN (transition).  Complete model is shown in Figure 2.

 From the encoder (Autoencoder) we chose a hidden layer of Gaussian probability distribution model of learning data embedding, especially to avoid the hyper-sphere low probability and discrete areas of the focus at the origin, this regional presence would prevent the hidden layer of continuous transformation model of learning. Variational Autoencoder[1] and related activities [19][21] in the hidden layer of the original data using Gaussian prior models model (generative model). However, the Gaussian assumption in the original data and applies to process natural images, so VAE predicted the result will look fuzzy (see figure III). The other hand, the fight network (GAN) [22] and related activities [2][3] learn to build model together with the generator cost functions.  So generative alternating with discriminator networks can be trained.

Generative model to hidden distribution of sample data into the dataset, discriminator distinguishing network centralized the data samples from all discrimination out of the sample generator, generator to be able to play the role of fool discriminator, therefore discriminator can also be seen as a cost function generator.

Not only do we need to learn from hidden layer to image space generator, also need to be able to encode the image back to the hidden layer, so you'll need to VAE and GAN networks combined. Intuitively, a simple way is to use VAE method with a cost function directly. Donahue et.Al [23], proposes a model of learning and two-way GAN bijective transformation network. Lamb et. Al.[24] proposes distinguishing network (discriminator generative networks), previously trained classifier differences as part of the cost function. Finally, Larsen et.Al[25] is trained to VAE and GAN networks, this encoder can optimize hidden layer of Gaussian prior models, and GAN networks extracted feature similarities. Generator will output the hidden random samples as the input, and outputs the encoder network, optimized to fool discriminator, and minimize the original image and similarity of the decoded image. Arbiter is always trained to distinguish type – judging the authenticity the authenticity of the image.

We use Larsen et.Al. [25] the methods to train the Autoencoder, schematic diagram in Figure 2 illustrates this model. In their paper as described in [25], encoders (Enc), generator (Gen) and arbiter (Dis) network optimization makes the following minimum cost function:

In the above formula,

Meet coding output distribution q (z|x) and prior distribution p (z) of Kullback-Liebler divergent, are is VAE are is of matrix, p (z) meet n (0,1) Gaussian distribution, we with reparemetrization to optimization its regularizer, so in training process in the always meet z = µg + ∈ σ, in test process is meet z = μ (formula in the µg and σ is coding network of output, ∈ is is and µg, and Σ Gaussian random vectors of the same dimension)

Second error is a calculated value represents the l layer hides the activation value in the discrimination network, the value with a legitimate image x and the corresponding encoding-decoding the value of Gen (Dis (x)) is calculated.

Assumptions:

Can be obtained:

In the course of training, in order to avoid steps that are too cumbersome, Dis is usually constant.

Last LGAN is generated against the network (GAN) costs [22], the cost function is expressed between Gen and Dis game. When the Dis are trained, Enc and Gen always remains fixed at:

U meet the normal distribution n (0,1) random variable, the first part of the formula is the log-likelihood function of Dis, used to distinguish legitimate images, the remaining two parts are random vectors u coded values or z = Enc (x) value used to distinguish whether the fake of image samples.

At the time of training the Gen, Dis and Enc and always maintain a fixed value:

Said Gen fool Dis identification network, [25] second Enc in the equation (x) in the training process is usually set to 0.

Our train is 200 times of the Autoencoder, each iteration containing 10000 gradient update, increment size is 64, as described in the previous section, sample random sampling from the driving data. We use optimized Adam [4], from the encoder network architecture reference Radford et.Al[3]. generator from 4 layers to a volume composed of grass-roots, following the normalization of samples after each layer and activate functions that leaky-ReLU. Made up of multi-volume primary arbiter and the encoder, and immediately after the first layer is the normalized sample of operations, the activation function used here is ReLU. Disl is the third rolled grass-roots network decoder output, then the normalized sample ReLU operation. Distinguishing the output size is 1, it cost function is binary cross-entropy function, encoding the output size is 2048, compact (compact representation) into the raw dimensions of 1/16. Details can be viewed in Figure 2 or this article synchronization code, coding-decoding and image of the sample is shown in Figure 3.

After the Autoencoder training, we fixed all the weights and Enc as a training model preprocessing steps, we will discuss in the next section switching model.

 Conversion models (transition model)

Training Autoencoder later, we got used to convert the data set, use Enc XT-> training RNN:zt,ht,ct-> Zt+1 ZT to indicate the encoding transformation of space.

