Shulou(Shulou.com)06/03 Report--

2019-12-31 17:59:04

Three issues a week, a detailed explanation of artificial intelligence industry solutions, so that AI closer to you.

The solutions are selected from the Pro industry database of Machine Heart.

Option 1: global population distribution map and density system

Introduction to the solution:

The system uses deep learning to analyze satellite image data with the goal of creating a map of the global population distribution to guide the company's drone program and help more people access the Internet. By training its model with information about a relatively small number of photos (about 8000 pictures taken over India), the trained neural network can identify evidence of human life in photos taken from 20 other countries.

Solution details:

At present, the system has analyzed a total of 15.6 million images representing 21.6 million square kilometers of earth land. The error rate of identifying artifacts is less than 10%. After that, we plan to combine other data to generate an accurate population density system.

The project is led by Tobias Tiecke, an Facebook engineer and optical physicist, and Yael Maguire, the director of Facebook Connectivity Lab.

The researchers simply judge whether the sample photos are inhabited and tagged, and train the neural network to simply choose one of the two between "yes" and "yes", instead of tagging house, car and other artificial evidence. After the training, we can get a simple classifier (Classifier). At this time, the system is deployed, it automatically analyzes the satellite image of the earth's surface, determines whether there are artificial traces in the image, and then combines the census and other data to analyze, and finally generates a more accurate population density map.

Plan 2: big data cleaning software service

Introduction to the solution:

A data conversion platform that automatically cleans up data, creating interfaces that can be used by multiple different platforms (traditional relational databases, Hadoop clusters). Trifacta can create SQL queries or map reduce code that can run on multiple physical data storage and processing systems.

The platform takes the user experience into account in the underlying design, and provides services that free data scientists from the dirty work of data purification, and allows data analysts to focus on data processing. There is no need to develop complex pipes to clean up the data and put them into the data warehouse. It is the first company that successfully combines back-end data technology with intuitive front-end user visual interface.

Solution details:

Automated sampling of data from big data, using visualization to allow analysts to find interesting patterns in a very short time.

Machine learning algorithms are applied to provide suggestions for reorganizing information and organizing. Analysts can group data sets into logical parts of information, normalize them each time, and display them in a friendly interface during their work. Summarizing the entire data set is the last step, which will eventually form a semi-structured data set and eventually take shape.

Trifacta has more than 50 enterprise users, including Cisco, sports camera manufacturers GoPro and Juniper, healthcare systems integration insurance company Kaiser Permanente, supply, information and health management products and services provider McKesson, PepsiCo, Pfizer and Procter & Gamble. Trifacta charges are based on the organization's volume of data, ranging from $100000 to $150000.

In October 2015, Trifacta also launched a free and simplified version of big data cleaning software, which currently has more than 5000 users in 3000 companies in 105 countries around the world.

Plan 3: intelligent fitting room

Introduction to the solution:

The smart fitting room helps users find clothes that suit their size, color and consumption scene through mirrors with touch screens and lighting adjustments. The main hope is to enhance the experience of offline services through intelligent means and hands-on experience, and create a differentiated competitiveness of offline services compared with online e-commerce.

Solution details:

Consumers enter the store, browse all the goods in the store through a mirror, submit a fitting application, and the clothes will be placed in the fitting room by the shopping guide. Customers can adjust the brightness and color of the lights to simulate the use of the scene, the mirror senses the RFID tag on the clothes and displays it on the screen, and then the mirror gives matching suggestions.

If you need to try clothes of other colors or sizes, you can also send them through the instructions on the screen and ask the shopping guide to send them.

When customers are satisfied with trying on, they can pay through Paypal directly on the mirror, and the clothes they have tried on will be saved in their personal account. Kinect is also installed in the fitting room to record and track the movements of the dresser.

Solution 4: malware protection software-CylancePROTECT series

Introduction to the solution:

The existing network threats are designed to bypass the existing protection measures, but simple post-detection is difficult to ensure security. The company uses a mathematical model to give its risk factors for each document, based on which machine learning algorithms are used to distinguish between "good" files and "bad" files. Use the extensible big data architecture to identify file patterns.

The solution is suitable for different industries, such as critical infrastructure, education, energy, finance, health, retail, federal government, etc.

Solution details:

Machine learning requires a total of four stages: collection, extraction, learning and classification.

1. Learn to use files from various industries covering a variety of formats and authors for manual training. The files are divided into three parts: known and validated, known and verified maliciously and unknown. The classification of these files needs to be guaranteed to be correct, otherwise it will lead to deviation.

two。 Extract the most likely feature set that uses the computing power of the machine and data processing technology to identify the file. Extract the unique features of the file according to its file type (.exe, .dll, .com, .pdf, .java, .doc, .xls, .ppt, and so on). Eliminate the deviation of manual classification. Thousands of features are used to identify malicious files, which greatly increases the cost of malware makers

3. After the learning attributes are collected and normalized, the irrelevant features are eliminated by vectorization and machine learning to speed up the analysis. Mathematicians establish statistical models according to the classification results and calculate the risk factors of each document.

4. The classification is based on the risk factors of the documents obtained by the statistical model, and the credibility of the users to which the files belong is analyzed to achieve optimization.

Scheme 5: data modeling platform-- DataBrain

Introduction to the solution:

DataBrain data modeling platform includes machine learning technology. According to its data analysis service system, a data science workflow is established to help data analysts, data engineers and business analysts improve their modeling ability.

The platform provides models suitable for credit risk control, precision marketing, personalized recommendation, investment strategy and customer clustering, which can help banks and financial institutions manage business and improve operational efficiency.

Solution details:

The DataBrain data analysis system consists of the following seven steps:

1. Problem definition: select a problem that is suitable for solving with a data model, and transform a special business problem into a data science problem.

two。 Data preparation: extract key data from multiple data sources, clean, process and process key data

3. Algorithm tuning: select the algorithm suitable for specific data, automatically find the optimal parameters, and establish an efficient data model.

4. Knowledge discovery: transform the data model into the basis of decision-making, and discover new knowledge other than business experience from the data.

5. Effect analysis: analyze the effectiveness and computational efficiency of the model based on different effect measurement indicators, and evaluate the business value of the model.

6. Online deployment: standardize model input and output +, seamlessly connect the model with the business system, and learn automatically from massive data

7. Model update: monitor model stability, model effect, model logic and data changes, collect feedback data, and constantly update the model.

While the data modeling platform has built-in automatic parameter tuning robot Atom, when modeling, it will find the corresponding parameters of different algorithms, find the optimal algorithm and parameter combination, and improve the efficiency of parameter adjustment. Atom automatically distributes the tasks of parameter adjustment and selection algorithm to multiple computing nodes through distributed computing, which greatly shortens the computing time, and monitors the model effect in real time and iterates online. In addition, the platform can automatically extract data features from multiple data sources, transform the data analysis results into business knowledge, and provide a basis for customer decision-making.

Https://www.toutiao.com/i6776540749019218445/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.