CVEDIA Wins a 2020 Edison Award for its Synthetic Data Algorithms to Train AI

CVEDIA Wins a 2020 Edison Award for its Synthetic Data Algorithms to Train AI

Author: Eric Walz   

On April 8, 2020, the annual Edison Award winners were announced, which recognize innovations each year in the fields science, technology, design and engineering. The Edison Awards was founded in 1987. 

Among this year's recipients was Arlington, Virginia-based company CVEDIA. The company received a bronze award for the AI Applications category for its synthetic algorithms for training AI and machine learning models. CVEDIA's synthetic algorithms were chosen as a winner by a panel of over 3,000 leading business executives from around the world.

CVEDIA is a software as a service (SaaS) company that specializes in creating synthetic machine learning algorithms for computer vision applications where data is limited or unavailable. The company creates synthetic data for computer vision applications that normally require massive amounts of training data

The company's products are used by 30 Global 500 companies, many of which are working in Silicon Valley on autonomous driving technology. 

The company's technology uses computer-generated imagery to train AI algorithms. This breakthrough allows clients to purchase off-the-shelf AI applications for applications where data collection isn't possible. CVEDIA creates custom datasets for each client's needs. 

"After a thorough review, the Edison Awards Judges recognize CVEDIA SynApps algorithms as a game-changing innovation standing out among the best new products and services launched in their category," said Frank Bonafilia, Executive Director of the Edison Awards.

CVEDIA developed a powerful simulation platform called "SynCity", which is used to generate data for neural network training and validation. Developers can use SynCity to generate highly random scenes, conditions, and metadata for training machine learning models, allowing developers of autonomous vehicles to test out their software in the safety of a simulator instead of on public roads. 

For developers of autonomous vehicles, SynCity enables real-time Hardware-In-the-Loop (HWIL), Human-In-the-Loop (HITL) or Software-In-the-Loop (SIL) simulations even for complex sensor configurations. These types of computer simulation are also being developed by chipmaker NVIDIA for autonomous machines. Applications include perception systems used by autonomous vehicles and advanced driver assist systems (ADAS) found on modern vehicles. 

Using Synthetic Data for Training AI

Although its been around for nearly three decades, synthetic data is being used more frequently today for advanced machine learning applications, especially to train camera and sensor-based AI, which is used for autonomous vehicle perception systems and automotive safety features such as vehicle and pedestrian detection. These systems however, require a significant amount of training data.

The data is computer-generated rather than being manually collected from cameras and other vehicle sensors. Because the synthetic data is attached to its source, the origin can be easily tracked. 

Synthetic data also provides a means of benchmarking other data by allowing developers to compare it to see what discrepancies exist. This process is known as cross-validation, to see how well a machine learning model performs with the data it's given.

How it Works

A machine learning model is first trained on a synthetically generated dataset with the intention of transfer learning to real data. Once the synthetic environment is ready, it is fast and inexpensive to produce as much training data as required to achieve the desired accuracy of the machine learning algorithm. 

Another benefit to using synthetic data is that it can have perfectly accurate labels, eliminating the time consuming task of manually labeling data. Most training data is labeled manually by humans. 

For example, an AI-based computer vision system used for perception must be taught what it's seeing. When trained with images, all of the relevant objects in the image are identified and highlighted by bounding boxes. Each object is then labeled so a computer knows what it's seeing.

Synthetic data can be substituted for labeling that may be very expensive or impossible to obtain by conventional methods, which are known as edge cases. In addition, the synthetic environment can be modified as needed to improve the model and training.

Using synthetic data also improves security for applications that call for a higher level of overall security. Synthetic data can be used as a substitute for certain real data that contains sensitive information. 

For example, a public dataset of vehicle images used to train perception systems may contain sensitive information. This is one of the reasons Google Maps blurs the license plates of vehicles and the faces of people that appear in its Street View images.

CVEDIA says its synthetic algorithms outperform traditional algorithms because synthetic data allows for feature-based design. That means the company's models are safer and quicker to market. The synthetic algorithms are rigorously validated by data scientists and customizable without requiring further data collection.

To demonstrate real-world application of its technology, CVEDIA created a neural network that's never been provided training data of actual cars. It was trained using only synthetic data, skipping the need for expensive training data collection and annotations. Yet it was still able to correctly identify vehicles in moving traffic.

CVEDIA said it took less than 48 hours to build, train and demonstrate the neural network. 

The company's work extends beyond the automotive industry. CVEDIA creates neural networks across multiple industries and can customize solutions depending on the scope of the project.

Eric Walz
Eric Walz
Originally hailing from New Jersey, Eric is a automotive & technology reporter covering the high-tech industry here in Silicon Valley. He has over 15 years of automotive experience and a bachelors degree in computer science. These skills, combined with technical writing and news reporting, allows him to fully understand and identify new and innovative technologies in the auto industry and beyond. He has worked at Uber on self-driving cars and as a technical writer, helping people to understand and work with technology.
Prev:Autonomous Driving Startup Zoox Agrees to Settle Lawsuit Filed By Tesla Over Stolen IP Next:Honda Partners With SNAM to Expand EV Battery Recycling Plan
    view more