W,V,U,A for training the right values in the formula, and HT is the hidden RNN, CT signal directly controls the vehicle speed and steering angle, and LSTM,GRU, as well as CT multiplication with the ZT iterations will further research in the future, now used to optimize training weighted cost function is the mean square error (MSE):

Obviously the formula is optimal, because when we train Autoencoder, the distribution of code z Lprior Gauss constraint is imposed. In other words, the mean square error equal to the value of a normally distributed random variable. If the predicted encoded values for:

Forecast picture frames can be expressed as

We use a sequence of video frames of up to 15 to train the model, after the first 5 frames of output as input to 10 frames after learning network, that is, Enc (XT) when the function Z1,...,Z5 is calculated, continue to enter as a follow-up to get

Feedback to continue as input. In RNN's literature, will continue as input output feedback is called RNN hallucination. In order to avoid complicated operations, we'll continue as input the output feedback in the process of the gradient is set to 0.

| Test results

In this study, we will spend most of our energy on how to make Autoencoding structure to hold on the texture of the road, as mentioned above, we have studied the different cost functions, although their mean square error is about the same, but using GAN network cost function is to get the Visual effects for best results. As shown in Figure 3, we show two set of training models corresponding to different cost functions of decoded picture, as expected, the image is fuzzy neural networks based on MSE, making multiple lane mark recognized by the error into a long single lane.

In addition, the fuzzy reconstruction cannot be preserved behind to the edges of the image, so this method cannot be used to promote the most important reason is difficult to achieve distance and estimation of the distance between the vehicle. The other hand, the learning curve is plotted with the MSE line speeds faster than models based on fight network. Perhaps in learning with steering angle information to pixels is encoded can be immune from the problem. We will keep this issue for future research.

Once we get a good Autoencoder, you can begin to train the model. Predict the results picture frame as shown in Figure 4, we use the 5Hz video of training the model, after learning of the transformation model even after 100 frames are consistently maintained road frame structure. When the seed sampling frames from converting model is different, we have observed through the lane, close to the vehicle in front, and front drive open, such as driving, but the model does not simulate scenes out of corners. When we drive in the corners of the image frame when you initialize the model, model fast lane straight, straight and start simulation. Under this model, although no accurate optimization cost function in pixel space, we can still learn from the video conversion. We also believe that the reliance on the more powerful model (such as depth RNN, the LSTM, GRU) coding the context and contextual encoding (sensor-assisted video sample plus the steering angle and speed) will appear more realistic simulations.

This paper release data set that contains this method during the experiment, all the necessary sensors.

| Conclusion

Comma.AI in learning is introduced in this paper preliminary research of vehicle driving simulator, video forecast model based on Autoencoder and RNN. We do not have end-to-end (End-to-End) associated with all things, instead of based on the fight network (GAN) cost function is to train Autoencoder, lifelike images, and then we train the RNN model in the embedded space. While the Autoencoder and the outcome of conversion model looks very realistic, but to simulate all the events associated with the drive still needs to do more research. Automatic driving in order to stimulate further research on, we publish the video sampling as well as vehicle speed, driving, steering angle sensor data, such as data sets, and is currently being trained neural network is open source source code.

"References"

[1] Diederik P Kingma and Max Welling, "Auto-encoding variational bayes," arXiv preprint

arXiv:1312.6114, 2013.

[2] Emily L Denton, Soumith Chintala, Rob Fergus, et al., "Deep generative image models using laplacian pyramid of adversarial networks," in Advances in Neural Information Processing Systems, 2015.

[3] Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434, 2015.

[4] Diederik Kingma and Jimmy Ba, "Adam: A method for stochastic optimization." arXiv

preprint arXiv:1412.6980, 2014.

[5] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio, "Generative adversarial nets," in Advances in Neural Information Processing Systems, 2014.

[6] Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian Goodfellow, "Adversarial Autoencoders," arXiv preprint arXiv:1511.05644, 2015.

[7] Jan Koutn´ ık, Giuseppe Cuccu, Jurgen Schmidhuber, and Faustino Gomez, "Evolving large- scale neural networks for vision-based reinforcement learning," Proceedings of the 15th annual conference on Genetic and evolutionary computation, 2013.

[8] Volodymyr Mnih, Koray Kavukcuoglu, David Silver, et al., "Human-level control through deep reinforcement learning," Nature, 2015.

[9] David Silver, Aja Huang, Chris Maddison, et al., "Mastering the game of Go with deep neural networks and tree search," Nature, 2016.

