Technologies

ScF Network Deep Learning Algorithm

FPGA

PLK can provide an integrated autonomous driving solution in the form of
non-memory semiconductor Field-Programmable Gate Arrays (FPGA).
FPGA complements the performance of general-purpose processors and combines the
characteristics of custom semiconductors. Therefore, it is recently spotlighted tremendously.
FPGA provides customers with price competitiveness and fast processing capabilities.

A large amount of data required for machine learning implementation can be processed at
high speed without any confusion or bottlenecks of the system.
So it can provide highly accurate AI solutions.

GPU

PLK is developing a deep learning algorithm with a GPU that processes a large amount
of data at high speed. Through GPU that is advantageous to simultaneous calculation and
parallel computing, numerous image-recognition training sessions are performed to
improve recognition speed and accuracy. Mainly, we achieved remarkable performance
improvement in the pedestrian recognition field, and it can detect the near and far
pedestrians at real-time and the pedestrians covered or overlaid partially by another object.

Besides, we can prepare for various dangerous situations by simultaneously recognizing
different road objects, including vehicles, lanes, signs, and pedestrians.


ALGORITHM

CNN Convolution Neural Network

PLK can provide an integrated autonomous driving solution in the form of non-memory semiconductor Field-Programmable Gate Arrays (FPGA). FPGA complements the performance limitations of general-purpose processors and combines the characteristics of custom semiconductors. Therefore, it is recently spotlighted tremendously. FPGA provides customers with price competitiveness and fast processing capabilities.

RNN Recurrent Neural Network

RNN is a method of deep learning that is specialized in iterative and sequential data learning. It has an internal circulation structure. Therefore, it is suitable for handling data with sequential information. It can recognize the flow of regular data and extract abstract information, which can be used to analyze continuous input data such as video, voice, and language. Even if learning progresses, it is advantageous to reflect the continuous flow of information in learning without losing information of past learning. In recent years, dynamic data has been increasing in various industrial fields, and PLK is going to utilize it for the automotive external/internal image and voice recognitions.

RL Reinforcement Learning

RL is a learning method that improves one's behavior according to the mutual relationship between oneself and the environment. Likewise doing complimented behaviors more and doing punishable behaviors less, it means reinforcing learning through adaptability. It is the method used a lot in robots and artificial intelligence. Since the method learns through the process that gives a reward if followed well and gives punishment if not followed well, the input and the output do not have a clear relationship like CNN and RNN and it learns the method to maximize rewards through a series of steps.

GAN Generative Adversarial Network

GAN is a method of learning that draws attention as the way to lead future deep learning. It is not a supervised learning method that learns through various data like CNN and RNN, so it is considered to be an unsupervised learning method. Unlike the way that learns the pair of data and label, it is an active way of gaining knowledge from unlabeled data itself and identifying the characteristics of things on its own. Through this method, out of recognition and distinction, it can create on its own, which can show a variety of benefits including image creation, editing, conversion, and restoration.