Backbone Fine Tuning

Backbone fine-tuning focuses on improving the performance of pre-trained neural network backbones for downstream tasks by selectively adjusting their parameters. Current research emphasizes efficient methods, exploring architectures like convolutional networks (ResNets, ConvNeXts), transformers (ViTs), and recurrent networks (GRUs), and employing techniques such as low-rank adaptation, gradient clipping, and singular value decomposition to optimize speed, accuracy, and generalization across diverse datasets and hardware platforms. This area is crucial for advancing various applications, including image generation, object detection, human pose estimation, and brain-computer interfaces, by enabling more efficient and robust models for resource-constrained environments and improved generalization to unseen data.

Papers

January 21, 2023

Open-World Multi-Task Control Through Goal-Aware Representation Learning and Adaptive Horizon Prediction
Shaofei Cai, Zihao Wang, Xiaojian Ma, Anji Liu, Yitao Liang
Open World Goal Conditioned Policy Goal Conditioned Backbone Fine Tuning Multi Task Agent

September 30, 2022

PointPillars Backbone Type Selection For Fast and Accurate LiDAR Object Detection
Konrad Lis, Tomasz Kryjak
3D Object Detection LiDAR Point Cloud LiDAR Data 3D Object Detector Backbone Fine Tuning LiDAR Object Detection

June 15, 2022

Lessons learned from the NeurIPS 2021 MetaDL challenge: Backbone fine-tuning without episodic meta-learning dominates for few-shot learning image classification
Adrian El Baz, Ihsan Ullah, Edesio Alcobaça, André C. P. L. F. Carvalho, Hong Chen, Fabio Ferreira, Henry Gouk, Chaoyu Guan, Isabelle Guyon, Timothy Hospedales, Shell Hu, Mike Huisman, Frank Hutter, Zhengying Liu, Felix Mohr, Ekrem Öztürk, Jan N. van Rijn, Haozhe Sun, Xin Wang, Wenwu Zhu
Deep Neural Network Meta Learning Shot Image Classification Backbone Fine Tuning Effective Transfer Learning Metalearning Algorithm

June 13, 2022

Singular Value Fine-tuning: Few-shot Segmentation requires Few-parameters Fine-tuning
Yanpeng Sun, Qiang Chen, Xiangyu He, Jian Wang, Haocheng Feng, Junyu Han, Errui Ding, Jian Cheng, Zechao Li, Jingdong Wang
Parameter Efficient Fine Tuning Many Parameter Shot Segmentation Pre Trained Backbone Backbone Fine Tuning

April 7, 2022

Solving ImageNet: a Unified Scheme for Training any Backbone to Top Results
Tal Ridnik, Hussam Lawen, Emanuel Ben-Baruch, Asaf Noy
Supervised ImageNet Unified Framework CNN Model Computer Vision Model Key Result Backbone Fine Tuning Numerous Cutting Edge Backbone ImageNet Challenge

January 3, 2022

Stochastic Weight Averaging Revisited
Hao Guo, Jiyong Jin, Bin Liu
Gradient Descent Stochastic Gradient Descent Neural Network Weight Backbone Fine Tuning Stochastic Weight Averaging

Backbone Fine Tuning

Papers

Open-World Multi-Task Control Through Goal-Aware Representation Learning and Adaptive Horizon Prediction

PointPillars Backbone Type Selection For Fast and Accurate LiDAR Object Detection

Lessons learned from the NeurIPS 2021 MetaDL challenge: Backbone fine-tuning without episodic meta-learning dominates for few-shot learning image classification

Singular Value Fine-tuning: Few-shot Segmentation requires Few-parameters Fine-tuning

Solving ImageNet: a Unified Scheme for Training any Backbone to Top Results

Stochastic Weight Averaging Revisited