Mahdiyar Ali Akbar Alavi
Education
University of Tehran Tehran, Iran
-
Bachelor of Science in Electrical Engineering September 2017 - July 20222017 - 2022
- Overall GPA: 16.75/20 (3.52/4)
- GPA of the Last 2 Years: 17.75/20 (3.79/4)
-
Minor in Computer Engineering September 2019 - February 20222019 - 2022
- Overall GPA: 18.29/20 (3.8/4)
National Organization for Development of Exceptional TalentsNODET Tehran, Iran
-
Diploma in Mathematics and Physics Discipline September 2013 - June 20172013 - 2017
- Overall GPA: 19.88/20 (4/4)
Research Interests
- Machine LearningML
- Deep LearningDL
- Natural Language ProcessingNLP
- Computer Vision
- Vision and Language
- Human-Robot InteractionHRI
Relevant Courses
(Graduate courses are indicated by ✝)
- Artificial Intelligence | Score: 19.75/20
- Neural Networks✝ | Score: 16/20
- Linear Algebra | Score: 18.7/20
- Operational Research | Score: 18/20
- Advanced Programming | Score: 19/20
- Data Structures | Score: 20/20
- Discrete Mathematics | Score: 17.2/20
- Operating Systems | Score: 17.1/20
- Computer Architecture | Score: 19.3/20
- Modern Control Systems | Score: 19.1/20
- Distributed Systems | Audited
- Fundamentals of Mechatronics EngineeringFund. of Mechatronics Eng. | Score: 16.5/20
Research Experience and Notable Projects
-
B.Sc. Thesis: Improving OOD Generalization of Transformer-based NLI Models using Dataset Cartography February 2022 - July 20222022 Supervisor: Dr. Yadollah Yaghoobzadeh
Through Dataset Cartography, I fine-tuned RoBERTa-base and Bert-base-uncased models using various datasets in different ways. For example, first, I fine-tuned models with the whole dataset and then with its most ambiguous samples. Furthermore, I evaluated the models using evaluation sets, such as HANS and SuperGLUE Diagnostic Dataset. By employing this method, mnli-6-var33-3 was born (a fine-tuned RoBERTa-base model for MNLI, which reported an improvement of 2.53% in the evaluation process using HANS).
Description
Through Dataset Cartography, I fine-tuned RoBERTa-base and Bert-base-uncased models using various datasets in different ways. For example, first, I fine-tuned models with the whole dataset and then with its most ambiguous samples. Furthermore, I evaluated the models using evaluation sets, such as HANS and SuperGLUE Diagnostic Dataset. By employing this method, mnli-6-var33-3 was born (a fine-tuned RoBERTa-base model for MNLI, which reported an improvement of 2.53% in the evaluation process using HANS).
-
Gesture-controlled Robot using Arduino and Python (MediaPipe) November 20222022 Supervisor: Dr. Mehdi Tale Masouleh
I built a 4-wheel drive gesture-controlled mobile robot during the Arduino Instruction work experience at Robotech Academy. This robot was simply controllable with hand gestures through Arduino Uno and Python MediaPipe Computer Vision library.
This project was designed to assess students of the introductory robotics tutorial course.Description
I built a 4-wheel drive gesture-controlled mobile robot during the Arduino Instruction work experience at Robotech Academy. This robot was simply controllable with hand gestures through Arduino Uno and Python MediaPipe Computer Vision library.
This project was designed to assess students of the introductory robotics tutorial course. -
Solving a Linear Programming Problem (Optimal Vehicle Routing) using Python (PuLP) February 20222022 Course Title: Operational Research
First, a dataset was collected through Google Maps, which included ten famous sites in Tehran. Then, inspired by the Network Flow Problem, I coded a Python script using Linear Programming to find the shortest path between two arbitrary given places in the dataset.
Description
First, a dataset was collected through Google Maps, which included ten famous sites in Tehran. Then, inspired by the Network Flow Problem, I coded a Python script using Linear Programming to find the shortest path between two arbitrary given places in the dataset.
-
An Instagram Bot (InstaCrawler) for Automatic Data Collection using Python (Selenium) September 20212021 Supervisor: Dr. Reshad Hosseini
During my internship at HARA AI, I developed a Python module using the Selenium library, which was able to log into the Instagram website, like new posts, visit unseen stories, and download pictures. Collected pictures could then be used to train deep learning models in the field of Computer Vision.
