Ramesh Basnet

Researcher and developer of an AutoML system for embedded devices focused on neural architecture search, automated hyperparameter optimization, parallel computation, quantization, and deployment of computer vision models.

Delivered production-ready tooling to accelerate model iteration speed while meeting strict power and latency constraints on edge hardware.

Spearheaded development of advanced ML algorithms for cardiovascular healthcare, including 3D medical image segmentation, motion artifact remediation, and disease biomarker detection across CMR and CCTA data.

Collaborated with cardiologists and research teams in North America and Europe, producing published research integrated into the cvi42® platform.

Developed state-of-the-art deep learning algorithms for medical image processing with journal publications in IEEE TMI and BSPC.

Ranked second in a performance metric and placed among the top eight participants globally in the MICCAI iSeg 2019 challenge.

Designed ML models for continuous emotional state prediction from audio-visual data, delivering robust performance and first-authored publications in Pattern Recognition Letters and ICTCS.

Presented news, commentaries, economic reviews, and special talks live on air. Organized opening, musical, and closing sessions of the program.

Collaborated with artists of different nationalities and languages to translate news.

Thesis: A Parameter-Efficient Deep Dense Residual Convolutional Neural Network for Volumetric Brain Tissue Segmentation from Magnetic Resonance Images.

Supervisors: Dr. M. Omair Ahmad and Dr. M.N.S. Swamy.

Thesis: Estimation of Instantaneous Affective Dimension from Audio-Visual Data Using Two-Stream CNN.

Supervisor: Dr. S. M. Mahbubur Rahman.

Host a weekly radio program dedicated to promoting Nepali culture, traditions, and heritage.

Conduct interviews with guests to share insights and experiences on successful settlement in Canada, diverse opportunities, and business endeavors.

Collected relief funds for the victims of the April 2015 Nepal earthquake from universities in Bangladesh.

Supplied relief materials with support from the Embassy of Nepal, Dhaka, and organized a candlelight vigil in memory of lost lives.

Evaluates incorporating explicit anatomic localization into SSCL pretraining for CMR to improve downstream diagnostic performance and saliency.

Compares DL and radiomics approaches to classify ischemic vs non-ischemic cardiomyopathy using CMR short axis cine imaging.

Introduces a compact 3D U-Net style architecture with dense connections and residual links for efficient brain tissue segmentation.

Reviews top-performing methods and highlights challenges in multi-site infant brain segmentation consistency.

Investigates two-stream CNNs and feature selection for predicting continuous emotional state dimensions from audiovisual data.

Presents mutual information-based selection of CNN features for frame-by-frame emotional state prediction.