Alexander Wong's research while affiliated with University of Waterloo and other places

... In this work, we separate synonyms from both acronyms and abbreviations due to their importance in medical domain (Yu et al., 2002). While synonymous relations could be implicitly learned from pretrained language model (LM) (Michalopoulos et al., 2022;Li et al., 2022), previous researches show that language models are only limited to biomedical (Sung et al., 2021) or clinical knowledge bases (Yao et al., 2022) due to the data sparsity challenge in the medical domain. An explicit way of adding such medical knowledge into language model should be explored. ...

... Table 2 shows the results for the ranking task on SemEval-07 and SWORDS. Michalopoulos et al. (2022) harness WordNet (Fellbaum, 1998) to obtain synsets of the target word and also their glosses, and employ BERT and RoBERTa (Liu et al., 2019) to rank candidates. The models proposed by Lacerra et al. (2021a) are different from the others in that they fine-tune BERT on lexical substitution data sets. ...

... There exists an accumulation effect of quantization error in quantization [37], which is manifested by the increasing tendency of quantization error in the network influenced by the quantization of previous layers. Since PTQ does not contain quantization training, this accumulation effect is more obvious in reconstruction loss. ...

... UmlsBERT (Michalopoulos et al., 2021) also integrates external knowledge resources to improve biomedical language models. The authors updated the masked language modeling in the pre-training step with the associations between the words specified in the UMLS. ...

... Although a geometry-guided postprocessing approach based on accurate edges could lead to an effective and fast building polygon reconstruction, it has not been fully explored. Most recent deep learning based approaches tend to ignore this explicit knowledge integration process [3], [7], [16], and thereby cannot efficiently leverage the learned knowledge regarding the prior geometry shape for building outline vectorization. In fact, given accurate building edges with relatively low commission and omission error, a geometry-guided approach can efficiently locate and estimate building corners by leveraging the interaction among adjacent edges and shape prior knowledge learned from building segmentation. ...

... In the fight against the pandemic, computer-aided screening of patients using radiography images has served as a complementary approach to standard polymerase chain reaction (PCR) test. Despite our efforts in improving the performance of deep learning models [1,2], a compromise often must be made between the performance and model trust [3]. ...

... Although this review has focused on OCT in a comprehensive manner, there are similar GAN applications in other areas of ophthalmic image analysis including for retinal fundus photography (RF) [77][78][79][80][81][82][83] and OCT angiography (OCTA) . 84 The area of GANs is rapidly evolving and this review aims to provide a window into their wide and increasing level of application to optical coherence tomography. ...

... Deep neural networks have recently been widely used, especially for multi-step prediction. Standard deep neural network (DNN) models, such as convolution neural networks (CNN), long short-term memory (LSTM), gated recurrent units (GRU), encoder-decoder, and transformer models, have been commonly used in computer vision [14], image classification [15], time series prediction [16], natural language processing [17], and other fields. Compared to methods for building PDEs, deep learning demonstrates powerful modeling capabilities with large datasets that can be deployed on modern computer systems. ...

... Existing studies on soil water content simulations in the root zone of crops can be divided into mechanistic models based on the physical laws of water transport and machine learning models based on data characteristics. Machine learning soil moisture prediction methods represented by deep learning models are emerging areas of research [7][8][9]. However, the high cost (data acquisition), computing power (model training), and threshold (model building) required for deep learning limit its application in soil water content (SWC) prediction [10]. ...

... data predefined and formatted to a set structure) healthcare data [143]. While unstructured data present some significant challenges for biomedical informatics, several solutions based on text-mining have been reported [144,145]. Text mining is an AI technology that uses the natural language processing technique to extract facts, relationships, and assertions from unstructured text in documents or databases and transform this type of information into structured data that is suitable for analysis, such as ML models [146]. ...