gmgu / graph_edit_distance

This project aims at exact graph edit distance (GED) computation and GED verification (verify whether the GED between two graphs is smaller than a given threshold), where all edit operators are assumed to have unit costs.

Research Interests

My primary research efforts have been devoted to developing fast algorithms. I have developed fast algorithms for graph isomorphim, graph isomorphism query processing, and multiple pattern Cartesian tree matching during my Ph.D. study. At LG Electronics, I have developed AI coding assistant using large language model (LLM).

Work Experience

LG Electronics - Artificial Intelligence Lab (Senior Researcher)

• Aug. 2022 – Present: Fine-tuning large language models for AI coding assistance using PyTorch FSDP and DeepSpeed (for full parameter fine-tuning), and LoRA (for parameter-efficient fine-tuning) on AWS SageMaker. Development of data augmentation techniques for causal language models. Inference server development for large language model using NVIDIA Triton, FastTransformer, and FastAPI. Inference optimization for transformer models using flash attention and paged attention.
• Apr. 2022 – Dec. 2022: Training a small language model (from scratch) for AI coding assistance using PyTorch. Web UI development using Python Streamlit. Inference server development for language models using Flask and ShannonAI/service-streamer.

Seoul National University – Institute of Computer Technology (Post-Doctoral Assistant)

• Jan. 2022 – Mar. 2022: Algorithm development for graph isomorphism query processing (Efficient Graph Isomorphism Query Processing using Degree Sequences and Color-Label Distributions, IEEE ICDE 2022).

Tech/Skills

Competitive Programming

Programming Languages

• C/C++, Python, CUDA C++, Rust, C#, Java, Shell Script, LaTeX

Libraries

• PyTorch, TensorFlow, Triton (OpenAI), Seaborn, Pandas, PySpark, HuggingFace Transformers, DeepSpeed, NVIDIA Triton, NVIDIA Faster Transformer, FastAPI, gtest

Others

• AWS (SageMaker, EC2, Lustre, S3)

CV

GeonmoGu_CV