Agent workloads scale with tasks, users, and repeated tool calls.
Research Statement
Agentic Small Language Model
Empowering small language models through distillation, tool integration, and context optimization.
sLM
+
Tools
Agent
Minki Kang Ph.D. Student, KAIST AI
Small models are the deployment path for everyday agents.
My industry experience repeatedly pointed to the same constraint: useful agents must often run close to users and products.
Private memories and user data should not always leave controlled environments.
Games, medical AI, and copilots need responsive agents under real deployment limits.
Why are small models weak?
Small models are bottlenecked by parametric memory.
They are not simply unable to reason. Many reasoning tasks secretly require closed-book memory: facts, calculations, intermediate checks, and verification signals.
KARD shows that external knowledge reduces what a small model must memorize internally.
T1 shows that tools turn memorization-heavy verification into tool-use and interpretation.
The goal is to convert memory-heavy problems into agentic problems.
Small models need tools more, but use tools worse.
Tool need sLM > LLM
Small models need external memory, computation, and verification to overcome limited capacity.
Agentic capability sLM < LLM
Without training, they are worse at retrieving, computing, checking, and recovering.
My research trains small models to use agentic scaffolding reliably enough that external memory and computation compensate for limited parametric capacity.
Retrieve missing knowledge
Compute and verify exact steps
Distill reliable agentic behavior
Research Program
Building blocks for agentic small language models.
KARD: Knowledge-Augmented Reasoning
Retrieve knowledge for rationale generation.
NeurIPS 2023T1: Tool-integrated Self-verification
Use tools for calculation and fact-checking.
ICLR 2026Distilling LLM Agent into Small Models
Distill retrieval, code, observation, and recovery traces.
NeurIPS 2025 SpotlightACON: Optimizing Context Compression
Compress context for long-horizon productive work.
ICML 2026Future Vision
Agentic scaffolding can turn small models into self-improving agents.
Use tools, memory, verification, and planning reliably.
Design data, objectives, evaluations, and training recipes.
Turn research feedback into better agentic behavior.
Takeaway: once an SLM masters agentic scaffolding, it can conduct research on model training and use that feedback to improve its own capabilities.
1 / 6