Boosting the Robustness of Capsule Networks with Diverse Ensemble

Author(s):  
Yueqiao Li ◽  
Hang Su ◽  
Jun Zhu ◽  
Jun Zhou
Keyword(s):  
2019 ◽  
Vol 11 (11) ◽  
pp. 1259 ◽  
Author(s):  
Eike Jens Hoffmann ◽  
Yuanyuan Wang ◽  
Martin Werner ◽  
Jian Kang ◽  
Xiao Xiang Zhu

This article addresses the question of mapping building functions jointly using both aerial and street view images via deep learning techniques. One of the central challenges here is determining a data fusion strategy that can cope with heterogeneous image modalities. We demonstrate that geometric combinations of the features of such two types of images, especially in an early stage of the convolutional layers, often lead to a destructive effect due to the spatial misalignment of the features. Therefore, we address this problem through a decision-level fusion of a diverse ensemble of models trained from each image type independently. In this way, the significant differences in appearance of aerial and street view images are taken into account. Compared to the common multi-stream end-to-end fusion approaches proposed in the literature, we are able to increase the precision scores from 68% to 76%. Another challenge is that sophisticated classification schemes needed for real applications are highly overlapping and not very well defined without sharp boundaries. As a consequence, classification using machine learning becomes significantly harder. In this work, we choose a highly compact classification scheme with four classes, commercial, residential, public, and industrial because such a classification has a very high value to urban geography being correlated with socio-demographic parameters such as population density and income.


2020 ◽  
Vol 34 (04) ◽  
pp. 4264-4271
Author(s):  
Siddhartha Jain ◽  
Ge Liu ◽  
Jonas Mueller ◽  
David Gifford

The inaccuracy of neural network models on inputs that do not stem from the distribution underlying the training data is problematic and at times unrecognized. Uncertainty estimates of model predictions are often based on the variation in predictions produced by a diverse ensemble of models applied to the same input. Here we describe Maximize Overall Diversity (MOD), an approach to improve ensemble-based uncertainty estimates by encouraging larger overall diversity in ensemble predictions across all possible inputs. We apply MOD to regression tasks including 38 Protein-DNA binding datasets, 9 UCI datasets, and the IMDB-Wiki image dataset. We also explore variants that utilize adversarial training techniques and data density estimation. For out-of-distribution test examples, MOD significantly improves predictive performance and uncertainty calibration without sacrificing performance on test data drawn from same distribution as the training data. We also find that in Bayesian optimization tasks, the performance of UCB acquisition is improved via MOD uncertainty estimates.


2018 ◽  
Vol 3 (1) ◽  
pp. 81-82
Author(s):  
Soraya Fallah ◽  
Cklara Moradian ◽  
Wendy Murawski

Ability, equity, and culture: Sustaining inclusive urban education reform, edited by Kozleski and Thorius (2014), is a remarkable compilation of work, written by a diverse ensemble of educators, researchers, practitioners, and advocates. A thought-provoking book that looks critically at urban education reform, the authors challenge readers to have a broader understanding of what the term inclusivity entails. The editors present the work of 17 authors who were all part of the National Institute for Urban School Improvement (NIUSI). These authors shed light on various aspects of systemic urban reform in policy, pedagogy, and practice. Issues discussed ranged from the micro to the macro change initiatives to classroom environments and district culture, as well as successful models of student-centered programs around the country. Using data from 12 years of research conducted under the sponsorship of NIUSI, the contributors paint a hopeful, if daunting, a portrait of what equitable, inclusive, and culturally responsive education should and could look like. Ultimately, the contributors of this book believe that sustainable, scalable, successful systemic educational reform is attainable, provided that all stakeholders are committed to cultural responsivity and inclusivity for all students. In order to achieve this goal, the authors posit that reform needs to combat discrimination based on socially constructed notions of difference, such as gender, race, ethnicity, ability, class, and sexual orientation.


2019 ◽  
Author(s):  
Nazanin Donyapour ◽  
Nicole Roussey ◽  
Alex Dickson

Conventional molecular dynamics simulations are incapable of sampling many important interactions in biomolecular systems due to their high dimensionality and rough energy landscapes. To observe rare events and calculate transition rates in these systems, enhanced sampling is a necessity. In particular, the study of ligand-protein interactions necessitates a diverse ensemble of protein conformations and transition states, and for many systems this occurs on prohibitively long timescales. Previous strategies such as WExplore that can be used to determine these types of ensembles are hindered by problems related to the regioning of conformational space. Here we propose a novel, regionless, enhanced sampling method that is based on the weighted ensemble framework. In this method, a value referred to as “trajectory variation” is optimized after each cycle through cloning and merging operations. This method allows for a more consistent measurement of observables and broader sampling resulting in the efficient exploration of previously unexplored conformations. We demonstrate the performance of this algorithm with the N-dimensional random walk and the unbinding of the trypsin-benzamidine system. The system is analyzed using conformation space networks, the residence time of benzamidine is confirmed, and a new unbinding pathway for the trypsin-benzamidine system is found. We expect that REVO will be a useful general tool to broadly explore free energy landscapes.


Author(s):  
Nazanin Donyapour ◽  
Nicole Roussey ◽  
Alex Dickson

Conventional molecular dynamics simulations are incapable of sampling many important interactions in biomolecular systems due to their high dimensionality and rough energy landscapes. To observe rare events and calculate transition rates in these systems, enhanced sampling is a necessity. In particular, the study of ligand-protein interactions necessitates a diverse ensemble of protein conformations and transition states, and for many systems this occurs on prohibitively long timescales. Previous strategies such as WExplore that can be used to determine these types of ensembles are hindered by problems related to the regioning of conformational space. Here we propose a novel, regionless, enhanced sampling method that is based on the weighted ensemble framework. In this method, a value referred to as “trajectory variation” is optimized after each cycle through cloning and merging operations. This method allows for a more consistent measurement of observables and broader sampling resulting in the efficient exploration of previously unexplored conformations. We demonstrate the performance of this algorithm with the N-dimensional random walk and the unbinding of the trypsin-benzamidine system. The system is analyzed using conformation space networks, the residence time of benzamidine is confirmed, and a new unbinding pathway for the trypsin-benzamidine system is found. We expect that REVO will be a useful general tool to broadly explore free energy landscapes.


Sign in / Sign up

Export Citation Format

Share Document