Samsung Research

Contribution to AI

Distributing a laboratory across seven cities within a single year was the first thing worth noticing. Between May and October 2018, AI centres opened in Cambridge, Moscow, Toronto, New York and Montreal alongside the existing Seoul and Silicon Valley operations, each attached to a local academic cluster and headed by someone recruited from it — Andrew Blake, formerly director of Microsoft's Cambridge laboratory, chaired the British centre; Victor Lempitsky led the Russian one. The premise was that the talent would not move, so the laboratory should. Moscow produced the most-cited result: a 2019 method for animating a head from a handful of photographs, or a single one, by meta-learning across a large corpus of talking-head video rather than training a model per person. The Mona Lisa clips that circulated afterwards put synthetic video in front of a general audience and sharpened the argument about fabricated footage considerably earlier than it would otherwise have arrived. The more durable contribution concerns where inference happens. Insisting that generative models run on the handset rather than in a data centre — announced in 2023, shipping as live call translation on the Galaxy S24 in January 2024 — turned quantisation, distillation and small-model design into a manufacturing constraint at a scale of hundreds of millions of units. The institutional lesson is bleaker: Moscow, Montreal and New York have since closed.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).