StFX Scholar (St. Francis Xavier University)
Not a member yet
55051 research outputs found
Sort by
Bonita (MacDonald) Stone
No full textposted: SaltWire, Jan. 17, 2022; funeral arrangements by C. L. Curry Funeral Services; "Bonnie"
Eileen Margaret Cecelia Hurley
No full textposted: SaltWire, Jan. 19, 2022; funeral arrangements by G. W. Giffin Funeral Home; interpreter at Historic Sherbrooke Village Museu
Angus L. Macdonald Library’s John Stewart Collection
No full textThis volume was created by the students of English 226: From Tablets to Tablets: Texts and Technologies, working with Professor Laura Estill and Susan Cameron, Special Collections Librarian, St Francis Xavier University
A Comparison of Multiple Supervised Learning Techniques for Conducting Text Classification
No full textNatural language processing (NLP) is a subfield of Computer Science research which utilises computers to understand, and use natural human language to accomplish some task [4]. Some examples of the various applications of NLP include machine translation, text processing and summerisation, information retrieval, speech recognition and much more [4]. Text classification as a subject of research in and of itself is rather trivial, with Jurafsky discussing in their book, “Speech and Language Processing” that in NLP, logistic regression is the baseline supervised machine learning algorithm for classification [10]. This thesis aims to compare the results of multiple implementations of classification, including logistic regression, K-Nearest Neighbours, and classification with an LSTM Recurrent Neural Network, to predict the genre of a film, using the film’s plot synopsis. Three different experiments will be conducted. One using random training and testing splits of the imbalanced training data set, one using the complete imbalanced training data set for training and the complete imbalanced test data set for testing, and one using random training and testing splits of a balanced version of the training data set. The accuracy of the three models, as well as the learning capabilities of each model, will be discussed in this thesis, where the goal to investigate which of the logistic regression, K-Nearest Neighbours or LSTM models are best suited for text classification
David Hubert Mattie
No full textposted: SaltWire, Apr. 5, 2022; funeral arrangements by C. L. Curry Funeral Service
Joseph Henry Parker
No full textposted: SaltWire, Jan. 13, 2022; funeral arrangements by Dennis Haverstock Funeral Homes; "Joe"
Effective Vaccination Strategy for Infectious Diseases by Analyzing the Age and Comorbidity Attributes of Individuals on Social Network
No full textMarie Amelia McPhee
No full textposted: SaltWire, Mar. 18, 2022; funeral arrangements by Green's Funeral Hom
Elizabeth Bridget (Martin) Smith
No full textposted: SaltWire, Apr. 18, 2022; funeral arrangements by MacIsaac Funeral Home; "Bette"
Novel metrics for evaluation and validation of regression-based supervised learning
No full textError consistency is a validation metric for evaluating the sample-based error variability across machine learning models trained as part of in-lab validation. Many machine learning (ML) based regression algorithms are likely to be inconsistent with each other when trained repeatedly on the same task as part of standard cross validation, in part due to sampling, but also, potentially associated with the inclusion of randomness in their training paradigms, which is common in many learning techniques. In this work, we propose a novel approach to validation and evaluation of regression-based learning algorithms, called regression ‘error consistency’ (EC) to assist in assessing sample-wise consistency of errors as part of in-lab validation. We have applied novel EC metrics on simulated datasets generated from different distributions, with three regression algorithms (i.e., Knn-1, Knn-5, and Linear regression) and found weak to moderate correlations with well-known goodness of fit (GoF) metrics like mean absolute error (MAE), and mean squared error (MSqE). We have also used EC metrics on six real-world datasets with six different regression models and compared the results with previously developed classification EC. The results demonstrate that, out of six models, the random forest achieved high accuracy but exhibited less consistency in its error profiles. This finding matches with classification based EC results. In addition, we compare the correlation between classification EC and regression EC using three real-world datasets and four ML model pairs. The findings demonstrate similar behaviour between the classifiers and the regression models. Moreover, we applied the EC metrics on the MNIST digits dataset using a convolutional neural network (CNN) as part of a preliminary deep learning experiment. Though MNIST is typically treated as a classification dataset, we considered this dataset as a regression problem and the CNN model developed demonstrated good performance. Furthermore, we applied the novel EC metrics to the result of the regression U-Net (rUNet) for longitudinal image estimation of patients being monitored for neural degeneration, and generated parametric map images from the EC metric result. The parametric map images were able to identify the region of the predicted image with contrast defined by the error consistency itself, providing the user with a visual assessment of the spatial distribution of the consistency of incorrect predictions. Finally, with the MIMIC-III dataset, a long short-term memory (LSTM) recurrent neural network (RNN) was employed to assess the consistency of temporal prediction in the future. The intersection-union related EC metrics demonstrated that the consistency of the LSTM model was decreasing, while the ratio related EC metrics showed that the consistency was increasing as predictions extend into the future, and all EC metrics report the fluctuations between trained model prediction error consistency over time. We believe that the proposed EC metrics will be useful in evaluating and validating regression algorithm error consistency, including in deep learning, and will hopefully guide the machine learning research community to develop more reproducible and predictable (in terms of the errors they will make) regression algorithms