Browsing by Subject "transfer learning"

Now showing 1 - 6 of 6
  • Thumbnail Image
    Item
    Cycle-consistent Conditional Adversarial Transfer Networks
    (ACM, 2019-10) Li, Jingjing; Chen, Erpeng; Ding, Zhengming; Zhu, Lei; Lu, Ke; Huang, Zi; Computer Information and Graphics Technology, School of Engineering and Technology
    Domain adaptation investigates the problem of cross-domain knowledge transfer where the labeled source domain and unlabeled target domain have distinctive data distributions. Recently, adversarial training have been successfully applied to domain adaptation and achieved state-of-the-art performance. However, there is still a fatal weakness existing in current adversarial models which is raised from the equilibrium challenge of adversarial training. Specifically, although most of existing methods are able to confuse the domain discriminator, they cannot guarantee that the source domain and target domain are sufficiently similar. In this paper, we propose a novel approach named cycle-consistent conditional adversarial transfer networks (3CATN) to handle this issue. Our approach takes care of the domain alignment by leveraging adversarial training. Specifically, we condition the adversarial networks with the cross-covariance of learned features and classifier predictions to capture the multimodal structures of data distributions. However, since the classifier predictions are not certainty information, a strong condition with the predictions is risky when the predictions are not accurate. We, therefore, further propose that the truly domain-invariant features should be able to be translated from one domain to the other. To this end, we introduce two feature translation losses and one cycle-consistent loss into the conditional adversarial domain adaptation networks. Extensive experiments on both classical and large-scale datasets verify that our model is able to outperform previous state-of-the-arts with significant improvements.
  • Thumbnail Image
    Item
    Full Training versus Fine Tuning for Radiology Images Concept Detection Task for the ImageCLEF 2019 Challenge
    (2019) Sinha, Priyanshu; Purkayastha, Saptarshi; Gichoya, Judy; BioHealth Informatics, School of Informatics and Computing
    Concept detection from medical images remains a challenging task that limits implementation of clinical ML/AI pipelines because of the scarcity of the highly trained experts to annotate images. There is a need for automated processes that can extract concrete textual information from image data. ImageCLEF 2019 provided us a set of images with labels as UMLS concepts. We participated for the rst time for the concept detection task using transfer learning. Our approach involved an experiment of layerwise ne tuning (full training) versus ne tuning based on previous reported recommendations for training classi cation, detection and segmentation tasks for medical imaging. We ranked number 9 in this year's challenge, with an F1 result of 0.05 after three entries. We had a poor result from performing layerwise tuning (F1 score of 0.014) which is consistent with previous authors who have described the bene t of full training for transfer learning. However when looking at the results by a radiologist, the terms do not make clinical sense and we hypothesize that we can achieve better performance when using medical pretrained image models for example PathNet and utilizing a hierarchical training approach which is the basis of our future work on this dataset.
  • Thumbnail Image
    Item
    Knowledge Reused Outlier Detection
    (IEEE, 2019-03) Yu, Weiren; Ding, Zhengming; Hu, Chunming; Liu, Hongfu; Computer and Information Science, School of Science
    Tremendous efforts have been invested in the unsupervised outlier detection research, which is conducted on unlabeled data set with abnormality assumptions. With abundant related labeled data available as auxiliary information, we consider transferring the knowledge from the labeled source data to facilitate the unsupervised outlier detection on target data set. To fully make use of the source knowledge, the source data and target data are put together for joint clustering and outlier detection using the source data cluster structure as a constraint. To achieve this, the categorical utility function is employed to regularize the partitions of target data to be consistent with source data labels. With an augmented matrix, the problem is completely solved by a K-means - a based method with the rigid mathematical formulation and theoretical convergence guarantee. We have used four real-world data sets and eight outlier detection methods of different kinds for extensive experiments and comparison. The results demonstrate the effectiveness and significant improvements of the proposed methods in terms of outlier detection and cluster validity metrics. Moreover, the parameter analysis is provided as a practical guide, and noisy source label analysis proves that the proposed method can handle real applications where source labels can be noisy.