[10] Sergey Levine, Peter Pastor, Alex Krizhevsky, and Deirdre Quillen, "Learning Hand-Eye Coordination for Robotic Grasping with Deep Learning and Large-Scale Data Collection,"

arXiv preprint arXiv:1603.02199, 2016.

[11] Brian L Stevens, Frank L Lewis and Eric N Johnson, "Aircraft Control and Simulation: Dynamics, Controls Design, and Autonomous Systems," John Wiley & Sons, 2015.

[12] Eric R Westervelt, Jessy W Grizzle, Christine Chevallereau, et al., "Feedback control of dynamic bipedal robot locomotion," CRC press, 2007.

[13] HJ Kim, Michael I Jordan, Shankar Sastry, Andrew Y Ng, "Autonomous helicopter flight via reinforcement learning," Advances in neural information processing systems, 2003.

[14] Junhyuk Oh, Xiaoxiao Guo, Honglak Lee, et al., "Action-conditional video prediction using deep networks in atari games," Advances in Neural Information Processing Systems, 2015.

[15] Manuel Watter, Jost Springenberg, Joschka Boedecker and Martin Riedmiller, "Embed to control: A locally linear latent dynamics model for control from raw images," Advances in Neural Information Processing Systems, 2015.

[16] Jurgen Schmidhuber, "On learning to think: Algorithmic information theory for novel com- binations of reinforcement learning controllers and recurrent neural world models," arXiv preprint arXiv:1511.09249, 2015.

[17] Michael Mathieu, Camille Couprie and Yann LeCun, "Deep multi-scale video prediction beyond mean square error," arXiv preprint arXiv:1511.05440, 2015.7

[18] Ramon van Handel, "Probability in high dimension," DTIC Document, 2014.

[19] Eder Santana, Matthew Emigh and Jose C Principe, "Information Theoretic-Learning Autoencoder," arXiv preprint arXiv:1603.06653, 2016.

[20] Eder Santana, Matthew Emigh and Jose C Principe, "Exploiting Spatio-Temporal Dynamics for Deep Predictive Coding," Under Review, 2016.

[21] Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly and Ian Goodfellow, "Adversarial Autoencoders", arXiv preprint arXiv:1511.05644, 2015.

[22] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, et al., "Generative adversarial nets," Advances in Neural Information Processing Systems, 2014.

[23] Jeff Donahue, Philipp Krahenb ¨ uhl and Trevor Darrell, "Adversarial Feature Learning," ¨ arXiv preprint arXiv:1605.09782, 2016.

[24] Alex Lamb, Vincent Dumoulin Vincent and Aaron Courville, "Discriminative Regularization for Generative Models," arXiv preprint arXiv:1602.03220, 2016.

[25] Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle and Ole Winther, "Autoencoding beyond pixels using a learned similarity metric," arXiv preprint arXiv:1512.09300, 2015.

[26] Jose C Principe, Neil R Euliano, W Cur Lefebvre, "Neural and adaptive systems: fundamentals through simulations with CD-ROM" John Wiley 

Lei feng's network (search for "Lei feng's network", public interest) Note: please contact the authorized and keep the source and author, no deletion of content.

Chop hands down Guess Express send gift

Our readers will note that we have recently started scouring the sea products. Some readers ask, love machine, machine or a cell phone?

Between a, have and everyone said about, love make machine of machine has not Council limited to phone of range, but various consumption electronic equipment and the novelty products, Shang to AR Drone 2.0 four axis aircraft (small series has chop hand) such bully male PA female of loaded b goods, Xia to sucking electrostatic key buckle (small series again chop has a times hand) such of small stuff, all in we of recommended range within.

Of course, what we recommend is what they will play, for some time, we buy things, including:

Health bracelet SEIKO SSC001 three-eye chronograph Angel Sensor (shipped by April 2014) Sennheiser CX880 headphones LEGO Minecraft 21102 "my world" blocks static Keychain AR Drone 2.0SONY USH-1 SHDC 32G Express card Beats Tour In-Ear noodles headphones Anki Drive (that went to the Apple store in Hong Kong but didn't buy) ... ... Juicy iPhone 5 case

Over so many things, some make their own play, some friends, some plan to sell on our shop ... ... Actually said has so more, is to to everyone convey a information, we began side play machine side do sea Amoy has, if you see what novelty of digital products, you can in micro-letter Shang recommended to we; if you hope through Haitao buy to you right of digital products is and start, also can to we sent micro-letter, we himself sea Amoy of when may will way help you with oh.

