DENNIS (XIMING) DONG
Toronto
Summary
- MSc graduate Machine Learning Engineer providing 2 years of professional experience and 5 years of research expertise in areas including but not limited to Natural Language Processing (NLP), Deep Learning, Model Interpretation, and Compression.
- Demonstrates excellent proficiency in Data Engineering and Data Pre-processing, capable of gathering and
processing large-scale datasets for training and evaluating NLP models. - Expert in Large Language Models (LLM) and experienced in experimental design.
- Known for being hard-working and solving complicated problems.
Overview
4
4
years of professional experience
Work History
Machine Learning Researcher
Huawei Canada
04.2023 - Current
- Owned the design and development of the LLMSafeGuard framework to enhance content safety in language models
- This resulted in a 29.7% reduction in toxic content for GPT-2 and a 56.2% reduction in copyright risks for Llama-2
- Optimized LLMSafeGuard content generating process by skipping potentially unnecessary safety checks, reducing the number of execution cycles. This optimization improved processing efficiency by 24%. A paper detailing this work was submitted to ICSE 2025 and will be published next year
- Developed an innovative algorithm using Captum's libraries to quantify the influence of input prompts on model outputs, hosted and executed efficiently on platform for model interpretation management
- Customized language model outputs, incorporating dynamic text generation termination and exclusion features
- Developed and implemented a feature using soft prompts trained via PEFT, optimized for seamless deployment on the platform
- Automated the construction of structured and customized prompts for language models by designing and implementing a comprehensive project
- Architected a Flask API for managing a Qdrant database with integrated Sentence-Transformers and RAG design, enhancing data retrieval accuracy and response times by about 40%
- Utilized Jenkins on a Huawei cloud server for robust project management and debugging, improving project delivery timelines by about 30% while maintaining high-quality standards.
Deep learning Researcher
University of Manitoba
09.2020 - 12.2022
- Developed a Knowledge Distillation (KD) utilizing BERT and ALBERT models to train compact 'Student Models' like CNN and LSTM, achieving a model 50 times smaller and faster than full-size BERT with a 5% improvement in accuracy, precision, recall, and F1-score, and retaining approximately 97% of the Teacher model's performance
- Enhanced text similarity studies by incorporating BERT instead of Glove for contextual language modeling, resulting in a 3.7% increase in accuracy over traditional LSTM/CNN models, with research findings presented at conferences and considered for academic publication.
Machine Learning Engineer
PING AN Technology
05.2022 - 10.2022
- Implemented unsupervised learning with RoBERTa on a vast unlabelled text corpus, optimizing the model with Early Stopping and Ray Tune
- This strategic fine-tuning on select labeled data achieved a 15.1% improvement in sentiment analysis accuracy
- Introduced and applied 'Prompt Learning' from Open Prompt to seamlessly transition from pre-training to fine-tuning of language models, resulting in a 3.5% increase in document classification accuracy
- Streamlined the deployment of advanced machine learning models by converting them to ONNX format for integration with Ping An's cloud platform
- This conversion enhanced online inference capabilities and boosted processing efficiency by about 20%
Education
Master of Science - Machine Learning & Deep Learning
University of Manitoba
Winnipeg, MB01.2023
Bachelor of Science - Computer Science (Hornors)
University of Manitoba
Winnipeg, MB12.2019
Skills
- Programming Language: Python, SQL, Java, C
- Framework: PyTorch, Transformers, Qdrant, TensorFlow, scikit-learn, Ray, PEFT, Megatron-LM, Dspy, Langchain, Accelerate
- Data Visualization: Pandas, Numpy, Matplotlib
- Other: Hugging Face, Linux, Slurm, Docker, Git, Latex
Publications
- Ximing Dong, Olive Huang, Parimala Thulasiraman, Aniket Mahanti, :"Improved Knowledge Distillation
via Teacher Assistants for Sentiment Analysis" in lEEE Symposium Series on Computational 1ntelligence
(SSCI),2023. (https://ieeexplore.ieee.org/document/10371965) - Ximing Dong, Dayi Lin, Shaowei Wang, Ahmed E. Hassan, :"A Framework for Real-time Safeguarding
the Text Generation of Large Language Model" (https://arxiv.org/abs/2404.19048)
Timeline
Machine Learning Researcher
Huawei Canada
04.2023 - Current
Machine Learning Engineer
PING AN Technology
05.2022 - 10.2022
Deep learning Researcher
University of Manitoba
09.2020 - 12.2022
Master of Science - Machine Learning & Deep Learning
University of Manitoba
Bachelor of Science - Computer Science (Hornors)
University of Manitoba
Similar Profiles
Wallace LiraWallace Lira
Graphics Engineer at Huawei CanadaGraphics Engineer at Huawei Canada
Abdulmuizz YusufAbdulmuizz Yusuf
Prompt Engineer at Huawei Canada Noah's Ark LabPrompt Engineer at Huawei Canada Noah's Ark Lab
Lingyao WangLingyao Wang
R&D Material Management Specialist at Huawei Technologies Canada Co., Ltd.R&D Material Management Specialist at Huawei Technologies Canada Co., Ltd.