| 1. |
|
|
| 2. |
|
|
| 3. |
|
|
| 4. |
- Bengtsson Bernander, Karl
(författare)
-
Equivariant Neural Networks for Biomedical Image Analysis
- 2024
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- While artificial intelligence and deep learning have revolutionized many fields in the last decade, one of the key drivers has been access to data. This is especially true in biomedical image analysis where expert annotated data is hard to come by. The combination of Convolutional Neural Networks (CNNs) with data augmentation has proven successful in increasing the amount of training data at the cost of overfitting. In this thesis, equivariant neural networks have been used to extend the equivariant properties of CNNs to more transformations than translations. The networks have been trained and evaluated on biomedical image datasets, including bright-field microscopy images of cytological samples indicating oral cancer, and transmission electron microscopy images of virus samples. By designing the networks to be equivariant to e.g. rotations, it is shown that the need for data augmentation is reduced, that less overfitting occurs, and that convergence during training is faster. Furthermore, equivariant neural networks are more data efficient than CNNs, as demonstrated by scaling laws. These benefits are not present in all problem settings and which benefits will occur is somewhat unpredictable. We have identified that the results to some extent depend on architectures, hyperparameters and datasets. Further research may broaden the performed studies to explain how the results occur with new theory.
|
|
| 5. |
- Bengtsson Bernander, Karl
(författare)
-
Improving Training of Deep Learning for Biomedical Image Analysis and Computational Physics
- 2021
-
Licentiatavhandling (övrigt vetenskapligt/konstnärligt)abstract
- The previous decade has seen breakthroughs in image analysis and computer vision, mainly due to machine learning methods known as deep learning. These methods have since spread to other fields. This thesis aims to survey the progress, highlight problems related to data and computations, and show techniques to mitigate them.In Paper I, we show how to modify the VGG16 classifier architecture to be equivariant to transformations in the p4 group, consisting of translations and specific rotations. We conduct experiments to investigate if baseline architectures, using data augmentation, can be replaced with these rotation-equivariant networks. We train and test on the Oral cancer dataset, used to automate cancer diagnostics.In Paper III, we use a similar methodology as in Paper I to modify the U-net architecture combined with a discriminative loss, for semantic instance segmentation. We test the method on the BBBC038 dataset consisting of highly varied images of cell nuclei.In Paper II, we look at the UCluster method, used to group sub- atomic particles in particle physics. We show how to distribute the training over multiple GPUs using distributed deep learning in a cloud environment.The papers show how to use limited training data more efficiently, using group-equivariant convolutions, to reduce the prob- lems of overfitting. They also demonstrate how to distribute training over multiple nodes in computational centers, which is needed to handle growing data sizes.
|
|
| 6. |
- Bengtsson Bernander, Karl, et al.
(författare)
-
Replacing data augmentation with rotation-equivariant CNNs in image-based classification of oral cancer
- 2021
-
Ingår i: Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications. - Cham : Springer International Publishing. - 9783030934194 - 9783030934200 ; , s. 24-33
-
Konferensbidrag (refereegranskat)abstract
- We present how replacing convolutional neural networks with a rotation-equivariant counterpart can be used to reduce the amount of training images needed for classification of whether a cell is cancerous or not. Our hypothesis is that data augmentation schemes by rotation can be replaced, thereby increasing weight sharing and reducing overfitting. The dataset at hand consists of single cell images. We have balanced a subset of almost 9.000 images from healthy patients and patients diagnosed with cancer. Results show that classification accuracy is improved and overfitting reduced if compared to an ordinary convolutional neural network. The results are encouraging and thereby an advancing step towards making screening of patients widely used for the application of oral cancer.
|
|
| 7. |
|
|
| 8. |
- Sunneborn Gudnadottir, Olga, et al.
(författare)
-
Distributed training and scalability for the particle clustering method UCluster
- 2021
-
Ingår i: EPJ Web of Conferences. - : EDP Sciences. - 2100-014X. ; 251, s. 02054-02054
-
Tidskriftsartikel (refereegranskat)abstract
- In recent years, machine-learning methods have become increasingly important for the experiments at the Large Hadron Collider (LHC). They are utilised in everything from trigger systems to reconstruction and data analysis. The recent UCluster method is a general model providing unsupervised clustering of particle physics data, that can be easily modified to provide solutions for a variety of different decision problems. In the current paper, we improve on the UCluster method by adding the option of training the model in a scalable and distributed fashion, and thereby extending its utility to learn from arbitrarily large data sets. UCluster combines a graph-based neural network called ABCnet with a clustering step, using a combined loss function in the training phase. The original code is publicly available in TensorFlow v1.14 and has previously been trained on a single GPU. It shows a clustering accuracy of 81% when applied to the problem of multi-class classification of simulated jet events. Our implementation adds the distributed training functionality by utilising the Horovod distributed training framework, which necessitated a migration of the code to TensorFlow v2. Together with using parquet files for splitting data up between different compute nodes, the distributed training makes the model scalable to any amount of input data, something that will be essential for use with real LHC data sets. We find that the model is well suited for distributed training, with the training time decreasing in direct relation to the number of GPU’s used. However, further improvements by a more exhaustive and possibly distributed hyper-parameter search is required in order to achieve the reported accuracy of the original UCluster method.
|
|
