Fostering Virtual Collaboration with Synthetic Data-Powered AI

Synthetic data powered AI is driving virtual collaboration systems. In March 2020, companies across the US abruptly closed office doors and sent employees home to work remotely due to the COVID-19 pandemic. At first, many thought the shutdowns would last a couple of months. But a year and a half later, millions of employees are still working remotely. The seemingly more permanent shift to remote work is leaving a mark on where and how we do our jobs — more so than anyone could have anticipated. It has also irrevocably changed how we communicate with colleagues, which has primarily been through video conferencing since the start of the pandemic.

As the world grows more remote, new technologies are sure to improve — and even disrupt — how we communicate. Companies are working on next-generation, AI-enabled features to enhance the telecommunication experience. To improve picture quality and more effectively utilize bandwidth, companies are using photorealistic avatars and neural rendering techniques to recreate images with improved resolution and lighting. Others are looking to develop emotion-sensing technology to characterize attentiveness and meeting effectiveness. Simpler features like virtual background will soon get a significant boost in quality as companies develop better facial segmentation and matting models. However, the most significant limitation in developing these next-generation features is access to high-quality, privacy-compliant data to train AI models.

Synthetic data is poised to accelerate how we build AI models and usher in the next generation of virtual collaboration platforms. Synthetic data is computer-generated image data that models the real world. Technologies from the visual effects industry are coupled with generative neural networks to create vast amounts of photorealistic and automatically labeled image data at a fraction of the cost and time of current approaches. The emerging technology enables the efficient prototyping, building, and testing of complex computer vision systems.

Since the data is computer-generated, underlying consumer privacy concerns are mitigated. As regulatory measures limit the ability of companies to use real-world data, synthetic data will step in and play a crucial part in developing video and teleconferencing systems.

Building advanced features will also require highly accurate 3D labels that traditional human-labeling approaches cannot provide. With synthetic data approaches, an expanded set of pixel-perfect labels is automatically generated, including 3D facial landmarks, depth maps, segmentation masks, and surface normals, enabling more capable AI models.

With more companies having a “remote first” approach to work, having a reliable and engaging teleconference and video experience will be essential. Synthetic data can play a key role in enhancing current tools to help boost collaboration and productivity during virtual meetings. It’s hard to replicate the connection with meeting face-to-face, but synthetic data can help bridge the gap.

[To share your insights with us, please write to sghosh@martechseries.com]