Multi-taskLearning and Ensemble Approach to Predict Cognitive Scores for Patients with Alzheimer's Disease

The objective of this research was to explore the efficacy of integrating 1) using multi-task learning, 2) neuroimaging data and common risk factors to predict the ADAS-cog 11 score in Alzheimer's patients. We studied the magnetic resonance imaging (MRI) scans of 798 participants ranging between 0 and 96 months from the initial diagnosis. This enabled us to exploit the benefits of the U-Net architecture, a structure historically known to perform well on medical imaging tasks using limited data with heavy image augmentations. The multi-task model simultaneously performed segmentation of white matter, gray matter, and cerebrospinal fluid on the MRI input and regression to predict the ADAS-cog 11 score. There is a body of literature highlighting the independent relationships between gray and white matter volumes and dementia severity; the trajectory was to explore if the multi-task model structure for these interrelated tasks would boost performance for each individual task. The final model integrates the deep learning results with a Gradient Boosting model trained on demographic data. A considerable performance improvement from the initial multi-task U-Net on the ADAS-cog 11 prediction task was achieved across our experiments.

Research on Real-Time System and Related Graph Task Model

Generally speaking, real-time system is considered to be able to influence the environment by receiving and processing data, and returning calculation results rapid enough, so as to control the environment. In computer science, real-time system describes the software and hardware system affected by time constraints, and its correctness relies on the logical correctness of the function and the time when the result is generated. According to the main characteristics of real-time operating system, such as time constraint, predictability and reliability, it puts forward higher requirements for the time accuracy and reliability of real-time operating system. This paper first introduces the real-time system from its main characteristics, related concepts and scheduling algorithm. Then five classical graph based task models of real-time system are introduced. Finally, this paper introduces the directed graph realtime task model from two aspects of definition and semantics. As an extension of realtime system task model, directed graph real-time task model is considered to be able to provide real-time systems with stronger expressive power and support the formal study of time constraint problems.

Biologically relevant transfer learning improves transcription factor binding prediction

Background Deep learning has proven to be a powerful technique for transcription factor (TF) binding prediction but requires large training datasets. Transfer learning can reduce the amount of data required for deep learning, while improving overall model performance, compared to training a separate model for each new task. Results We assess a transfer learning strategy for TF binding prediction consisting of a pre-training step, wherein we train a multi-task model with multiple TFs, and a fine-tuning step, wherein we initialize single-task models for individual TFs with the weights learned by the multi-task model, after which the single-task models are trained at a lower learning rate. We corroborate that transfer learning improves model performance, especially if in the pre-training step the multi-task model is trained with biologically relevant TFs. We show the effectiveness of transfer learning for TFs with ~ 500 ChIP-seq peak regions. Using model interpretation techniques, we demonstrate that the features learned in the pre-training step are refined in the fine-tuning step to resemble the binding motif of the target TF (i.e., the recipient of transfer learning in the fine-tuning step). Moreover, pre-training with biologically relevant TFs allows single-task models in the fine-tuning step to learn useful features other than the motif of the target TF. Conclusions Our results confirm that transfer learning is a powerful technique for TF binding prediction.

Improving Heterogeneous Network Knowledge Transfer Based on the Principle of Generative Adversarial

Deep learning requires a large amount of datasets to train deep neural network models for specific tasks, and thus training of a new model is a very costly task. Research on transfer networks used to reduce training costs will be the next turning point in deep learning research. The use of source task models to help reduce the training costs of the target task models, especially heterogeneous systems, is a problem we are studying. In order to quickly obtain an excellent target task model driven by the source task model, we propose a novel transfer learning approach. The model linearly transforms the feature mapping of the target domain and increases the weight value for feature matching to realize the knowledge transfer between heterogeneous networks and add a domain discriminator based on the principle of generative adversarial to speed up feature mapping and learning. Most importantly, this paper proposes a new objective function optimization scheme to complete the model training. It successfully combines the generative adversarial network with the weight feature matching method to ensure that the target model learns the most beneficial features from the source domain for its task. Compared with the previous transfer algorithm, our training results are excellent under the same benchmark for image recognition tasks.

A FPGA-Based Cloud-Edge Collaboration Platform in Cloud Manufacturing

To address the issue that many devices are connected to the cloud during the manufacturing process, which causes severe delays in analyzing massive manufacturing data in the cloud, an FPGA-based architecture of cloud edge collaboration is proposed. In this architecture, manufacturing equipment is connected to the cloud through an FPGA-based embedded edge node. The device data obtained by the edge node is processed by the FPGA module and the embedded system module according to the time-sensitivity. Considering the limited computing power of a single edge node, to realize cloud-edge collaborative computing, a communication-oriented task model and a computing model for edge nodes are designed. The task model learns cloud to edge and edge-edge communication, and the task model realizes the function of migrating computing tasks to other nodes. The edge node system’s design is realized based on the communication-oriented task model and the computing model for edge nodes. The cloud edge collaboration method is researched and explored based on this system. A series of comparative experiments, comparing the time delay of the FPGA module and embedded system module processing the same data, the framework’s usability and data processing ability can be verified.