And finally play a game: left is an order of small make up just last night, including AMICO static key chain is the gift of the game to send: order status, at present Amazon not shipping, anyone shipping time, forwarding companies, storage and transport companies delivery time was United States local time? Please in micro-letter Shang put you guess of time to we (for example: 1) 10.31 13:00 2) 11.1 18:20 3) 11.3 9:10), most close (30 minutes within) of readers can get we sent out of AMICO sucking electrostatic key buckle a (total 3 a, can and have) Oh, if zhiqian no concern we of micro-letter, that on took up you of phone scan following two dimension code participate in we of game's. Juicy Couture

 

Recent price Philips Norelco 1280X 181 99

Philips Norelco 1280XCC is a subsidiary of Philips high-end domestic flagship series, 1280X more than the flagship 1290 sent the travel lock. 1280X can be wet, plot system can automatically plot cutter, GyroFlex 3D smart fit facial contours. 1280X also has flagship chrome shell, and LED display. Ted Baker iPhone 6 cover Ted Baker iPhone 6 cover

Philips Norelco 1280X recent low price of $ 211.99, then below are a State minus $ 30 discount, that is, price $ 181.99, about 1100 Yuan, domestic prices in more than 1600 Yuan, the price advantage.

Purchase address

Different killer Galaxy Note5 S Pen experience

In the Android's fledgling years, Samsung to try to do a lot of exploring, among the most famous result is a Galaxy Note series. In addition to the iconic giant screen, S-Pen is the most obvious feature, the former is also a name of Note in one, and it has been accompanied by the Note series, undergone a full 5 generation of change.

 

In September 2011, the Android 2.3, mainstream screen sizes are 3.5 inch era of early Galaxy Note popped 16:10 5.3-inch screen and the addition of S-Pen shocked the industry. Early S-Pen to support the original send and graffiti handwriting features, screenshot support circles and so on.

Note2 released a year later, S-Pen also have a substantial update, upgrade to 1024, introduced the iconic Air suspension View preview function: S-Pen in 1-2cm at the top of the screen to be induced, and the screen displays the tip of point and can volley hover complete album/video and a detailed preview of the calendar. In addition, the Air suspension preview View is triggered, finger capacitive touch screen ignored to prevent writing hand when hit, greatly improving the experience of writing on the touch screen.

When the Galaxy Note3, Samsung introduced a fan-shaped "floating window command", intelligence memorandum, with the cut with stickers, screen writing, s, search for common operations, such as joined together. Memo added the function of turn handwriting to text, you can copy-and-paste handwritten words into text. In addition, S-Pen also has circle opening calculator, browser shortcut function. Hello Kitty Note 4 Cases

 

Galaxy Note4, double S-Pen pressure sensitive levels, upgrading to 2048 levels of pressure-sensitive and the software UI and functionality made a lot of improvement adjustments. Galaxy Note5 released this year, the S-Pen from the original drawing into the pop-up, joined out screen writing, scroll the screen and many other useful features.

 

Rush matured floating window command/smart memo/s Note

Metal and glass elements more Note5, S-Pen push the CAP has become a metal material. Press feels and true ball-point pen spring and very close to spare nothing and thinking very addictive, friends jokingly referred to this experience, I can play for a year.

 

Design of the barrel itself changed little, and body paint of the same color, silver tail is the familiar smell of hot. Writing can still change randomly comes with 5 additional written and the replacement of the clamps. Tip is similar to silicone material, you can increase the scratch resistance across the screen, provide the friction needed to write and support. Actual writing feels very soft, but friction is smaller than the real writing.

 

Will vibrate and beep after taking S-Pen, intelligent functions by "side keys, strokes, double-click the floating window and the command" 3 elements interact mainly hosted by the s Note.

Spare S-Pen pen close to the side of the screen click on the button can be called a floating window command, intelligence memorandum, of intelligent multi-screenshots and screenshots to prepare 3 preset function, users can add their own 3 application shortcuts. Previously operating speed is high, but is likely to make many users Miss screenshots of operation long press the side key, and press the side key functions have been canceled in the screen shot.

 

Stroke/size/color options, as well as "floating sticky notes"

Intelligence memorandum supports the reminder settings, you can customize the background color and stroke, and can even be used as a desktop plugin paste it to the desktop. Suspending the entire page, you can adjust the size and position, watching Twitter app and take notes. If you like it, all handwritten notes are "thumbnail" form on the table, the convenience advantage is obvious, but the bias speed record, didn't offer a wide choice of strokes.

