Green's operator reveals how multitask networks share computations
Disentangling Computation in Multi-Task Neural Networks with the Green's Operator
Machine Learning
Summary
Scientists wanted to understand how neural networks handle different tasks and reuse knowledge over time. They used a mathematical tool called the Green's operator to trace how small changes in the network affect later behavior. This approach showed how the network naturally reuses certain patterns across tasks and over time as it learns. The Green's operator gives a global picture of the network's computation instead of focusing on just snapshots of activity.
What this means in practice
- •For machine learning engineers: Analyze how multitask models reuse computations internally to optimize training and architecture design.
- •For cognitive modelers: Map causal pathways in learned neural dynamics to better simulate human multitasking cognition.
Authors
James Hazelden
Abstract
How is computation organized and reused across tasks and time in a trained recurrent network? Most analyses emphasize the geometry of neural activity, dynamical motifs, or local perturbation growth. We instead study the network's global first-order perturbation response. The finite-horizon Green's operator maps perturbations at each source along a trajectory to their downstream state-space responses and therefore directly represents perturbation routing. Simple reductions of this operator provide task-to-task and time-to-time views of the same computation, while matrix-free products make these views accessible without constructing the full operator. In a flexible multitask recurrent network, task reductions reveal structured reuse of known computational motifs, while temporal reductions reveal causal pathways and how they emerge during training. Our main point is simple: the Green's operator provides a global response geometry for mapping the organization of learned dynamical computation.