PMF-CL: Pareto-Minimal-Forgetting Continual Learner for Conflicting Tasks

Srijith Nair, Atilla Eryilmaz, Jia (Kevin) Liu · 2026

Read PMF-CL: Pareto-Minimal-Forgetting Continual Learner for Conflicting Tasks by Srijith Nair, Atilla Eryilmaz, Jia (Kevin) Liu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-19145.txt.

Description

Srijith Nair, Atilla Eryilmaz, and Jia Liu reframe continual learning as weighing clashing tasks. Most methods store past-task memory, yet what must be stored to cut loss remains unclear. When all tasks share one point, loss can vanish, but tasks rarely share one, so some loss occurs. Their Pareto-minimal-forgetting method finds plans that forget tasks least, for linear and basis-function regression. It uses memory that does not grow with task count, gives guarantees for quadratic losses and bounded loss if not, and beats prior methods.

Catalog record

CREATORSrijith Nair, Atilla Eryilmaz, Jia (Kevin) Liu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93996
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.19145

Browse the full library · About the Tower of Babel