June 6, 2017 at 12:2740+ business machine learning technology applications
Translation fasting Philip Hodzhetta, speaking recently at a conference Hollywood Professional Association Tech Retreat. I hope that the list of relevant services ready for integration into your projects and examples of a working business based on machine learning gathered in one place will be useful to developers. I propose to share your own results of the successful implementation of projects related to deep learning.
Google and Microsoft Cortana have similar services .
M.O. used in healthcare to predict waiting times for a patient in an emergency room . For prediction, factors such as staffing levels, patient data, emergency room schedules, and even room plans are used.
The Online Privacy Foundation has sponsored a psychopathic tweet detection contest , and the results are encouraging .
Researchers at IBM have learned to extract criteria for diagnosing heart attacks from the text of medical records .
Singaporean startup has launched an application that can send an alarm message when shaking the phone . The algorithm of M.O. in order to distinguish a valid gesture of calling a medical aid from the usual movements of the phone.
Google’s in-depth learning algorithm was used to diagnose cancer , and the results were astounding (clinical accuracy 48%, Google scoring 89%).
Indicates patients at high risk of re-visiting the hospital .
Researchers at Stanford trained a neural network to detect skin cancer from photographs .
The tasks of automatic machine translation have been relevant for a very long time, but in-depth training has proven to be most effective in the following areas:
The task of generating handwritten text for an arbitrary phrase, subject to training on a large set of sample manuscripts.
An interesting task of generating text based on the analysis of a large body of texts. Known methods of generating text word-for-word and letter-by-letter generation. Models learn grammar, punctuation, sentence formation and even imitate the stylistics of body texts.
Text classification or thematic modeling allows thousands of news items to be automatically grouped in news aggregators. It is also used to group keywords within a given taxonomy.
H&R Block trained IBM Watson to find the best tax deduction .
The technology will be able to read tens of thousands of medical histories and isolate the duration of inpatient treatment, medical appointments and procedures before accruing insurance benefits .
Dunnhumby is trying to predict whether the launch of the product on the market will be successful .
Benchmark Solutions is trying to predict the value of US corporate bonds .
Legal Robot translates the legal text into simple human, and they try to determine what provisions are missing in the contract, whether there are any extra provisions, such as a waiver of royalties or a non-disclosure agreement.
Other NechCrunch articles on artificial intelligence in law ...
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 established business practices in a huge data stream. For example, the discovery of insider trading in the stock market.
Machine learning can improve customer service by understanding the exact needs and problems of the customer. Lumidatum predictive analytics solutions provider reports that it can easily distinguish a customer starting to use your product from an experienced user, as well as recognize problems and start a proactive response as they arise.
The system based on deep learning synthesizes the 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 will see an automatically generated description of the photos.
In-depth training is used to color objects based on their surroundings, similar to how color artists work.
The Facebook Lumos computer vision platform is used to organize image searches on content . This means that users can find images not only by tags and text signatures, but by the description of objects in the images.
Jukedeck is one of many companies that write music using artificial intelligence. They train neural networks by completing assignments, much like a child learns.
Amazon sponsored a competition designed to address the issue of the possibility of automating the assignment and revocation of access rights for personnel.
Operators of video surveillance systems can skip dangerous objects, but they can’t hide from machine learning! Machine learning is able to flexibly adjust to seasonal changes in baggage and its contents, as well as the special requirements of controlled rooms. Company www.qylur.com aimed at reducing the number of false positives.
According to Kaspersky Lab , 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 to use M.O. According to TechCrunch , more than $ 10 billion has already been invested in 1,500 startups related to M.O. and artificial intelligence. In 2017, an increase of four times is predicted! I wanted to share this list with you ...
IBM Watson Learning Services
- Speech recognition
- Speech synthesis
- Natural language processing (mood, keywords, named entities, high-level concepts)
- Natural language classification (understands the meaning of the text and returns the corresponding classification)
- Chat bots
- Dialog (scenarios of branched communication between the user and the application)
- Machine translate
- Judgments about a person based on how a person writes (to search for people, products, opportunities, and to customize the user interface)
- Search and ranking of the most useful information from a collection of documents
- Analysis of tonality (using linguistic analysis to determine emotions, social trends, style, emotional context of conversations / conversations)
- Recognition of visual images (understanding the contents of an image for tagging, searching for faces, determining sex and age group, searching for similar images in a collection; a learning system for customization in specific applications; TheTake launched a website for shopping for goods noticed by users in films ).
