Seyyed Saleh Hosseini

Seyyed Saleh Hosseini

Machine Learning R&D
Canada

About Me

Proficient in modeling and optimizing modern generative ML architectures for CV and medical applications. Experienced in developing large NLP-based neural networks for sentiment analysis in healthcare services. Coding ski…

Experience

Machine Learning R & D

University Health Network (UHN), Toronto
Sep 2022 - Present · 3 years 10 months

Designed and developed an advanced denoising diffusion probabilistic model (DDPM) with a UNet backbone including ResNet and Attention blocks to generate synthesized CTs from MRIs.
Reduced discrepancy between real CT and sCT to less than 1% in dosage projections by RayStation, ensuring accuracy in cancer treatment.
Developed and optimized an all-around GAN framework including Conditional GANs, CycleGANs, and StyleGANs with multiple serially-connected ResNets or UNets to generate high-resolution synthesized 512×512 CTs from 256×256 MRI patches.
Achieved PSNR of 30dB and SSIM of 0.9.
Partially automated sCT generation from MRI via U-Mamba auto-segmentation and shape matching (Hu Moments metrics).
Streamlined decision-making on sCTs for specialists.
Conducted preprocessing on DICOM images, including conversion into slices and application of image normalization techniques such as registration and min-max normalization.
Performed post-translation reconversion to DICOM format for seamless integration into medical imaging systems.
Distributed data-parallel (DDP) training on Azure to parallelize the training of advanced generative algorithms such as Cycle GAN, Conditional GANs, and DDPM.
Accelerated the training process and allowed for quicker convergence and optimization of the models.
Refined BERT for patient sentiment prediction through analysis of healthcare professionals’ notes to enhance the accuracy and depth of insights into patients’ satisfaction levels.
Achieved an 85% F1 score by mitigating the data imbalance through rigorous precision and recall analysis.

Machine Learning R&D

UHN
Feb 2022 - Present · 4 years 5 months

"N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A"

AI Research Associate

McGill University & InterDigital, Inc., Montreal
Sep 2017 - Sep 2022 · 5 years

Designed and developed advanced CNN-based methods for channel tracking in THz systems, leveraging VGGNets.
Achieved 67% improvement in RMSE.
Pioneered a low-complexity deep neural network (DNN) for wideband Direction of Arrival estimation.
Dramatically reduced processing time.
Achieved a 96% improvement in processing time compared to Root-MUSIC.
Innovated Bert architectures with CNN and DNN placed at the top for anomaly detection in cell-free systems.
Increased detection probability by 132.5% at low transmission powers.
Improved processing time by 72%.

Skills

Machine Learning Algorithms Machine Learning Research & Development Generative Adversarial Networks Feature Engineering Anomaly Detection Deep Learning Transformers Self-Attention Natural Language Processing Data Visualization Statistical Analysis Problem-Solving Critical Thinking Model Development Deployment Diffusion Models Python C++ SQL MATLAB DNNs CNNs ResNets GANs BERT DDPM UNets U-Mamba MLOps NLP Progressive Growing of GANs PyTorch LibTorch Keras TensorFlow Pandas Azure NumPy SciPy Scikit-Learn FastAPI OpenCV DDP Git Docker
Report this Profile?