Enabling Various Data Analysis Using AI

ReNom | Deep Learning

IoT has progressed in many places, and vast amounts of data are being collected every day. AI can recognize various characteristics and features of existing data, replace human inference, and analyze high dimensional data which is beyond human capability. Beneficial information can be acquired from data analysis of image, video, sensor, text or other datasets using several Deep learning methods.


Computer Vision


In the field of computer vision, algorithms such as Fast RCNN, CNN, SSD, YOLO, and segmentation can be used for many applications such as tracking people and cars from camera images, predicting mechanical part failures in the industries, detecting objects from drone aerial photography, and supporting diagnostics by analyzing X-ray and CT scan images in the medical field.


Time Series Data Analysis · Natural Language Processing

Periodic data can be acquired through analyzing data obtained from continuous
observations. By learning the phenomena from multiple time dependent datasets, such as operation data of industrial machines or power plants, the probability of the next event can be predicted. It can help users to develop optimal maintenance cycle, advanced preventive maintenance etc. With natural language processing, users can analyze textual data, customer feedback, or generate texts automatically using information from previous words.


Hyperparameter Search

When constructing Deep Neural Network, data scientists are required to tune
hyperparameters by trial and error such as the number of units in the hidden layer or the learning rate. ReNom can automatically search the hyperparameter which reduces the burden on developers and time required for analysis. Hyperparameters can be searched by setting the search methods to Grid search, Random search, Bayesian search, etc.


Generative Model

In machine learning and deep learning, many datasets are used for classification and regression problems. Generative model, on the other hand, learns the distribution of the data, thus they are effective when the dataset is insufficient.Generative models can learn the probabilistic model that generates
the feature data of a particular class. By sampling data from the probabilistic model, a data of a particular class can be generated.



To achieve a model for identification or prediction purpose with high accuracy,
preprocessing collected raw data into a suitable form for analysis and learning is
essential. However, data scientists are tasked to build programs individually to
preprocess raw data into an appropriate data, conduct analysis, and improve
accuracy, which takes great deal of effort. ReNom has various preprocessing utilities that can support data scientists in analyzing tasks.


Algorithm Mix

We believe that using a single method (algorithm) is not suitable for diversified and complicated problems. ReNom is based on the concept of Algorithm Mix: a combination of multiple algorithms in a single platform which can be applied to tasks of varying businesses and industries. In ReNom, the necessary algorithms for AI development are implemented, such as Deep Learning and Reinforcement Learning, TDA (TopologicalData Analysis), etc.


Page top