Morteza Heidari

Professional Experience

Staff Machine Learning Engineer, Intel Corporation

Folsom, CA, USA | May 2024 – Present

Implementing and designing attention-based or transformer-based modeling (e.g., state-of-the-art NLP and Vision models or any other sequence-to-sequence-based deep learning models).

Optimizing large language models (LLMs) for deployment, focusing on time and space efficiency.

Developed transformer-based video sign language translation models that convert sign language video frames into American Sign Language labels, followed by a fine-tuned and quantized LLM to generate English text in real-time.

Designed and implemented a novel algorithm that reduced LLM model size by over 200%, exceeding the original goal of 10%, enabling significant efficiency improvements in model deployment.

Pioneered a latency-reduction technique that cut LLM inference time on CPU by over 25%, tripling the initial goal of 5%, facilitating faster real-time applications.

Leadership: Designed tasks for junior engineers on ML/Software optimizations, defining performance metrics, and benchmarking algorithms.

Tech Lead: Leading an ISV in designing and developing a deployment framework for Mixtral-8x7 on AI-enabled PCs.

Leading multiple ISVs in enabling AI features on inference phase for AI PCs and edge devices.

Imaging Scientist II, Philips Pharma Solutions

Rochester, NY, USA | May 2021 – May 2024

Developer: Design computer vision and deep learning methods to develop algorithms for 3D stitching, regression, classification, out-painting, registration, segmentation, super-resolution, parameter estimation, etc.

Developing an Image generation algorithm to reconstruct the full leg and calculate knee angle with robust results for the Philips for the first time.

Developing 3D image alignment to align 3D MRI images and reduce the cost of redoing MRI images.

Developing a pipeline for loading datasets, categorizing them, and stitching the related ones. Applicable to reduce data loading time for the company from 2 hours to 10 minutes.

Designed advanced NLP models for text classification, sentiment analysis, and language generation tasks.

Implemented and fine-tuned Large Language Models (LLMs) for complex language understanding and generation problems.

Researcher: Research and develop new algorithms for image processing, computer vision, and deep learning.

Tech Lead: MRI and CT Imaging evaluation.

Medical Imaging Analytical Intern, Siemens Healthineers

Malvern, PA, USA | June 2019 – August 2019

Development and assessment of a new deep learning algorithm to classify different views of echocardiogram imaging with 95% classification accuracy.

Presenting image processing tools to prepare the dataset for the purpose of classification.

Analyzing and proposing the best deep learning algorithm effective for this task.

Research Assistant & Software Developer, University of Oklahoma

Norman, OK, USA | January 2017 – May 2021

Published 13+ papers in top-tier journals in machine learning, deep learning, artificial intelligence, medical imaging, and data science.

Developed deep learning model to predict the likelihood of COVID-19 using chest X-ray images with preprocessing algorithms. The system provided 94.5% accuracy for three classes of classification.

Developed a hybrid technique to optimize CNN (Xception, InceptionV3) features with classical data optimization techniques for lesion classification. 91% of accuracy is achieved for lesion classification for this system.

Feature extraction, optimization, and data evaluation:

Applied a random projection algorithm to optimize the machine learning model for breast lesion classification. An overall AUC value of 0.84 showed the effectiveness of my method.

Proposing data regeneration as an optimal way to reduce redundancy of features for breast mass classification and predicting the likelihood of malignant cases.

Developed and assessed a new global mammographic image feature analysis scheme to predict the likelihood of malignant cases. The algorithm provided 92% of accuracy for malignant analysis.

Designed a bilateral scheme to mimic radiologists' vision system to predict risk score of the likelihood of cancer in breast images. The algorithm showed a 9.7% improvement in risk prediction accuracy in comparison with the baseline.

Designed a machine learning-based algorithm for Speech segmentation and enhancement to detach vowels and consonants and prepare them for steganography systems.

Developed an algorithm for image steganalysis based on frequency domain features and machine learning algorithms.