 

Illustrator stroke and paint

 

Illustrations polished

General s Note notes a variety of strokes, choice of color, thickness, and can even screen colors (color referred to copy the contents of the screen tip), which is a versatile "graffiti painting". Combined with pressure-sensitive rich stroke, used to paint graffiti and writing is self-evident.

Simulation of watercolor and oil painting brush strokes

  Hello Kitty Note 4 Case

Killer feature: write fast/smart card screen section/character recognition

Note5 new card screen function: rate of screen out the s Note do not require any additional operation minimalist recording mode is available. This mode will ignore the touch of a finger, only S-Pen can write, the machine becomes a real Note (Note notes).

Meet the AMOLED screen black not power-hungry features, card screen was about even long has no obvious impact on the electricity, very suitable for high speed and long period of record. For users who often need to record inspirations and take notes, this is a killer feature.

To the input box will pop up ORC input button

ORC text recognition, when tip refers to handwriting recognition area appears on the box, content will be direct to recognize handwriting to text, call log the contact, and do simple input of time, if not or too lazy to type you can use S-Pen to complete all operations.

 

Screenshot of graffiti and scroll screen

Scrolling screen is now a bring your own capabilities, maximum 22 page, you could almost meet the long tweets, long Web pages are recorded. Full page screenshot can also be edited and written. More violence is, the machines can recognize text in the screenshot, you can copy text contents separately.

 

Automatic cutout screenshots and text recognition

Intelligent multi-screen you can press square, oval and random shapes capture screen contents. More similar to PS in the generation of automatic cutout and clip art library function, which can automatically identify and intercept a target profile, which is the reinforcement of the pasteboard, you can put together multiple screenshots/cutout for editing/scrawling.

 

Left to local amplification, right table insert and clip art library

There are local amplification, simple charts and other features were as an optional plug-in for users to choose. Numerous functions seamlessly, Note5 when I take notes in meetings/classes, with unmatched convenience of cellphones or computers.

 

As a rolling 5-generation products will gradually feel S-Pen to use under the surface there are many designs and details in the use of: System records the S-Pen pull out against losing at the time and after the lock screen reminder, pull back will display a translucent floating Windows, which greatly reduces missed opportunities.

Overall, Note5 S-Pen on this very easy to use. This experience toxic ... ... It experiences even more than the surface and iPad Pro, mobile phone has more competitors in the field. In the cell phone age differences are getting smaller, and have to say that S-Pen is a different case. It's a series, built high on a barrier and confirmed the sentence: S-Pen deserve Note.

Tips

Micro sweep sweep, author tips bar ~

Every day we are in the personalization this is what road

Lei feng's network (search for "Lei feng's network" public attention) by writer Xie Defu (micro signal: beancurd191), more than 9 years Internet experience, 6 years relevant experience in personalization, big data. Enter Huawei worked 6 years after graduation, in 2013 start date, focus on large data technology in film and TV, culture, video, entertainment and other fields of application. Vera Bradley phone case

Qian days rumors to was Tencent acquisition of today headlines many people should not strange, I is today headlines of faithful user, why I think today headlines do have than other news/information client better, is because it show to I of content are is I wants to see of, with I in above of behavior increasingly more, it show to I of content more accurate, today headlines using of most core of technology is personalized recommended technology.

With the rise of mobile Internet, many of the behavior gradually shifted from PC end user mobile. People spend more and more time on the phone. People everywhere use mobile phones, when in your car, go to the bathroom, eating, even when you walk in you are using mobile phones. Relative to the PC side, mobile features a narrow screen, user time fragmented. And as the volume of information is increasing, it is hard to find what they want quickly through large amounts of information. This experience was very bad, if you are a product manager, if you face the same problem, hope it is helpful to talk about your next.

What is the recommended engine?

If you bought a book on Amazon, you may encounter such a situation, when added to the shopping basket when you choose a book, it will automatically recommend other books to you. For example: people who bought this book also bought XXXX, you may also like XXXX, combining recommendation, buy the book there are several other combinations can enjoy a favorable combination of the book price. These uses are recommended system, it simply recommended system is the study of all acts of the user on the platform, portraits of users, as well as research platform/product. While matching user and product processes.

Recommended application range of engines?

