The Normal Distribution, Vol. 5

Synthetic Data for Competitive Advantage

The historic limiting factor in the development of perception systems has been?compute: Building and training sophisticated (and even unsophisticated) CV models requires a significant amount of computational power. As costs have dropped in cloud, on-prem and hybrid deployments for both CPU-and GPU-based usage, organizations have begun shifting attention and resources to CV's second biggest historic limitation: Access to high-quality training data.

Agile businesses are now sourcing proprietary ML training data as a form of competitive advantage. Better training data produces more performant, more accurate models. By generating proprietary data, organizations maintain control over how it's used, with the added benefit of synthetics being ethically sourced and auditable for?content provenance.?Synthetic data from Synthesis AI gives ML practitioners an edge when developing CV models.

Modern Generative AI, Part III

Our?series?on generative AI looks at the underlying technologies that make many machine learning models function. Building on earlier discussions of?variational autoencoders?(VAEs) and?discrete latent spaces,?Part III in the series?shifts focus to applied AI in real-world models, in this case, OpenAI's DALL-E.?Read the latest from our Research team?on generative AI and how text-to-image models actually work.

Synthesis Labs: Text-to-3D Beta Program

Earlier in April we were?thrilled to launch?Synthesis Labs, the first release from which is high-resolution text-to-3D generation. We were even more thrilled by the enthusiastic response we received as measured by applications to the?beta program. Thanks to everyone who's signed up to date; the application is?still open. We'll be sending a short survey to applicants to get a better sense of how you wish to use 3D assets.

OpenSynthetics: Community for ML

If you're a student or academic, or just looking to network with other machine learning practitioners, OpenSynthetics is a valuable resource.?Open datasets?to support non-commercial ML development are?available.