While deep learning methods continue to improve in predictive accuracy on a wide range of application domains, significant issues remain with other aspects of their performance including their ability to quantify uncertainty and their robustness. Recent advances in approximate Bayesian inference hold significant promise for addressing these concerns, but the computational scalability of these methods can be problematic when applied to large-scale models. In this paper, we describe initial work on the development of URSABench (the Uncertainty, Robustness, Scalability, and Accuracy Benchmark), an open-source suite of benchmarking tools for comprehensive assessment of approximate Bayesian inference methods with a focus on deep learningbased classification tasks. 1
With the advent of powerful analysis tools, intelligent medical diagnostics for neurodegenerative disease (NDs) diagnosis are coming close to becoming a reality. In this work, we describe a state-of-theart machine-learning system with multiclass diagnostic capabilities for the diagnosis of NDs. Our framework for multiclass subject classification comprises feature extraction using principal component analysis, feature selection using Fisher discriminant ratio, and subject classification using least-squares support vector machines. A multisite, multiscanner data set containing 2540 patients clinically diagnosed as Alzheimer Disease (AD), healthy controls (HC), Parkinson disease (PD), mild cognitive impairment (MCI), and scans without evidence of dopaminergic deficit (SWEDD) was obtained from Parkinson's Progression Marker Initiative and Alzheimer's Disease Neuroimaging Initiative. Our work assumes significance since studies have primarily focused on comparing only two subject classes at once, i.e., as binary classes. To profile the diagnostic capabilities for realtime clinical practice, we tested our framework for multiclass disease diagnostic capabilities. The proposed method has been trained and tested on this cohort (2540 subjects), the largest reported so far in the literature. For multiclass diagnosis, our method results in highest reported classification accuracy of 87.89 ± 03.98% with a precision of 82.54 ± 08.85%. Also, we have obtained accuracy of up to 100% for binary class classification of NDs. We believe that this study takes us one step closer to translating machine learning into routine clinical settings as a decision support system for ND diagnosis.
Approximate Bayesian deep learning methods hold significant promise for addressing several issues that occur when deploying deep learning components in intelligent systems, including mitigating the occurrence of overconfident errors and providing enhanced robustness to out of distribution examples. However, the computational requirements of existing approximate Bayesian inference methods can make them ill-suited for deployment in intelligent IoT systems that include lower-powered edge devices. In this paper, we present a range of approximate Bayesian inference methods for supervised deep learning and highlight the challenges and opportunities when applying these methods on current edge hardware. We highlight several potential solutions to decreasing model storage requirements and improving computational scalability, including model pruning and distillation methods.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.