Geoffrey Hinton

Foundational researcher whose contributions to neural‑network training and representation learning underpinned the deep‑learning revolution. The formalization of backpropagation, insights into hidden‑layer representations and early work on probabilistic neural models helped establish which classes of networks are trainable, generalize well and are amenable to compression or distillation—properties critically relevant for on‑chain deployment. Projects like Cortex, which seek to make ML models a first‑class asset in a blockchain economy, depend on an ecosystem where models can be reliably trained off‑chain, represented compactly on‑chain, and verified for correctness and performance. The methodological lineage from early deep‑learning research informs verification strategies (e.g., black‑box vs white‑box testing), acceptable approximation techniques for cheaper inference, and expectations about failure modes and robustness under distributional shift. Additionally, the conceptual framing of neural networks as probabilistic function approximators influenced how model uncertainty and reward mechanisms are encoded in decentralized marketplaces: reputation, challenge‑response proof systems and staking incentives for model providers all require an understanding of model behavior under adversarial or shifted inputs. While not directly engaged with Cortex’s engineering, the cumulative effect of foundational deep‑learning research made possible the kinds of compact, efficient and verifiable models that Cortex aspires to integrate within a permissionless blockchain environment.