Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userChannel Scaling: A Scale-and-Select Approach for Transfer Learning
112
0
0.0
(
0
)
Added by
Ken C. L. Wong
Publication date
2021
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Transfer learning with pre-trained neural networks is a common strategy for training classifiers in medical image analysis. Without proper channel selections, this often results in unnecessarily large models that hinder deployment and explainability. In this paper, we propose a novel approach to efficiently build small and well performing networks by introducing the channel-scaling layers. A channel-scaling layer is attached to each frozen convolutional layer, with the trainable scaling weights inferring the importance of the corresponding feature channels. Unlike the fine-tuning approaches, we maintain the weights of the original channels and large datasets are not required. By imposing L1 regularization and thresholding on the scaling weights, this framework iteratively removes unnecessary feature channels from a pre-trained model. Using an ImageNet pre-trained VGG16 model, we demonstrate the capabilities of the proposed framework on classifying opacity from chest X-ray images. The results show that we can reduce the number of parameters by 95% while delivering a superior performance.
rate research
Read More
Transfer learning allows the reuse of deep learning features on new datasets with limited data. However, the resulting models could be unnecessarily large and thus inefficient. Although network pruning can be applied to improve inference efficiency, existing algorithms usually require fine-tuning and may not be suitable for small datasets. In this paper, we propose an algorithm that transforms the convolutional weights into the subspaces of orthonormal bases where a model is pruned. Using singular value decomposition, we decompose a convolutional layer into two layers: a convolutional layer with the orthonormal basis vectors as the filters, and a layer that we name BasisScalingConv, which is responsible for rescaling the features and transforming them back to the original space. As the filters in each transformed layer are linearly independent with known relative importance, pruning can be more effective and stable, and fine tuning individual weights is unnecessary. Furthermore, as the numbers of input and output channels of the original convolutional layer remain unchanged, basis pruning is applicable to virtually all network architectures. Basis pruning can also be combined with existing pruning algorithms for double pruning to further increase the pruning capability. With less than 1% reduction in the classification accuracy, we can achieve pruning ratios up to 98.9% in parameters and 98.6% in FLOPs.
Skip connections in deep networks have improved both segmentation and classification performance by facilitating the training of deeper network architectures, and reducing the risks for vanishing gradients. They equip encoder-decoder-like networks with richer feature representations, but at the cost of higher memory usage, computation, and possibly resulting in transferring non-discriminative feature maps. In this paper, we focus on improving skip connections used in segmentation networks (e.g., U-Net, V-Net, and The One Hundred Layers Tiramisu (DensNet) architectures). We propose light, learnable skip connections which learn to first select the most discriminative channels and then attend to the most discriminative regions of the selected feature maps. The output of the proposed skip connections is a unique feature map which not only reduces the memory usage and network parameters to a high extent, but also improves segmentation accuracy. We evaluate the proposed method on three different 2D and volumetric datasets and demonstrate that the proposed light, learnable skip connections can outperform the traditional heavy skip connections in terms of segmentation accuracy, memory usage, and number of network parameters.
Domain similarity measures can be used to gauge adaptability and select suitable data for transfer learning, but existing approaches define ad hoc measures that are deemed suitable for respective tasks. Inspired by work on curriculum learning, we propose to emph{learn} data selection measures using Bayesian Optimization and evaluate them across models, domains and tasks. Our learned measures outperform existing domain similarity measures significantly on three tasks: sentiment analysis, part-of-speech tagging, and parsing. We show the importance of complementing similarity with diversity, and that learned measures are -- to some degree -- transferable across models, domains, and even tasks.
Transfer learning is a standard technique to improve performance on tasks with limited data. However, for medical imaging, the value of transfer learning is less clear. This is likely due to the large domain mismatch between the usual natural-image pre-training (e.g. ImageNet) and medical images. However, recent advances in transfer learning have shown substantial improvements from scale. We investigate whether modern methods can change the fortune of transfer learning for medical imaging. For this, we study the class of large-scale pre-trained networks presented by Kolesnikov et al. on three diverse imaging tasks: chest radiography, mammography, and dermatology. We study both transfer performance and critical properties for the deployment in the medical domain, including: out-of-distribution generalization, data-efficiency, sub-group fairness, and uncertainty estimation. Interestingly, we find that for some of these properties transfer from natural to medical images is indeed extremely effective, but only when performed at sufficient scale.
Offline Signature Verification (OSV) is a challenging pattern recognition task, especially when it is expected to generalize well on the skilled forgeries that are not available during the training. Its challenges also include small training sample and large intra-class variations. Considering the limitations, we suggest a novel transfer learning approach from Persian handwriting domain to multi-language OSV domain. We train two Residual CNNs on the source domain separately based on two different tasks of word classification and writer identification. Since identifying a person signature resembles identifying ones handwriting, it seems perfectly convenient to use handwriting for the feature learning phase. The learned representation on the more varied and plentiful handwriting dataset can compensate for the lack of training data in the original task, i.e. OSV, without sacrificing the generalizability. Our proposed OSV system includes two steps: learning representation and verification of the input signature. For the first step, the signature images are fed into the trained Residual CNNs. The output representations are then used to train SVMs for the verification. We test our OSV system on three different signature datasets, including MCYT (a Spanish signature dataset), UTSig (a Persian one) and GPDS-Synthetic (an artificial dataset). On UT-SIG, we achieved 9.80% Equal Error Rate (EER) which showed substantial improvement over the best EER in the literature, 17.45%. Our proposed method surpassed state-of-the-arts by 6% on GPDS-Synthetic, achieving 6.81%. On MCYT, EER of 3.98% was obtained which is comparable to the best previously reported results.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
