Showing 1–2 of 2 results for author: Bakos, B

Enhancing pretraining efficiency for medical image segmentation via transferability metrics

Authors: Gábor Hidy, Bence Bakos, András Lukács

Abstract: In medical image segmentation tasks, the scarcity of labeled training data poses a significant challenge when training deep neural networks. When using U-Net-style architectures, it is common practice to address this problem by pretraining the encoder part on a large general-purpose dataset like ImageNet. However, these methods are resource-intensive and do not guarantee improved performance on th… ▽ More In medical image segmentation tasks, the scarcity of labeled training data poses a significant challenge when training deep neural networks. When using U-Net-style architectures, it is common practice to address this problem by pretraining the encoder part on a large general-purpose dataset like ImageNet. However, these methods are resource-intensive and do not guarantee improved performance on the downstream task. In this paper we investigate a variety of training setups on medical image segmentation datasets, using ImageNet-pretrained models. By examining over 300 combinations of models, datasets, and training methods, we find that shorter pretraining often leads to better results on the downstream task, providing additional proof to the well-known fact that the accuracy of the model on ImageNet is a poor indicator for downstream performance. As our main contribution, we introduce a novel transferability metric, based on contrastive learning, that measures how robustly a pretrained model is able to represent the target data. In contrast to other transferability scores, our method is applicable to the case of transferring from ImageNet classification to medical image segmentation. We apply our robustness score by measuring it throughout the pretraining phase to indicate when the model weights are optimal for downstream transfer. This reduces pretraining time and improves results on the target task. △ Less

Submitted 6 June, 2025; v1 submitted 24 October, 2024; originally announced October 2024.

Comments: An error was discovered in the aggregation process of our results, particularly affecting the experiments involving the advanced pretraining method. This impacts the main conclusions of the paper, and we are therefore withdrawing the submission

ACM Class: I.4.6

On a Communication Complexity problem in Combinatorial Number Theory

Authors: Bence Bakos, Norbert Hegyvári, Máté Pálfy

Abstract: The original knapsack problem is well known to be NP-complete. In a multidimensional version one have to decide whether a $p\in \N^k$ is in a sumset-sum of a set $X \subseteq \N^k$ or not. In this paper we are going to investigate a communication complexity problem related to this. The original knapsack problem is well known to be NP-complete. In a multidimensional version one have to decide whether a $p\in \N^k$ is in a sumset-sum of a set $X \subseteq \N^k$ or not. In this paper we are going to investigate a communication complexity problem related to this. △ Less

Submitted 15 March, 2021; originally announced March 2021.