SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2307.11342 · ICCV · 2023

Tuning Pre-trained Model via Moment Probing

Qilong Wang, Qinghua Hu, Jingbo Zhou, Pengfei Zhu, Mingze Gao, Zhenyi Lin

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 13 functions out of this paper's own repositories and ran 8 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
mingzeg/moment-probing canonical 8 of 11
mingzeG/Moment-Probing — 0 of 2
FunctionStatusWhere it lives
MyNorm Ran mingzeg/moment-probing/models/as_mlp.py
pointer only (licence: NONE) · get_code("c3f4c265067c1717")
get_init_weights_vit Ran mingzeg/moment-probing/models/vision_transformer.py
pointer only (licence: NONE) · get_code("def23f7d7d042968")
get_mce_from_accuracy Ran mingzeg/moment-probing/utils/mce_utils.py
pointer only (licence: NONE) · get_code("a9e48df75d2e823e")
init_ssf_scale_shift Ran mingzeg/moment-probing/models/vision_transformer.py
pointer only (licence: NONE) · get_code("896b5eec245f1bab")
param_groups_weight_decay Ran mingzeg/moment-probing/optim_factory.py
pointer only (licence: NONE) · get_code("88f4962784e09b88")
ssf_ada Ran mingzeg/moment-probing/models/vision_transformer.py
pointer only (licence: NONE) · get_code("61e8364b7e9bcb70")
window_partition Ran mingzeg/moment-probing/models/swin_transformer.py
pointer only (licence: NONE) · get_code("993ae96666b00cb5")
window_reverse Ran mingzeg/moment-probing/models/swin_transformer.py
pointer only (licence: NONE) · get_code("609922bd93c75117")
Covariance Not yet run mingzeG/Moment-Probing/models/representation/MP.py
pointer only (licence: NONE) · get_code("80a1d69c480f2031")
Moment_Probing_ViT Not yet run mingzeG/Moment-Probing/models/representation/MP.py
pointer only (licence: NONE) · get_code("28fa2c5680a21935")
checkpoint_filter_fn Not yet run mingzeg/moment-probing/models/convnext.py
pointer only (licence: NONE) · get_code("7619bf7f57c5d2b6")
group_parameters Not yet run mingzeg/moment-probing/optim_factory.py
pointer only (licence: NONE) · get_code("61a14e104fefeb56")
group_with_matcher Not yet run mingzeg/moment-probing/optim_factory.py
pointer only (licence: NONE) · get_code("1fd037cb67e5a59e")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Recently, efficient fine-tuning of large-scale pre-trained models has attracted increasing research interests, where linear probing (LP) as a fundamental module is involved in exploiting the final representations for task-dependent classification. However, most of the existing methods focus on how to effectively introduce a few of learnable parameters, and little work pays attention to the commonly used LP module. In this paper, we propose a novel Moment Probing (MP) method to further explore the potential of LP. Distinguished from LP which builds a linear classification head based on the mean of final features (e.g., word tokens for ViT) or classification tokens, our MP performs a linear classifier on feature distribution, which provides the stronger representation ability by exploiting richer statistical information inherent in features. Specifically, we represent feature distribution by its characteristic function, which is efficiently approximated by using first-and second-order moments of features. Furthermore, we propose a multihead convolutional cross-covariance (MHC 3 ) to compute second-order moments in an efficient and effective manner. By considering that MP could affect feature learning, we introduce a partially shared module to learn two recalibrating parameters (PSRP) for backbones based on MP, namely MP + . Extensive experiments on ten benchmarks using various models show that our MP significantly outperforms LP and is competitive with counterparts at lower training cost, while our MP + achieves state-of-the-art performance.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2307.11342")
get_code_for_paper("2307.11342")
have("2307.11342")

Connect an agent — have() is free.