40+ Business Learning Technology Applications
Translation of a post by Philip Hodgett, who spoke at the recent Hollywood Professional Association Tech Retreat conference. I hope that a list of relevant services that are ready for integration into your projects and examples of running a business based on machine learning collected in one place will be useful for developers. I propose to share your own results of successful implementation of projects related to in-depth training.
Similar services are at Google and Microsoft Cortana .
M.O. used in healthcare to predict the waiting time for a patient in the emergency room waiting room . For prediction, factors such as staffing levels, patient data, emergency room schedules, and even room plans are used.
The Online Privacy Foundation sponsored a tweet-based psychopathy detection contest , and the results are encouraging .
')
Researchers from IBM have learned how to extract the criteria for diagnosing heart attacks from the text of medical records .
Singapore startup launched an application that can send an alarm message when you shake the phone . The algorithm of M.O. in order to distinguish the actual gesture of calling for medical aid from the usual movements of the phone.
Google’s depth learning algorithm was used to diagnose cancer , and the results were astounding (48% clinical accuracy, Google’s scoring 89%).
Indicates a patient with a high risk of re-admission to hospital .
Researchers at Stanford have trained a neural network to detect skin cancer from photographs .
The tasks of automatic machine translation are relevant for a very long time, but in-depth training has proven to be most effective in the following areas:
The task of generating handwriting for an arbitrary phrase, provided training on a large set of samples of manuscripts.
The interesting task of generating text based on the analysis of a large body of texts. Known methods for generating text word-for-word and literal generation. Models are trained in grammar, punctuation, sentence formation, and even imitate the style of body texts.
Text classification or thematic modeling allows thousands of news notes to be automatically grouped into news aggregators. It is also used to group keywords within a given taxonomy.
H & R Block trained the IBM Watson machine to search for optimal tax deductions .
The technology will be able to read tens of thousands of case histories and isolate the duration of inpatient treatment, medical appointments and procedures before calculating insurance payments .
Dunnhumby trying to predict whether the withdrawal of goods to the market successful .
Benchmark Solutions is trying to predict the value of US corporate bonds .
Legal Robot translates the legal text into plain human, and they try to determine which clauses are missing in the contract, whether there are any redundant clauses, such as a waiver of royalties or non-disclosure agreements.
Other NechCrunch articles on AI in jurisprudence ...
PayPal uses in-depth training to prevent fraud and money laundering at all levels of detail. The company is able to accurately detect unscrupulous buyers and sellers.
Machine learning is used to detect a variety of transactions that do not meet current business practice in a huge data stream. For example, the detection of insider trading in the stock market.
Machine learning can improve customer service by understanding the exact needs and problems of the client. Predictive analytics solutions provider Lumidatum reports that it can easily distinguish a customer starting to use your product from an experienced user, as well as recognize problems and begin a pro-active response as they arise.
The system on the basis of deep learning synthesizes sound corresponding to the video sequence.
The task of automatically describing a given image with text is marked by the explosive growth of publications since 2014. Now, if your Facebook page loads slowly, you see an automatically generated description of the photos.
In-depth training is used for coloring objects according to the surrounding context, in the same way as colorist artists work.
The Facebook Lumos computer vision platform is used to organize image search by content . This means that users can find images not only by tags and text captions, but by the description of objects in the images.
Jukedeck is one of the many companies involved in writing music using artificial intelligence. They train neural networks by completing assignments, much like a child is learning.
Amazon sponsored a contest designed to resolve the issue of the possibility of automating the assignment and cancellation of access rights for staff.
Video surveillance operators can skip dangerous items, but they cannot hide from machine learning! Machine learning is able to flexibly adjust to seasonal changes in baggage and its contents, as well as to the special requirements of controlled premises. Www.qylur.com aims to reduce the number of false positives.
According to Kaspersky Lab reports, in 2014 they detected over 325 thousand new malicious files every day. Only machine learning can cope with such volumes, especially given the fact that most new infections differ from old ones by 2%.
Trying to determine for myself how we could use machine learning in our software business, I made this list. I was slightly shocked by the variety of ways M.O. According to TechCrunch , more than 10 billion dollars have already been invested in 1,500 startups related to M.O. and artificial intelligence. In 2017, this amount is projected to increase fourfold! I wanted to share with you this list ...
IBM Watson Knowledge Services
- Speech recognition
- Speech synthesis
- Natural language processing (mood, keywords, named entities, high-level concepts)
- Classification of natural language (understands the meaning of the text and returns the appropriate classification)
- Chat bots
- Dialogue (scenarios of extensive communication between the user and the application)
- Machine translate
- Judgments about the person on the basis of how a person writes (to search for people, products, opportunities and the user interface)
- Search and rank the most useful information from the collection of documents
- Analysis of tonality (using linguistic analysis to determine emotions, social trends, style, emotional context of conversations / communication)
- Recognition of visual images (understanding of image content for tagging, searching for individuals, determining gender and age group, searching for similar images in the collection; learning system for setting up in specific applications; TheTake launched on the website to purchase products noticed by users in movies ).
