Soohyun Choi

PathBridger: Subgoal Bridges for Offline Goal-Conditioned Reinforcement Learning

Soohyun Choi, Seonvin Cho, and Songnam Hong

PathBridger plans across long horizons by selecting an intermediate subgoal, constructing a plausible bridge in state space, decoding the bridge into short executable action chunks with inverse dynamics, and replanning during execution. The method is evaluated on OGBench navigation and manipulation tasks.

Code arXiv

Multi-step Proximal Policy Improvement in Offline Reinforcement Learning

Soohyun Choi^*, Seonvin Cho^*, and Songnam Hong^†

^*Equal contribution · ^†Corresponding author

MPI interprets behavior-anchored offline actor updates as proximal policy improvement steps on a policy manifold and composes sequential re-centered refinements. It studies how geometry, step size, and critic error shape controlled advancement beyond dataset support.

Code arXiv

ALARM: Attention and Line Search Based Analog Circuit Optimization on a Riemannian Manifold

Soohyun Choi, Minseok Oh, and Ickhyun Song

ALARM treats circuits with the same designer-selected performance as an equivalence class and combines the induced quotient geometry with an attention-weighted coordinate lift and backtracking line search. In the archived active-inductor VCO study, it met the target in 10/10 runs using 33.2 SPICE evaluations on average.

Technical report Undergraduate capstone project

MART: Multi-Step Actor Refinement Trajectories in Offline Reinforcement Learning

MART studies how a fixed policy-improvement horizon can be traversed through multiple persistent local actors. By exposing only the first actor to Bellman targets and deploying the endpoint, it analyzes how refinement depth changes critic-facing bootstrap exposure, deployment reach, and offline training stability.

AMO: Adaptive Multiscale Policy Optimization in Offline Reinforcement Learning

AMO learns the total policy-improvement horizon T from an offline bilevel outer objective while treating N as a multiscale resolution parameter. Following MART's two-actor structure, it separates the critic-facing and actor scales: the critic uses the local T/N policy, while the actor represents the full-horizon T policy.

PathFlower: Goal-Conditioned State Flows

PathFlower develops the next direction from PathBridger: moving from an explicitly selected subgoal and bridge toward final-goal-conditioned state flows that generate useful intermediate states without fixing a subgoal in advance. This is an ongoing research direction in generative planning and long-horizon offline goal-conditioned RL.

REACT: Recursive Algorithm for RC Network Transfer Functions

Soohyun Choi, Joonghoon Lee, Sunggu Lim, Heatbit Park, Songnam Hong, and Jaeduk Han

First-author research conducted through an SK hynix–Hanyang University project. REACT develops a recursive closed-form transfer-function method for arbitrary RC trees and a dominant-pole reduced-order approximation validated against SPICE.