Recommender systems in a wide range of applications in various fields, such as e-commerce sites, video sites, video-streaming platform, news client, literary sites, music sites, and so on. Below is the recommended system at the famous e-commerce Web sites, video sites, some cases of application and effect.

Why recommendation systems are widely used in various fields?

1, directory or search for ways to find desired content, may have to turn on the small screen of the mobile terminal screen, high cost of finding interesting content, poor user experience.

2, approved the substance of recommendations presented to users, all users are interested in, and each user sees is not the same as Amazon CEO Bezos said, to 1000 users accessing the Amazon to see 1000 different Amazon.

3, users now have very many choices, you can choose the diversity and fragmentation of time users open the phone, such as failing to quickly find content of interest, will be leaving soon.

4, personalized recommendation algorithm for user interests recommend precision, helps users quickly find content of interest, and when you're done with a content, immediately recommend things to you, you can increase user stickiness.

5, to help users find more quality long-tail content, general user access only at the top of the 10% about the contents, many will never sink in not found in the database. Vera Bradley phone cases

6, to help balance the ecological platform, avoiding the Matthew effect, featured content is always to get more exposure, popular content never an opportunity is concerned, make content production ecological polarization.

Architecture and core algorithm for recommender systems

Here I did a product as an example to explain, in architecture, could each do is somewhat different, but used some of core algorithms, you should be the same. How to achieve, product managers do not need to focus so fine, just an idea of the principles can be.

Common algorithms including user preferences in Recommender System algorithms, collaborative filtering algorithm (item_base,user_base), Association Rules algorithms, clustering, similarity algorithm (content_base) and a number of other complementary algorithms. Final analysis of results is as follows:

1, according to users ' preference algorithms figure out users ' interests/product, and then recommend to the user.

2, according to Association Rules algorithms, worked out between support and confidence. The most common application is to mix and match, beer and diapers are classic examples.

3, item_base is in accordance with the collective user behavior to calculate the similarity between the goods and users have read and recommend or buy most similar articles to the user.

4, clustering algorithm based on user clustering or clustering of products. Clustering can be recommended for the category, or continues to calculate the relationship between user and product classes.

5, content_base for relevance calculation is based on the properties of the product itself, calculate the similarity between the goods, the most common kind of recommended applications.

6, user_base calculates the similarity between users is based on the collective behavior, such as calculate a and b are very similar, you can like the contents of b, but not seen, recommended to a.

Common scenarios Home guess you like recommending

Due to the smaller mobile screen, one screen to show less content, and users to find the content they are interested in a screen turned down one screen, show personalized recommendation of content in this place, you can quickly catch the user's eye.

Found in sections recommend content of interest to users, you can give users a pleasant surprise.

Association recommended/featured content page for details: details of the contents page to users recommending similar content with the current content.

Read end/video/live broadcast ends recommend: featured content similar to the current.

Recommended search page when search results, you can recommend its content of interest to users.

Personalized recommendation system in application of several key problems

Personalized recommendation system is a very complex system, which not only issues relating to data processing algorithms and system flexibility, but also robust, data sparsity problem, cold start problems, system accuracy and diversity issues.

1, trash data processing: the exception for system data, junk data services developed a set of rules.

2, cold start issue: due to the precipitation of new users access without data, making it difficult to recommend according to user behavior, currently prevalent method is provides interested labels to guide page when a new user first logs on, guiding the user to set, combined with the other algorithms. A better approach is to use social data for users on other platforms.

3, data sparsity problem: can l dimension calculation using a clustering algorithm, and combining it with other algorithms for recommendation.

4, recommended for accuracy and diversity: through the combination of multiple algorithms recommended. Ensure that the recommended result set accuracy and diversity.

Above is a basic knowledge about the personalized recommender system, and want to help.

This article published by everyone is a network product manager Lei feng. Unauthorized, prohibited reprint!

Low price Olympus E PM2 dual mirror set 349

Olympus E-PM2 Olympus is introduced in 2012 M43 micro model, which uses a Sony produced 16 million pixels COMS,3-inch 460,000-pixel LCD screen, telling AF anti-shake system, Olympus home of the fuselage, Cannon made a Freudian slip. Double-head set 14~42 and 40~150, respectively, the pursuit of small basic can meet the. Balenciaga iPhone Balenciaga iPhone

Olympus E-PM2 AIU a record low price of $ 349, about 2100 Yuan, to prices in the 2300 Yuan. Japan is Amazon 37800 yen, about 2240, at 2400 Yuan. Domestic B2C licensed 3K+, clear price advantage.

Purchase address