A5069802999- Autonomous Vehicle Technology and Safety
- Domain Adaptation and Few-Shot Learning
- Medical Imaging and Analysis
- Radiomics and Machine Learning in Medical Imaging
- COVID-19 diagnosis using AI
- Traffic control and management
- Advanced Neural Network Applications
- AI in cancer detection
- Brain Tumor Detection and Classification
- Advanced Neuroimaging Techniques and Applications
- Medical Image Segmentation Techniques
- Cell Image Analysis Techniques
- Reinforcement Learning in Robotics
- Natural Language Processing Techniques
- Traffic and Road Safety
- Advanced MRI Techniques and Applications
- Machine Fault Diagnosis Techniques
- Occupational Health and Safety Research
- Topic Modeling
- Traffic Prediction and Management Techniques
- Enhanced Oil Recovery Techniques
- Stellar, planetary, and galactic studies
- Machine Learning in Bioinformatics
- Artificial Intelligence in Healthcare
- Spinal Dysraphism and Malformations
Polytechnique Montréal
2018-2022
Gray matter (GM) tissue changes have been associated with a wide range of neurological disorders and was also recently found relevant as biomarker for disability in amyotrophic lateral sclerosis. The ability to automatically segment the GM is, therefore, an important task modern studies spinal cord. In this work, we devise modern, simple end-to-end fully automated human cord gray segmentation method using Deep Learning, that works both on vivo ex MRI acquisitions. We evaluate our against six...
Despite the fast developmental pace of new sentence embedding methods, it is still challenging to find comprehensive evaluations these different techniques. In past years, we saw significant improvements in field embeddings and especially towards development universal encoders that could provide inductive transfer a wide variety downstream tasks. this work, perform evaluation recent methods using linguistic feature probing We show simple approach bag-of-words with recently introduced...
Segmentation of axon and myelin from microscopy images the nervous system provides useful quantitative information about tissue microstructure, such as density thickness. This could be used for instance to document cell morphometry across species, or validate novel non-invasive magnetic resonance imaging techniques. Most currently-available segmentation algorithms are based on standard image processing usually require multiple steps and/or parameter tuning by user adapt different modalities....
Pyevolve is an open-source framework for genetic algorithms. The initial long-term goal of the project was to create a complete and multi-platform algorithms in pure Python. However, most recent developmental versions currently support also Genetic Programming (GP)[3]; accordingly, now aims at becoming Python evolutionary
The goal of autonomous vehicles is to navigate public roads safely and comfortably. To enforce safety, traditional planning approaches rely on handcrafted rules generate trajectories. Machine learning-based systems, the other hand, scale with data are able learn more complex behaviors. However, they often ignore that agents self-driving vehicle trajectory distributions can be leveraged improve safety. In this paper, we propose modeling a distribution over multiple future trajectories for...
It is not a secret that recent advances in deep learning (1) methods have achieved scientific and engineering milestone many different fields such as natural language processing, computer vision, speech recognition, object detection, segmentation, to name few. Different applications of medical imaging started appear first workshops, conferences then journals. According survey (2), the number papers grew rapidly 2015 2016. Nowadays, are pervasive throughout entire community, with...
The human spinal cord is a central nervous system structure that plays an important role in normal motor and sensory function, can be affected by many debilitating neurologic diseases. Due to its clinical importance, the frequently subject of imaging research. Common methods for visualizing anatomy pathology include histology magnetic resonance (MRI), both which have unique benefits drawbacks. Postmortem microscopic resolution MRI fixed specimens, sometimes referred as microscopy (MRM),...
Semantic segmentation is a crucial task in biomedical image processing, which recent breakthroughs deep learning have allowed to improve. However, methods general are not yet widely used practice since they require large amount of data for training complex models. This particularly challenging images, because and ground truths scarce resource. Annotation efforts images come with real cost, experts manually label at pixel-level on samples usually containing many instances the target anatomy...
Despite promising progress in reinforcement learning (RL), developing algorithms for autonomous driving (AD) remains challenging: one of the critical issues being absence an open-source platform capable training and effectively validating RL policies on real-world data. We propose DriverGym, OpenAI Gym-compatible environment specifically tailored driving. DriverGym provides access to more than 1000 hours expert logged data also supports reactive data-driven agent behavior. The performance...
Time-resolved spectra throughout the orbit of EF Eri during its low accretion state were obtained with solar blind channel on Advanced Camera for Surveys board Hubble Space Telescope. The overall spectral distribution exhibits peaks at 1500 and 1700 Å, while UV light curves display a quasi-sinusoidal modulation over binary orbit. Models white dwarfs (WDs) hot spot cyclotron emission attempted to fit variations A non-magnetic WD temperature ∼10,000 K central 15,000 generally match broad...
Recent advances in deep learning methods have come to define the state-of-the-art for many medical imaging applications, surpassing even human judgment several tasks. Those models, however, when trained reduce empirical risk on a single domain, fail generalize applied other domains, very common scenario due variability of images and anatomical structures, across same modality. In this work, we extend method unsupervised domain adaptation using self-ensembling semantic segmentation task...
This study describes the experimental application of Machine Learning techniques to build prediction models that can assess injury risk associated with traffic accidents. work uses an freely available data set accident records took place in city Porto Alegre/RS (Brazil) during year 2013. also provides analysis most important attributes a could produce outcome people involved accident.
The cost of wind energy can be reduced by using SCADA data to detect faults in turbine components. Normal behavior models are one the main fault detection approaches, but there is a lack consensus how different input features affect results. In this work, new taxonomy based on causal relations between and target presented. Based taxonomy, impact feature configurations modelling performance evaluated. To end, framework that formulates as classification problem also
Uncertainty quantification for deep neural networks has recently evolved through many techniques. In this work, we revisit Laplace approximation, a classical approach posterior approximation that is computationally attractive. However, instead of computing the curvature matrix, show that, under some regularity conditions, can be easily constructed using gradient second moment. This quantity already estimated by exponential moving average variants Adagrad such as Adam and RMSprop, but...