회사소개

NATURAL LANGUAGE PROCESSING

Research

Resource management is crucial to serving deep learning models to a wide range of users.

When deep learning is applied to mobile devices, drones or robots, there is a limit to available resources.
To maximise the user experiences, we need technology and efforts to properly distribute the resources between cloud server and user devices.
Of course, the research to achieve the world best deep learning performance is very important, but it also requires the high level of engineering to get these amazing AI technologies widely applied in real life. We do the work of presenting technology to users.

Cloud Server
It should automatically allocate resources to places where necessary and reduce unused resources precisely.

We provide services via consistent web API interfaces in IoT, mobile devices or PCs.
We design the API interfaces to support various tasks such as object detection, action recognition and voice recognition etc.
Client
We develop services and applications using built-in neural engines on Android or iOS.
We develop applications that run deep learning models on embedded devices such as ROS.
Data management
We manage training data set systematically and keep security of it.
We collect and keep data systematically so that it can be used in the next machine learning training based on users’ activity.

Cases of application

We have applied the machine learning technologies of object recognition, age/gender recognition and identity recognition to the embedded devices with Nvidia Jetson TX2 and Xavier or ROS platforms, and developed the applications using them. We operate a service that recognizes age/gender on the website.

  • 01

    Data Collection

    We enlarge the dataset by collecting the data from age/gender server and use it to train ML model to make it more fit to Korea.

  • 02

    Robot Application

    Using the camera image from Anki Vector, we are developing a demo application that responds to behavior recognition.