Zihan Ren (任子汉)

Non-observable, finite, resource-heavy assets — the subsurface today, and eventually off-Earth exploration — will only become more valuable as AI takes on part of the knowledge work and pushes everything into a much faster iteration cycle. I think of earth-asset utilization (say, mineral exploration or petroleum) as one coherent picture rather than isolated efforts. Part of that idea drives my own work: building generative priors of the subsurface for downstream tasks, wrapping predictive models as surrogates for optimization, and turning noisy measurements into decision-ready uncertainty (see portfolio).

I am open to collaboration on geo-foundation models, representation learning, and generative modeling — learning representations that are explainable and configurable for both generation and downstream tasks such as task identification, prediction, and inversion, with the decision loop embedded in the process rather than bolted on afterward. If that sounds interesting, reach out via email or LinkedIn.