AutoTune: Automatically Tuning Convolutional Neural Networks for Improved Transfer Learning

Basha, S. H. Shabbeer; Vinakota, Sravan Kumar; Pulabaigari, Viswanath; Mukherjee, Snehasis; Dubey, Shiv Ram

doi:10.1016/j.neunet.2020.10.009

Cited by 31 publications

(16 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…( 106 ) proposed a pre-training-based model for problems associated with insufficient data, which effectively reduces the overfitting risks. Several researchers worked with only the first few layers of a pre-trained CNN and then retrained the later ones in a new optimization task ( 107 ). Thus, transfer learning can first train a rough approximation model for a given task and serve as a basis for modeling a novel task.…”

Section: Deep Learning-based Radiomicsmentioning

confidence: 99%

Deep Learning With Radiomics for Disease Diagnosis and Treatment: Challenges and Potential

Zhang

et al. 2022

Front. Oncol.

View full text Add to dashboard Cite

The high-throughput extraction of quantitative imaging features from medical images for the purpose of radiomic analysis, i.e., radiomics in a broad sense, is a rapidly developing and emerging research field that has been attracting increasing interest, particularly in multimodality and multi-omics studies. In this context, the quantitative analysis of multidimensional data plays an essential role in assessing the spatio-temporal characteristics of different tissues and organs and their microenvironment. Herein, recent developments in this method, including manually defined features, data acquisition and preprocessing, lesion segmentation, feature extraction, feature selection and dimension reduction, statistical analysis, and model construction, are reviewed. In addition, deep learning-based techniques for automatic segmentation and radiomic analysis are being analyzed to address limitations such as rigorous workflow, manual/semi-automatic lesion annotation, and inadequate feature criteria, and multicenter validation. Furthermore, a summary of the current state-of-the-art applications of this technology in disease diagnosis, treatment response, and prognosis prediction from the perspective of radiology images, multimodality images, histopathology images, and three-dimensional dose distribution data, particularly in oncology, is presented. The potential and value of radiomics in diagnostic and therapeutic strategies are also further analyzed, and for the first time, the advances and challenges associated with dosiomics in radiotherapy are summarized, highlighting the latest progress in radiomics. Finally, a robust framework for radiomic analysis is presented and challenges and recommendations for future development are discussed, including but not limited to the factors that affect model stability (medical big data and multitype data and expert knowledge in medical), limitations of data-driven processes (reproducibility and interpretability of studies, different treatment alternatives for various institutions, and prospective researches and clinical trials), and thoughts on future directions (the capability to achieve clinical applications and open platform for radiomics analysis).

show abstract

Section: Deep Learning-based Radiomicsmentioning

confidence: 99%

Deep Learning With Radiomics for Disease Diagnosis and Treatment: Challenges and Potential

Zhang

et al. 2022

Front. Oncol.

View full text Add to dashboard Cite

show abstract

“…The hyperparameters that were optimized were the number of neurons in the FC layers (all neurons are connected to all the neurons in the next layer), weight decay, learning rate, batch size (number of examples used to estimate the gradient), and the number of epochs (number of cycles through the full training data). The number of nodes is an integer between 4 and 128 [32], the weight decay and the learning rate are sampled from a log-uniform distribution with bounds from 10 −10 to 10 −3 and from 10 −5 to 10 −1 [32], respectively, and the batch size is an integer in {4, 8, 16, 32, 64, 128} [35]. With these parameter ranges, the loss-function value stopped improving soon after 10 epochs.…”

Section: Bce(ymentioning

confidence: 99%

“…For this reason, we tried 10, 20, and 30 epochs during the optimization. We used 100 optimization rounds [32] including 20 fully-random start-up rounds [35] to determine the optimal parameter values. The best average accuracy between the two outputs was 0.972 on the validation set.…”

Section: Bce(ymentioning

confidence: 99%

Assessment of the Clusterability of Data Using a Multimodal Convolutional Neural Network

Reunanen¹,

Räty²,

Lintonen³

et al. 2022

IEEE Trans. Artif. Intell.

View full text Add to dashboard Cite

“…An Auto-tuning PID controller architecture with neural networks is shown schematically in Figure 5 . This architecture uses a feedback control unit and a Recurrent Network with Supervised Learning and Delayed Structure to adjust the parameters of the PID controller (how and why the combination was selected can be seen in [ 33 ]). When system parameters vary, the neural network can timely correct the PID controller parameters and properly choose parameters for the input layer based on the problem and system architecture.…”

Section: Introductionmentioning

confidence: 99%

Self-Tuning Control Using an Online-Trained Neural Network to Position a Linear Actuator

et al. 2022

View full text Add to dashboard Cite

Linear actuators are widely used in all kinds of industrial applications due to being devices that convert the rotation motion of motors into linear or straight traction/thrust motion. These actuators are ideal for all types of applications where inclination, lifting, traction, or thrust is required under heavy loads, such as wheelchairs, medical beds, and lifting tables. Due to the remarkable ability to exert forces and good precision, they are used classic control systems and controls of high-order. Still, they present difficulties in changing their dynamics and are designed for a range of disturbances. Therefore, in this paper, we present the study of an electric linear actuator. We analyze the positioning in real-time and attack the sudden changes of loads and limitation range by the control. It uses a general-purpose control with self-tuning gains, which can deal with the essential uncertainties of the actuator and suppress disturbances, as they can change their weights to interact with changing systems. The neural network combined with PID control compensates the simplicity of this type of control with artificial intelligence, making it robust to drastic changes in its parameters. Unlike other similar works, this research proposes an online training network with an advantage over typical neural self-adjustment systems. All of this can also be dispensed with the engine model for its operation. The results obtained show a decrease of 42% in the root mean square error (RMSE) during trajectory tracking and saving in energy consumption by 25%. The results were obtained both in simulation and in real tests.

show abstract

AutoTune: Automatically Tuning Convolutional Neural Networks for Improved Transfer Learning

Cited by 31 publications

References 30 publications

Deep Learning With Radiomics for Disease Diagnosis and Treatment: Challenges and Potential

Deep Learning With Radiomics for Disease Diagnosis and Treatment: Challenges and Potential

Assessment of the Clusterability of Data Using a Multimodal Convolutional Neural Network

Self-Tuning Control Using an Online-Trained Neural Network to Position a Linear Actuator

Contact Info

Product

Resources

About