on-device training has become an increasingly popular approach to machine
learning, enabling models to be trained directly on mobile and edge devices.
However, a major challenge in this area is the limited memory available on
these devices, which can severely restrict the size and comp