This project's most challenging part was making the bot behave like ordinary users, as the Instagram website can easily detect unusual behaviors and ban automatic data collectors.Description
During my internship at HARA AI, I developed a Python module using the Selenium library, which was able to log into the Instagram website, like new posts, visit unseen stories, and download pictures. Collected pictures could then be used to train deep learning models in the field of Computer Vision.
This project's most challenging part was making the bot behave like ordinary users, as the Instagram website can easily detect unusual behaviors and ban automatic data collectors. -
Race Recognition using Artificial Neural Networks in Python (Keras) June 20212021 Course Title: Artificial Intelligence
In the first phase of this project, I analyzed the UTKFace dataset through the Pandas, Seaborn, and PyPlot libraries. Then, I prepared the data for the training process and trained an artificial neural network with the processed data using the Keras API. This model could predict each person's ethnicity by having their face image. Afterward, I tried to enhance the accuracy of predictions by changing different training parameters, such as the optimizer function or the kernel regularizer. In the last phase, I used the test data (30% of the dataset) to evaluate the model, which indicated an accuracy of 71%.
Description
In the first phase of this project, I analyzed the UTKFace dataset through the Pandas, Seaborn, and PyPlot libraries. Then, I prepared the data for the training process and trained an artificial neural network with the processed data using the Keras API. This model could predict each person's ethnicity by having their face image. Afterward, I tried to enhance the accuracy of predictions by changing different training parameters, such as the optimizer function or the kernel regularizer. In the last phase, I used the test data (30% of the dataset) to evaluate the model, which indicated an accuracy of 71%.
-
House Price Prediction using Multithreading in C++ (PThread) June 20212021 Course Title: Operating Systems
-
Multi-cycle Stack-based Processor Design in Verilog May 20212021 Course Title: Computer Architecture (Digital Systems II)
-
Classification of Persian (Farsi) Books using Naïve Bayes Classifier in Python May 20212021 Course Title: Artificial Intelligence
-
Image Restoration using Discrete Hopfield Network in Python December 20202020 Course Title: Neural Networks
-
A Two-Player Computer Game (Soccer Stars) using Object-Oriented Programming in C++ November 20202020 Course Title: Advanced Programming
-
Document Detection using MATLAB August 20202020 Course Title: Digital Signal Processing (DSP)
-
Red-Black Tree Implementation using Python July 20202020 Course Title: Data Structures and Algorithms
Teaching Assistant Experience
-
Fundamentals of Mechatronics Engineering | Role: Chief Teaching Assistant Spring 20222022 Instructor: Dr. Mehdi Tale Masouleh
-
Computer Architecture (Digital Systems II) | Role: Teaching Assistant Spring 20222022 Instructor: Dr. Saeed Safari
-
Computer Architecture Laboratory | Role: Laboratory Assistant Instructor Spring 20222022 Instructor: Dr. Saeed Safari
-
Introduction to Computing Systems and Programming | Role: Supervising Teaching Assistant Fall 20212021 Instructors: Dr. Hadi Moradi and Dr. Mostafa Tavassolipour
-
Electrical Measurement Laboratory I | Role: Laboratory Assistant Instructor Fall 20182018 Instructor: Dr. Hossein Iman-Eini
Work Experience
-
Robotech Academy
| Role: Part-time Arduino Instructor June 2022 - March 20232022-2023 Description: We are preparing an introductory robotics tutorial course, and I undertake the Arduino instruction part. -
Karyar College
| Role: Volunteer Front-end Development Teaching Assistant August 2021 - October 20212021 Description: I used to hold weekly Q&A sessions for students who were learning front-end development. -
HARA AI
| Role: Part-time Summer Intern July 2021 - September 20212021 Description: I created a Python module (InstaCrawler) to automatically collect data from Instagram.
Honors and Awards
-
Ranked 533rd among 148,000 contestants in the National University Entrance Exam in the field of Mathematics and Physics July 20172017
Licenses and Certificates
-
Natural Language Processing in Python Track | Presented by: DataCamp | View Certificate March 20222022
Programming Skills
-
Python (Selenium, PuLP, NumPy, Pandas, and Keras)
-
MATLAB
-
Verilog HDLC/C++ (Object-Oriented Programming)
-
Verilog HDLC/C++ (Object-Oriented Programming)
-
Front-end Development (HTML/CSS/jQuery)
-
Arduino
Languages
-
Persian (Farsi) | Native
-
English | Advanced
- IELTS Academic Band Score: 7.5 (L: 8.5, R: 8.0, S: 7.5, W: 6.5)IELTS Academic Band Score: 7.5
-
French | Elementary
References
Available upon request.