Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Loss Smoothing for Stable Adaptation Under Distribution Shift

About

In settings such as fine-tuning and reinforcement learning, neural networks are often adapted under distribution shift. Standard adaptation methods typically optimize the target objective directly, inducing an abrupt change from the source training objective. This abrupt transition can distort learned representations, including features that may still be useful for the new task. We investigate whether a more gradual transition can improve adaptation. We propose loss smoothing, a simple approach that interpolates between the source and target training objectives at the start of adaptation. This smooth transition helps to preserve useful features from the source distribution while still enabling the model to specialize to the target distribution. Across controlled supervised shifts, pretrained vision adaptation, offline-to-online and online reinforcement learning, and language model fine-tuning, we find that loss smoothing consistently improves performance, suggesting that smoother objective transitions are a broadly useful tool for model adaptation.

Darshan Patil, Ekaterina Lobacheva, Razvan Pascanu, Sarath Chandar• 2026

Related benchmarks

TaskDatasetResultRank
Offline-to-Online Reinforcement LearningD4RL Aggregate
Average Normalized Score77.6
17
Image ClassificationDomainNet ID single-source ViT
IQM Accuracy76.42
13
Image ClassificationImageNet ViT (retention)
IQM Accuracy83.86
13
Offline-to-Online Reinforcement LearningD4RL AntMaze
UMaze Success Rate80
7
Offline-to-Online Reinforcement LearningD4RL adroit
Pen Score118.2
7
Offline-to-Online Reinforcement LearningD4RL MuJoCo
HalfCheetah (M) Score48.2
7
Showing 6 of 6 rows

Other info

Follow for update