Google and Microsoft Cortana have similar services .
The medicine
Prediction of waiting times in an emergency room
M.O. used in healthcare to predict waiting times for a patient in an emergency 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 has sponsored a psychopathic tweet detection contest , and the results are encouraging .
Heart attack detection
Researchers at IBM have learned to extract criteria for diagnosing heart attacks from the text of medical records .
Call for help with stroke and epileptic seizures
Singaporean startup has launched an application that can send an alarm message when shaking the phone . The algorithm of M.O. in order to distinguish a valid gesture of calling a medical aid from the usual movements of the phone.
Cancer diagnosis
Google’s in-depth learning algorithm was used to diagnose cancer , and the results were astounding (clinical accuracy 48%, Google scoring 89%).
Prediction of patient re-visits to the hospital
Indicates patients at high risk of re-visiting the hospital .
Skin Cancer Diagnosis
Researchers at Stanford trained a neural network to detect skin cancer from photographs .
Word processing
Machine translate
The tasks of automatic machine translation have been 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 has put in-depth training on the phone
Handwriting
The task of generating handwritten text for an arbitrary phrase, subject to training on a large set of sample manuscripts.
Text Generation
An interesting task of generating text based on the analysis of a large body of texts. Known methods of generating text word-for-word and letter-by-letter generation. Models learn grammar, punctuation, sentence formation and even imitate the stylistics of body texts.
- Inexplicable efficiency of recurrent neural networks
- Automatic generation of SEM texts by recurrent neural networks.
Classification
Text classification or thematic modeling allows thousands of news items to be automatically grouped in news aggregators. It is also used to group keywords within a given taxonomy.
Business and Law
Tax optimization
H&R Block trained IBM Watson to find the best tax deduction .
Insurance claim calculation
The technology will be able to read tens of thousands of medical histories and isolate the duration of inpatient treatment, medical appointments and procedures before accruing insurance benefits .
Prediction of success when entering the market
Dunnhumby is trying to predict whether the launch of the product on the market will be 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 simple human, and they try to determine what provisions are missing in the contract, whether there are any extra provisions, such as a waiver of royalties or a non-disclosure agreement.
Other NechCrunch articles on artificial intelligence in law ...
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 established business practices in a huge data stream. For example, the discovery 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 customer. Lumidatum predictive analytics solutions provider reports that it can easily distinguish a customer starting to use your product from an experienced user, as well as recognize problems and start a proactive response as they arise.
Image processing
Automatic dubbing of silent films
The system based on deep learning synthesizes the sound corresponding to the video sequence.
Spawning 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 will see an automatically generated description of the photos.
- A picture is worth a thousand words: building a natural description of images
- Fast progress in automatic description of images.
- Youtube reports that more than one billion videos are automatically described in 10 languages.
Black and white colorization
In-depth training is used to color objects based on their surroundings, similar to how color artists work.
Convert Drawings to Photos
Search for images by content
The Facebook Lumos computer vision platform is used to organize image searches on content . This means that users can find images not only by tags and text signatures, but by the description of objects in the images.
Other Machine Learning Applications
Music writing
Jukedeck is one of many companies that write music using artificial intelligence. They train neural networks by completing assignments, much like a child learns.
Personnel Access Control
Amazon sponsored a competition designed to address the issue of the possibility of automating the assignment and revocation of access rights for personnel.
Total CCTV
Operators of video surveillance systems can skip dangerous objects, but they can’t hide from machine learning! Machine learning is able to flexibly adjust to seasonal changes in baggage and its contents, as well as the special requirements of controlled rooms. Company www.qylur.com aimed at reducing the number of false positives.
Fighting spam and malware
According to Kaspersky Lab , 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%.