Truths Lab
Research at Truths Lab is applied. We read the literature, we run small experiments where the answers are uncertain, and we publish when what we have learned is load-bearing for someone other than us. The areas below organize the work and the notes that follow.
Models that work under the constraints of real environments: limited light, novel viewpoints, restricted compute. The research that underwrites our product work.
Compiling research models to hardware that exists today, and to hardware that exists in places where the cloud is the wrong answer.
Better measurements for AI systems in their deployed setting, not their benchmark setting. Particularly methods that hold up across populations.
We do not publish on a schedule. The list grows when we have something worth recording. Sign up for occasional emails by writing to support@truthslab.com.
A short working note on how we measure detection systems in the environments they will run in, not the ones the dataset implies.