Scientific foundation models distill information from freely available unlabeled data into rich priors that can be used in downstream scientific discovery tasks. These models are trained to recapitulate known scientific information, much as a human scientist starts by learning basic facts and doing basic experiments. These trained representations are then leveraged to accelerate progress on real world challenges where much data may not be available.

Differentiable physics applies the techniques of differentiable programming to model physical, chemical, biological, and engineering systems. Differentiable physics methods span high dimensional partial and ordinary differential equation solution, deep representation learning, rich priors from scientific foundation models, and more. Deep Forest Sciences is a leader in the emerging field of differentiable physics.