Ahmed Masry
Data Scientist (NLP/ML) @ Arteria AI
Toronto, Ontario, Canada
Email: ahmed.elmasry24653@gmail.com
Latest Update
Presented ChartQA poster at TMLS 2023
About me
I am currently working as an NLP Data Scientist at Arteria AI. Prior to that, I completed my M.Sc. (Thesis-based) in Computer Science from York University where I was working as a research assistant under the supervision of Prof. Enamul Hoque in collaboration with Prof. Shafiq Joty, who is currently a research director at Salesforce AI Research. My research focused on multimodal charts understanding tasks, such as answering questions about charts, summarizing charts, and extracting data from charts.
Our research work has been published in several prestigious venues such as ACL 2022, CVPR 2021, and EuroVis 2022. Moreover, our work has gained considerable attention and was cited by several prestigious research AI teams in their papers including Google (Pix2Struct, DePlot, MatCha, Pali-X), Microsoft & Amazon (ChartReader). Additionally, our ChartQA benchmark has been recognized and cited by OpenAI's GPT-4 official blog post in evaluating their GPT-4 model reasoning capabilities.
Education
M.Sc. in Computer Science
York University, Canada
September 2020 to June 2022
GPA: Exceptional, A+
B.Sc. in Computer Engineering
Koc University (Ranked 1st in Turkey)
September 2017 to June 2022
GPA: 3.9/4.0
Work Experience
Data Scientist @ Arteria AI
May 2022 to Present
Research Assistant @ York University
September 2020 to June 2022
Data Scientist Intern @ ShallowAI
July 2020 to September 2020
Undergraduate Research Assistant @ Koc University
June 2019 to January 2020
Teaching Assistant @ York University
January 2021 to April 2022
Publications
UniChart: A Universal Vision-language Pretrained Model for Chart Comprehension and Reasoning
Ahmed Masry, Parsa Kavehzadeh, Xuan Long Do, Shafiq Joty, and Enamul Hoque
arXiv:2305.14761 (Accepted at EMNLP 2023)
Chart-to-Text: A Large-Scale Benchmark for Chart Summarization
Shankar Kantharaj*, Rixie Tiffany Leong*, Xiang Lin*, Ahmed Masry*, Megh Thakkar*, Enamul Hoque, Shafiq Joty
[*Equal Contribution]
Published at ACL 2022
ChartQA: A Benchmark for Question Answering about Charts with Visual and Logical Reasoning
Ahmed Masry, Do Long, Jia Qing Tan, Shafiq Joty, and Enamul Hoque.
Published at ACL 2022
Chart Question Answering: State of the Art and Future Directions
Enamul Hoque, Parsa Kavehzadeh, Ahmed Masry
Published at EuroVis 2022
Integrating Image Data Extraction and Table Parsing Methods for Chart Question Answering
Ahmed Masry, Enamul Hoque Prince
[Best Paper Award]
Published at ChartQA @ CVPR 2021
Chain FL: Decentralized Federated Machine Learning via Blockchain
C. Korkmaz, H. E. Kocas, A. Uysal, A. Masry, O. Ozkasap and B. Akgun
Published at BCCA 2020
Research Experience
Worked on transformer-based architectures for large-scale pre-training over charts and their application for downstream tasks such as summarization, question answering, and data extraction.
Worked on zero and few-shot prompting of LLMs for chart question answering and summarization.
Pretrained UniChart, a universal chart comprehension model, using novel self-supervised pretraining objectives on 627K charts crawled from diverse online platforms. UniChart surpasses the performance of the leading public model, Google’s MatCha, and achieves state-of-the-art results across various downstream tasks
Implemented a chart data extraction pipeline using key points detection, OCR, and the Hungarian algorithm.
Developed VisionTaPas, a novel state-of-the-art chart question answering (ChartQA) model, incorporating image and tabular data encoders (ViT & TaPas) to process the charts’ modalities (the visual features and underlying data table) and combining them using a cross-modality encoder.
Developed a layout-based chart summarization approach by adapting the T5 model to accommodate the spatial information of the charts’ textual tokens.
Skills
Languages: Python, Java, C
Deep Learning and Machine Learning: PyTorch, TensorFlow, Scikit-learn, Pandas, Numpy, Huggingface, NLTK.
Cloud Experience: Databricks, AWS, GCP, HPC
Information Visualization: Tableau, VegaLite
Achievements & ExtraCurricular
Al Ghurair Foundation for Education STEM Scholarship Program (AGFE)
A merit-based award for highly achieving Arab students to pursue a fully funded STEM degree.
Vehbi Koc Scholar And Dean’s Honor Roll
Merit-based Awards for Academic Excellence
CVPR 2021 Chart Question Answering Workshop
Achieved Best Paper Award.
Al-Ghurair Foundation (AGFE)
Student Ambassador
Robotics Craftsmanship International Academy (University of Coimbra)
Ranked 1st in the sumo robots competition, and in the top 10 out of 84 participants.
Recent News
Paper accepted at EMNLP 2023
October 2023Presented ChartQA Poster at TMLS 2023
June 2023Defended my M.Sc. Thesis at York University
June 2022Presented ChartQA Poster at CVR/VISTA 2022
June 2022Paper accepted at EuroVis 2022
May 2022Started working at ArteriaAI as an NLP Data Scientist.
May 2022
Paper accepted at ACL 2022
March 2022Paper accepted at ACL 2022
March 2022Won Best Paper Paper award at ChartQA@CVPR 2021
June 2021
Paper accepted at ChartQA@CVPR 2021
June 2021Joined as a Graduate Teaching Student at York University
January 2021Joined as a Graduate Research Assistant at York University
September 2020Started Fully Funded M.Sc. in Computer Science at York University
September 2020Started as a Data Scientist Intern at Shallow AI
July 2020