Source: ATYUN AI platform
The amount of data generated each day to more than incredible. IDC predicts that by 2025 the total will grow to 5.2 ZB (zettabyte is equivalent to 250 billion DVD.), And it is growing exponentially, only in the past two years, resulting in 90% of the existing data.
Data are helpful for anyone, especially the scientists responsible for data use the data to the training, validation and testing machine learning system. Two years ago, in order to keep things simple, software engineers Yosi Taguri with three colleagues --Shay Erlichmen, Joe Salomon, and Rahav Lussato co-founded Missinglink.ai.
"We are at an incredibly critical point, we need to address all the important issues need to be addressed, for example, to save lives by detecting cancer, and provide safer on the streets, more intelligent driving," Taguri said. " However, to find the meaning of all data through is very difficult, require too much manpower .MissingLink allow each engineer an unprecedented way to build complex AI machines. "
To this end, MissingLink.ai to-end management and deployment tools that simplify coding and model training process. It supports popular machine learning framework, such as Google's TensorFlow, Facebook's Caffe2, PyTorch and Keras, and synchronize data changes immediately, without having to manually copy the files. As the system automatically entrusted to experimental and computational resources available to run in parallel, they only need three lines of code to set.
"We are moving away from a lot of clumsy work, so that they can focus on bigger issues," Taguri said.
One of the highlights is its suite of powerful data management engine, Taguri called "version aware". In essence, it changes the mapping database over time, allowing engineers to query data for model training and compared against a specific version to run. It also cached in local streaming data and, using the CPU replicate experiments on the GPU.
Missing.AI another title is its visualization dashboard, finishing it ongoing experiments in the list view, which includes the start time, machine or cloud instance each test run, and total run time schedule, and other useful indicators. After the completion of the experiment, the results including source code, visualization and resources to automatically record to their offspring.
"Learning takes a lot of depth," Taguri said, "the company learned that it's not easy. They are basically insist on doing DevOps work, such as mobile data, tracking experiment, and try to get the machine running and GPU."
Taguri claimed that one of the customer's productivity has increased 20-fold.
MissingLink.ai providing a 1GB data hosting, a free hosting plan resources and a hosting organization. The most expensive cost $ 120 per month payment plan, and store and manage resource constraints were upgraded to 100GB and 5GB.
Taguri said, "One of our core principles to keep in mind is that we should not educate data scientists how deep learning, we should integrate seamlessly into their workflow."
This switched ATYUN artificial intelligence media platforms, the original link: MissingLink AI simplify data management, coding, and model training process
more recommendations
Artificial intelligence can become a tool for video surveillance to combat crime
Embedded deep learning: from cloud to device
Musk: We only up to a 10% chance to make the safety of artificial intelligence
A comprehensive restructuring, Samsung created a new Research Center for Artificial Intelligence
