Source: ATYUN AI platform
The new start-up companies Gamalon, Inc. Founder and CEO Ben Vigoda created a technology that makes writing process automation model, the technology shift from tool data scientist for the full machine learning techniques to learn from the data by itself.
Bernadette Esposito
Andrew Ng is one of Google Brain ancestors, but also in recent years, advocates revival of neural networks, he tells his origin story in TED speech. He said he tried to create an autopilot helicopter, but no matter how hard he studies his mathematical model of helicopter aerodynamics and mechanics, he could not let the helicopter fly by helicopter mathematical model of programming. Then, he decided to rely only helicopter sensor data to train a machine learning algorithm, helicopter fly very well. Ng said he learned from this experience is not complicated to try to write their own model, but on deep learning algorithms.
Located in Cambridge, Massachusetts's new start-up companies Gamalon, Inc. Ben Vigoda, founder and CEO Andrew Ng and form almost the same experience, but he came to completely different conclusions. Prior to his MIT machine learning chip startup Lyric Semiconductor, Inc. The company later acquired by ADI, took a long time and a lot of money to develop mathematical models for work signal processing applications. But Vigoda decided not to give up writing mathematical model. Instead, he believes what is needed is for the development and debugging of mathematical models and make it faster to adapt the tool data. Vigoda said, "Write these models to make them fit the data, what is wrong, then improved model is slow and arduous. The idea of trying to improve the model may take several weeks, usually it takes several years to work into the system. For conventional programming, we have all these technical compiler, debugger, profiler, etc., can be easily and quickly write a program to improve you are writing this support agile development environment for programming but mathematics (Bayesian) model - scientists to they adapt to the kind of model of written data, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective we have all these technical data compiler, debugger, profiler, etc., it can be easily and quickly write a program to improve you are writing this support agile development environment for programming but mathematics (Bayesian) model -... for scientists adapt their data models written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective the data we have all these technical compiler, debugger, profiler, etc., can be easily and quickly write a program to improve you are writing this support agile development environment for programming but mathematics (Bayesian) model - Scientists in order to meet their data model that written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. This development environment supports agile programming. But for mathematics (Bayesian) model - the model that scientists in order to meet their data written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. This development environment supports agile programming. But for mathematics (Bayesian) model - the model that scientists in order to meet their data written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective data. "the first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. Scientists in order to meet their data model that written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. This development environment supports agile programming. But for mathematics (Bayesian) model - the model that scientists in order to meet their data written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective data. "the first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. Scientists in order to meet their data model that written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. This development environment supports agile programming. But for mathematics (Bayesian) model - the model that scientists in order to meet their data written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective data. "the first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. Scientists in order to meet their data model that written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective data. "the first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. Scientists in order to meet their data model that written, we do not have any similar development tools. "Vigoda created Gamalon, and with funding from DARPA, and start building these tools." The first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data. "The first result is that we can quickly build and test the model. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which helps to explain the most effective data. "the first result is that we can quickly build and test models. We can write science (Bayesian) model and the data almost immediately tested to see which parts of the model and the model which most effectively contribute to interpret the data.
But that is just the beginning. When they realize they can begin to replace human modeler, the real "aha" moment came. "We found that we can actually be used for human development tools to build computer models to guide their changes, and self-test these new models. By automating the process of preparing the model, we will shift from technology as a tool for data scientists fully mature machine learning technology, which can be learned from the data. the system has performed very well in the traditional machine learning tasks such as image recognition and natural language processing. "Gamalon will this new invention called Bayesian integrated program (BPS), its performance and depth of learning is completely different. Compare drawing-aware applications that the company has a demo video, graphics system learn to recognize and Google DeepMind named "Quick, Draw!" Side by side. Gamalon the BPS system only from the number of training examples to learn, rather than the thousands or millions, from a person instead of thousands of people, on the iPad instead of running on hundreds of servers, and almost immediately rather than spend learning the number of days, weeks or months to complete the study calculated. These performance improvements are expected to make the machine learning community attention. And learn almost immediately, rather than spending a few days, weeks or months to complete its study computing. These performance improvements are expected to make the machine learning community attention. And learn almost immediately, rather than spending a few days, weeks or months to complete its study computing. These performance improvements are expected to make the machine learning community attention.
Gamalon now declare their use BPS technology, the first two commercial applications Gamalon Structure and Gamalon Match created. "This is a machine learning as a service," he explained. "We can host it on any of the major cloud providers."
The company can use the service to build, clean up, prepare and integrate data from different sources. Vigoda said: "More than 90 percent of corporate data can not be used because it is unstructured company has a large free-form text such as product descriptions, customer name and address, has been converted to a voice query text, explain insurance claims, doctor. notes, etc. they need to convert each text block database row with the right column. now is not a good way to do this, so the company mechanical Turk or outsourced to professional service firms, they will encounter a lot of mistakes. we new products Gamalon structure solved the problem of data structure. "
In addition, if you want to link and integrate multiple data sources together, you need to use data integration and data preparation software, "Vigoda explains." Then, you pay ten times the cost you pay for the software in order to review the results and eliminate errors and redundant. Integration takes months, then you can make dozens review. "Gamalon Match solved the problem of data integration.
A Gamalon customers with hundreds of physical stores in the US, and want to use Gamalon for home delivery service build inventory system. "They need to link to their store and find products on the shelves," Vigoda said. It may sound simple, until you consider "How many different ways to describe diet cola situation," Vigoda adds. "Our system access to all databases, reading all of the databases, and to determine the data available in each store, so when the driver reaches the store, they know the product will wait."
At the same time, a manufacturing and wholesale customers "want to know what their hundreds of dealers and their sales are," Vigoda said. "So, we enter and connect all these databases, they will be aligned with the list of contracts, and how to get the product through its distribution channels of incredible views."
Gamalon ultimate goal is to provide "ubiquitous middleware layer" for all SaaS software, Vigoda said. "Today, there are hundreds of different enterprise SaaS application is available for each application that stores data in different ways, each company to buy different combinations. We can provide a single global view, understand your company's sales and sales targets, inventory sources, and to pay your expenses, without the need to migrate to a single centralized database system. we can replace the database system records the record company with machine intelligence information indexes. we are delighted that these first products with us where to go! "
This switched ATYUN artificial intelligence media platforms, the original link: Gamalon invention and commercialization of a revolutionary machine learning techniques
more recommendations
Machine Learning Basics Tutorial: Neural Networks
AI chip start-up companies has shown explosive growth trend
Stanford Report: AI smarter than 5-year-old human, not even as a mouse
Also found a loophole! You can cheat AI facial recognition system by wearing glasses
