Seyed Soroush Majd

Education

Research Interests

• Natural Language Processing
• Machine Learning
• Deep Learning
• Trustworthy AI

Teaching Experience

Publications

Research Experience

1. Improving Semantic Textual Similarity Using Deep Learning

   Improving clinical Semantic Textual Similarity (STS) using metric learning. Developing training strategies to enhance model representation to overcome challenges such as limited clinical data.

---

2. Multi-Wavelength Video-Based Blood Perfusion Mapping in Human Body Skin using rPPG

   Researched different rPPG (Remote Photoplethysmography) signal extraction methods using videos from the skin at different light wavelengths to map blood flow below the skin with the cooperation of The Color and Visual Computing Laboratory (Colourlab) at the Norwegian University of Science and Technology (NTNU).

---

3. Heart Rate Measurement and Blood Perfusion Mapping in Parts of Human Body Skin using rPPG

   Estimated the heart rate and mapped blood flow below the skin using facial videos with a non-invasive method called rPPG. Involves detecting the Region of Interest (ROI) (using Neural Network, Thresholding, and Manual Selection), extracting the rPPG signal from the ROI (using two methods of calculating the average pixel intensity and Independent Component Analysis (ICA)), and calculating the Fourier Transform of the rPPG signal to identify the frequency with the highest amplitude as the estimated heart rate

Selected Coursework

Notable Projects

1. Robust FAQ Question Answering using Metric Learning (Angular Loss) and SPARQL-Based Ontology Querying
2. Evaluating Fairness, Backdoor Attack, and Out-Of-Distribution Detection in ResNet18 Image Classification
3. Model Interpretation using SHAP for Regression (MLP) and LIME for Image Classification (MobileNet v2)
4. Enhancing Robustness and Generalization in ResNet18 Classification with Angular Loss on Limited Training Data (20% of CIFAR-10 dataset)
5. Question Answering Using Cosine Similarity Between Input Questions and Constructed Sentences (Subject, Object, Predicate) from the RDF Ontology
6. Comparing Performance of Contextual and Static Embeddings in Analogical Reasoning Tasks
7. Image Colorization with Convolutional Neural Network
8. Tata Steel Ltd. Stock Price Prediction Using Hidden Markov Model
9. Text Classification on WELFake Dataset with BERT and RoBERTa
10. Multi-Layer Perceptron for Sign Language Image Classification
11. Human Activity Recognition with CNN-LSTM
12. Google Stock Price Prediction Based on GRU and LSTM models
13. Earthquake Prediction in Iran
14. Baseline Wander Removal in ECG Signals Using FIR Filter, IIR Filter, EMD, Moving Average, and Wavelet Transform
15. Comparing Object Recognition in Humans and Deep Convolutional Neural Networks Based on Kay et al. (2008)

Honors and Awards

• Achieved top 1% place among all applicants of the Nationwide University Entrance Exam (Konkour) for B.Sc. in Engineering among 162,879 applicants, Iran, 2016.
• Member of National Organization for Development of Exceptional Talents (NODET), Tehran, Iran, 2012--2016.

Hobbies

• Roller Skating
• Working Out
• Gaming

References