Hi, I'm Donghyeon Joo

Researching the Intersection of LLM Sparsity and Compute Efficiency

I am a fourth year Ph.D. Student in University of Maryland, College Park, advised by Professor Bahar Asgari in Computer Architecture and Systems Lab (CASL).

Research Focus

• • ML Level: Deriving sparsity in weights and KV cache to preserve model accuracy at high sparsity
• • System Level: GPU kernel support for sparse LLMs on existing compute platforms
• • Architecture Level: Novel architectural changes to better support sparsity and improve efficiency
• • Efforts in LLM reasoning, dynamically reconfigurable architecture, and RAG

⭐ I am a huge Star Wars fan, on a personal mission to name my papers with memorable planet names from the prequel/original trilogy ⭐

🔥 MUSTAFAR: Promoting Unstructured Sparsity for KV Cache Pruning in LLM Inference

Authors: Donghyeon Joo, Helya Hosseini, Ramyad Hadidi, Bahar Asgari

NeurIPS 2025

PaperCodePosterSlidesRecording

"Where Obi-wan had the high ground"

🌃 CORUSCANT: Co-Designing GPU Kernel and Sparse Tensor Core to Advocate Unstructured Sparsity in Efficient LLM Inference

Authors: Donghyeon Joo, Helya Hosseini, Ramyad Hadidi, Bahar Asgari

MICRO 2025

PaperCodePosterSlidesRecording

"Where Palpatine WAS the senate"

PIPIRIMA: Predicting Patterns in Sparsity to Accelerate Matrix Algebra

Authors: Ubaid Bakhtiar, Donghyeon Joo, Bahar Asgari

DAC 2025

SEGIN: Synergistically Enabling Fine-Grained Multi-Tenant and Resource Optimized SpMV

Authors: Helya Hosseini, Ubaid Bakhtiar, Donghyeon Joo, Bahar Asgari

CAL 2025

🖥️ Characterizing LLM Kernel Access and Memory Interaction in Multi-Partition NUMA GPUs

Authors: Donghyeon Joo, Sooraj Puthoor, Nuwan Jayasena, Bahar Asgari

Paper

🪴 ENDOR: Hardware-Friendly Sparse Format for Offloaded LLM Inference

Authors: Donghyeon Joo, Ramyad Hadidi, Soheil Feizi, Bahar Asgari

Paper

"Where Ewoks were really cute"