Home

Geonwoo Shin

M.S. student, Department of Industrial Engineering, Seoul National University · SLCF Lab

I study how deep learning can be used inside classical portfolio-based asset allocation frameworks. I am interested in how estimation error is amplified when a view is turned into an actual weight, and in allocations that stay stable when the market regime shifts.

I recently presented work at APIEMS 2025 on combining Black–Litterman with views that an LLM forms by reflecting on its own past rebalancing decisions. I am currently working on applying vector quantization to latent factor models.

shin000621@gmail.com · GitHub · Google Scholar

Research

Market-Adaptive Memory LLM for the Black–Litterman Model

APIEMS 2025, oral · KIIE 2025 Spring Joint Conference, poster

The agent stores its reflections on past rebalancing results in a vector memory, retrieves the ones from market regimes similar to the current one, forms a view on each stock, and blends it with market equilibrium returns through Black–Litterman. Retrieving only the similar-regime reflections worked best; adding the most recent ones on top made performance worse.

Preserving low-variance pricing information in vector-quantized factor models (in progress)

Compressing returns into a small set of representative vectors can discard directions that carry little variance but still matter for pricing. I use per-subspace codebooks to keep them.

Publications

  • W. Jeong, G. Shin, J. Lee. JoCE: Joint Counterfactual Explanations for Interpretable Time Series Anomaly Detection. Pattern Recognition, 177, 2026.
  • D. Kim*, G. Shin*, Y. Choi, S. Park, J. Lee. A Locally Tokenized Generative Model for Robust Time-Series Watermarking. Submitted to NeurIPS 2026. arXiv
  • D. Kim, S. Lee, G. Shin, J. Lee. Discretizing Continuous Time Series for Imputation with Masked Diffusion Training. Submitted to NeurIPS 2026. arXiv

* Equal contribution

Experience

  • TA, Korea Banking Institute data science program capstone — the team I advised won an excellence award
  • TA, SNU KDT Big Data · FinTech · AI program (machine learning and deep learning, three cohorts)
  • TA, Samsung DS² program (linear algebra and optimization)
  • Encouragement Award, NH Investment & Securities Big Data Competition — LLM-based ETF curation
  • WorldQuant IQC 2024 — 1st on campus, advanced to Stage 2
  • Vice president, UFEA financial engineering society — equity and interest rate derivatives
  • Best Paper Award, KIPA Autumn Conference — automatic IPC classification of patent documents
  • Excellence Award, Korea University Capstone Design

Education

  • M.S. in Industrial Engineering, Seoul National University (expected Feb 2027)
  • B.S. in Industrial and Management Engineering, Korea University