Similar services are at Google and Microsoft Cortana .
The medicine
Prediction of waiting time in the emergency room waiting room
M.O. used in healthcare to predict the waiting time for a patient in the emergency room waiting room . For prediction, factors such as staffing levels, patient data, emergency room schedules, and even room plans are used.
Prediction of psychopathy
The Online Privacy Foundation sponsored a tweet-based psychopathy detection contest , and the results are encouraging .
')
Heart attack detection
Researchers from IBM have learned how to extract the criteria for diagnosing heart attacks from the text of medical records .
Call for help with stroke and epileptic seizures
Singapore startup launched an application that can send an alarm message when you shake the phone . The algorithm of M.O. in order to distinguish the actual gesture of calling for medical aid from the usual movements of the phone.
Cancer diagnosis
Google’s depth learning algorithm was used to diagnose cancer , and the results were astounding (48% clinical accuracy, Google’s scoring 89%).
Prediction of repeated visits of patients to the hospital
Indicates a patient with a high risk of re-admission to hospital .
Diagnosis of skin cancer
Researchers at Stanford have trained a neural network to detect skin cancer from photographs .
Word processing
Machine translate
The tasks of automatic machine translation are relevant for a very long time, but in-depth training has proven to be most effective in the following areas:
- Automatic translation of texts
- Automatic image translation
- How Google put deep learning in the phone
Handwriting generation
The task of generating handwriting for an arbitrary phrase, provided training on a large set of samples of manuscripts.
Spawn text
The interesting task of generating text based on the analysis of a large body of texts. Known methods for generating text word-for-word and literal generation. Models are trained in grammar, punctuation, sentence formation, and even imitate the style of body texts.
- Unexplained efficiency of recurrent neural networks
- Automatic generation of SEM texts by recurrent neural networks.
Classification
Text classification or thematic modeling allows thousands of news notes to be automatically grouped into news aggregators. It is also used to group keywords within a given taxonomy.
Business and Law
Tax optimization
H & R Block trained the IBM Watson machine to search for optimal tax deductions .
Calculation of insurance payments
The technology will be able to read tens of thousands of case histories and isolate the duration of inpatient treatment, medical appointments and procedures before calculating insurance payments .
Predicting success when entering the market
Dunnhumby trying to predict whether the withdrawal of goods to the market successful .
Stock price prediction
Benchmark Solutions is trying to predict the value of US corporate bonds .
Understanding legal texts
Legal Robot translates the legal text into plain human, and they try to determine which clauses are missing in the contract, whether there are any redundant clauses, such as a waiver of royalties or non-disclosure agreements.
Other NechCrunch articles on AI in jurisprudence ...
Money laundering prevention
PayPal uses in-depth training to prevent fraud and money laundering at all levels of detail. The company is able to accurately detect unscrupulous buyers and sellers.
Anomaly detection
Machine learning is used to detect a variety of transactions that do not meet current business practice in a huge data stream. For example, the detection of insider trading in the stock market.
Customer service improvement
Machine learning can improve customer service by understanding the exact needs and problems of the client. Predictive analytics solutions provider Lumidatum reports that it can easily distinguish a customer starting to use your product from an experienced user, as well as recognize problems and begin a pro-active response as they arise.
Image processing
Automatic dubbing of silent films
The system on the basis of deep learning synthesizes sound corresponding to the video sequence.
Generation of text descriptions
The task of automatically describing a given image with text is marked by the explosive growth of publications since 2014. Now, if your Facebook page loads slowly, you see an automatically generated description of the photos.
- A picture is worth a thousand words: building a natural description of images
- Rapid progress in the automatic description of images.
- Youtube reports that more than one billion videos are automatically described in 10 languages.
Colorization of black and white images
In-depth training is used for coloring objects according to the surrounding context, in the same way as colorist artists work.
Convert images to photos
Search for images by content
The Facebook Lumos computer vision platform is used to organize image search by content . This means that users can find images not only by tags and text captions, but by the description of objects in the images.
Other Machine Learning Uses
Writing music
Jukedeck is one of the many companies involved in writing music using artificial intelligence. They train neural networks by completing assignments, much like a child is learning.
Personnel access control
Amazon sponsored a contest designed to resolve the issue of the possibility of automating the assignment and cancellation of access rights for staff.
Total video surveillance
Video surveillance operators can skip dangerous items, but they cannot hide from machine learning! Machine learning is able to flexibly adjust to seasonal changes in baggage and its contents, as well as to the special requirements of controlled premises. Www.qylur.com aims to reduce the number of false positives.
Anti-spam and malware
According to Kaspersky Lab reports, in 2014 they detected over 325 thousand new malicious files every day. Only machine learning can cope with such volumes, especially given the fact that most new infections differ from old ones by 2%.
Source: https://habr.com/ru/post/324694/
All Articles