Chuck Yoo

Research profile

Research summary

Chuck Yoo is a systems researcher at Korea University whose OpenAlex-indexed record spans 1999-2026 with no gap years, and whose recent output is heavily weighted toward machine-learning systems: the field split in the current snapshot is exactly even between distributed LLM training and computer systems performance evaluation. Topic clusters — cloud computing and resource management (86), network traffic and congestion control (61), parallel computing and optimization (54), software-defined networks and 5G (52), and data storage technologies (26) — trace a career that moved from networking and cloud infrastructure into LLM systems. Three 2026 arXiv papers in the Evolve catalog, each with a middle-author role, illustrate the current agenda: a characterization of LLM-agent resource and performance dynamics; CONA, a parallelism-strategy-chaining method that reaches target perplexity 1.4-9.6x faster; and PARSER, an output-importance-guided residual-sparsification scheme for compressing mixture-of-experts LLMs, which Evolve rates highly reproducible. The snapshot reports an h-index of 16 with an i10-index of 42; citation-based impact scoring is unavailable in this data. Collaboration is moderate — 15 distinct coauthors in the last three years, with twelve recurring partners — and the radar marks activity (98) as the strongest dimension, with the 2026 count of 15 works the highest on record. This is a name-resolved profile built from public bibliographic data and Evolve's paper reviews.

Research keywords

Highlights

• CONA chains parallelism strategies online for 1.4-9.6x faster time-to-perplexity

  In this 2026 arXiv study, CONA switches parallelism strategies during training, guided by gradient noise scale and Bayesian optimization, reaching target perplexity 1.4-9.6x faster. Evolve finds the results consistent but small-scale — single-run experiments on models at or below 1.3B parameters.

• PARSER compresses mixture-of-experts LLMs via output-importance-guided residual sparsification

  PARSER selects residual-pruning targets by output importance, narrowing accuracy gaps against state-of-the-art methods at equal memory. Evolve rates its reproducibility readiness high, while noting that the gains skew toward MMLU benchmarks.

• Characterizing LLM-agent resource dynamics shows bottlenecks are task-dependent

  This 2026 study profiles the resource and performance behavior of AI agents in tool containers, finding concurrency bottlenecks tied to task type and CPU allocation, and demonstrates admission control and allocation policies. Evolve calls it a careful study but notes it rests on one host, one LLM, and replayed trajectories.

Timeline

1. 1999Earliest year in the OpenAlex counts-by-year record, with 1 indexed work — the start of a 27-year publication span.
2. 2005Output reaches double digits for the first time, with 11 indexed works.
3. 2013Most productive year in the record before 2026, with 14 indexed works.
4. 20257 indexed works, before output jumps to a record 15 the following year.
5. 2026Three arXiv papers enter the Evolve catalog — LLM-agent resource characterization, CONA parallelism strategy chaining, and PARSER MoE compression — while counts-by-year reaches a record 15 works.

Papers in the Evolve catalog

• 2026-09-18

  Not All AI Agents Are Equal: Characterizing Resource and Performance Dynamics

• 2026-09-07

  Parallelism Strategy Chaining for Fast Training Convergence

• 2026-09-01

  Residual Sparsification via Output Importance for Compressing Mixture-of-Experts LLMs

Frequent co-authors