Beamr Boost for Machine Learning

Accelerating Neural Networks Training

Herzliya Israel, March 20, 2024 (GLOBE NEWSWIRE) -- Beamr Imaging Ltd. (NASDAQ: BMR), a leader in video optimization technology and solutions, today announced that it has released the results of a case study which highlights how Beamr tech enables accelerated machine learning training by using significantly smaller video files and without any negative impact on the video artificial intelligence (AI) process.

Machine learning for video is becoming an increasingly significant technology for businesses. But the players in this expanding arena face critical pain points, like storage and bandwidth bottlenecks or the difficulty to reach acceptable training and inferencing speeds.

In this case study, Beamr’s R&D team showed that training an AI network using video files compressed and optimized through Beamr’s Content-Adaptive Bitrate technology produced results that are as good as training the network with the original, larger files. The AI network was trained to fulfill the task of action recognition, such as distinguishing between people who are walking, running, dancing or doing many other day-to-day actions.

Beamr CTO, Tamar Shoham, explained: “It was important to us to define a test case that really uses the fact that the content is video, instead of an image. When viewing individual frames, it is not possible to differentiate between frames captured during walking and running, or between someone jumping or dancing. Therefore, in order to classify videos according to the action they show, the temporal component is needed, which is why 'action recognition' was our task of choice”.

The video files used for machine learning training were optimized by Beamr Cloud, reducing file sizes by 24%-67%. Such a reduction is beneficial when storing video files for future use and possibly performing other manipulations on them. Recently-launched Beamr cloud is an optimization and modernization software-as-a-service (SaaS) that enables automated, efficient and fast video processing, through no-code processes or customized pipelines to meet specific user needs.

Training performed with the smaller video files optimized by Beamr tech, provided results which were equivalent to those obtained with the larger and non-optimized files (for more details about the experiment, see the full case study).

The case study is part of Beamr’s ongoing commitment to accelerate adoption and increase accessibility of machine learning for video and video analysis solutions. A previous case study focused on the AI network inference stage, which is the phase of drawing conclusions from an AI network that has already been trained. The previous experiment found that video files that were downsized by 40% on average streamlined machine learning processes. This allowed significant savings in storage and costs.