  • Thumbnail Image
    Item
    Medication Adherence Prediction Through Online Social Forums: A Case Study of Fibromyalgia
    (JMIR, 2019) Haas, Kyle; Ben Miled, Zina; Mahoui, Malika; Electrical and Computer Engineering, School of Engineering and Technology
    Background: Medication nonadherence can compound into severe medical problems for patients. Identifying patients who are likely to become nonadherent may help reduce these problems. Data-driven machine learning models can predict medication adherence by using selected indicators from patients’ past health records. Sources of data for these models traditionally fall under two main categories: (1) proprietary data from insurance claims, pharmacy prescriptions, or electronic medical records and (2) survey data collected from representative groups of patients. Models developed using these data sources often are limited because they are proprietary, subject to high cost, have limited scalability, or lack timely accessibility. These limitations suggest that social health forums might be an alternate source of data for adherence prediction. Indeed, these data are accessible, affordable, timely, and available at scale. However, they can be inaccurate. Objective: This paper proposes a medication adherence machine learning model for fibromyalgia therapies that can mitigate the inaccuracy of social health forum data. Methods: Transfer learning is a machine learning technique that allows knowledge acquired from one dataset to be transferred to another dataset. In this study, predictive adherence models for the target disease were first developed by using accurate but limited survey data. These models were then used to predict medication adherence from health social forum data. Random forest, an ensemble machine learning technique, was used to develop the predictive models. This transfer learning methodology is demonstrated in this study by examining data from the Medical Expenditure Panel Survey and the PatientsLikeMe social health forum. Results: When the models are carefully designed, less than a 5% difference in accuracy is observed between the Medical Expenditure Panel Survey and the PatientsLikeMe medication adherence predictions for fibromyalgia treatments. This design must take into consideration the mapping between the predictors and the outcomes in the two datasets. Conclusions: This study exemplifies the potential and limitations of transfer learning in medication adherence–predictive models based on survey data and social health forum data. The proposed approach can make timely medication adherence monitoring cost-effective and widely accessible. Additional investigation is needed to improve the robustness of the approach and extend its applicability to other therapies and other sources of data. [JMIR Med Inform 2019;7(2):e12561]
  • Thumbnail Image
    Item
    Structure-Preserved Unsupervised Domain Adaptation
    (IEEE, 2019-04) Liu, Hongfu; Shao, Ming; Ding, Zhengming; Fu, Yun; Computer Information and Graphics Technology, School of Engineering and Technology
    Domain adaptation has been a primal approach to addressing the issues by lack of labels in many data mining tasks. Although considerable efforts have been devoted to domain adaptation with promising results, most existing work learns a classifier on a source domain and then predicts the labels for target data, where only the instances near the boundary determine the hyperplane and the whole structure information is ignored. Moreover, little work has been done regarding to multi-source domain adaptation. To that end, we develop a novel unsupervised domain adaptation framework, which ensures the whole structure of source domains is preserved to guide the target structure learning in a semi-supervised clustering fashion. To our knowledge, this is the first time when the domain adaptation problem is re-formulated as a semi-supervised clustering problem with target labels as missing values. Furthermore, by introducing an augmented matrix, a non-trivial solution is designed, which can be exactly mapped into a K-means-like optimization problem with modified distance function and update rule for centroids in an efficient way. Extensive experiments on several widely-used databases show the substantial improvements of our proposed approach over the state-of-the-art methods.
  • Thumbnail Image
    Item
    A Transfer Learning Approach to Object Detection Acceleration for Embedded Applications
    (2021-08) Vance, Lauren M.; Christopher, Lauren; King, Brian; Rizkalla, Maher
    Deep learning solutions to computer vision tasks have revolutionized many industries in recent years, but embedded systems have too many restrictions to take advantage of current state-of-the-art configurations. Typical embedded processor hardware configurations must meet very low power and memory constraints to maintain small and lightweight packaging, and the architectures of the current best deep learning models are too computationally-intensive for these hardware configurations. Current research shows that convolutional neural networks (CNNs) can be deployed with a few architectural modifications on Field-Programmable Gate Arrays (FPGAs) resulting in minimal loss of accuracy, similar or decreased processing speeds, and lower power consumption when compared to general-purpose Central Processing Units (CPUs) and Graphics Processing Units (GPUs). This research contributes further to these findings with the FPGA implementation of a YOLOv4 object detection model that was developed with the use of transfer learning. The transfer-learned model uses the weights of a model pre-trained on the MS-COCO dataset as a starting point then fine-tunes only the output layers for detection on more specific objects of five classes. The model architecture was then modified slightly for compatibility with the FPGA hardware using techniques such as weight quantization and replacing unsupported activation layer types. The model was deployed on three different hardware setups (CPU, GPU, FPGA) for inference on a test set of 100 images. It was found that the FPGA was able to achieve real-time inference speeds of 33.77 frames-per-second, a speedup of 7.74 frames-per-second when compared to GPU deployment. The model also consumed 96% less power than a GPU configuration with only approximately 4% average loss in accuracy across all 5 classes. The results are even more striking when compared to CPU deployment, with 131.7-times speedup in inference throughput. CPUs have long since been outperformed by GPUs for deep learning applications but are used in most embedded systems. These results further illustrate the advantages of FPGAs for deep learning inference on embedded systems even when transfer learning is used for an efficient end-to-end deployment process. This work advances current state-of-the-art with the implementation of a YOLOv4 object detection model developed with transfer learning for FPGA deployment.