Victor

I am software engineer in machine learning, currently working on YouTube recommender systems. My current work is on multi-task modelling and reinforcement learning. Prior to joining Google, my PhD from Caltech is in pure mathematics, on the goemetry and algebra.

My hobbies include contact improv dancing and taking care of my small dog in San Francisco.
Authored Publications
Sort By
  • Title
  • Title, descending
  • Year
  • Year, descending
Learned Task Relationships in Multi-task Models
Naijing Zhang
Bosun Adeoti
Mike Leung
Florian Raudies
Yiping Yuan
Recsys (2026)
Preview abstract We propose a framework that learns cross-task relationships in multi-task models by estimating the joint distribution of task labels. This approach improves performance via transfer learning and enhances information extraction. Although our framework applies to any multi-task system, we demonstrate its efficacy within YouTube’s production recommendation stacks. Experiments across the Notifications, Homepage, and Watch Next surfaces show improvements in both accuracy and satisfied user engagement. Finally, we propose a workflow template to facilitate broader future implementation. View details
Preview abstract We study various aspects of the representation theory of loop groups, all with the aim of giving geometric constructions, parameterized by conjugacy classes of the Weyl group, of the basic representation of the affine Lie algebras associated to a simply laced simple Lie algebra as a restriction isomorphism on dual sections of the level 1 line bundle on the affine Grassmannian. Along the way, we obtain various results on the structure of loop tori, the definition of a notion of a Heisenberg Central extension as an alternative for twisted modules over the lattice vertex algebra and the determination of their representation theory, some computations on central extensions of a torus over a field by K 2 , and a new proof of the classification of the conjugacy classes of the Weyl group by parabolic induction. View details
×