Jiayi (Joyee) Wang

Research

Publications.

My research spans multi-agent LLMs, LLMs for healthcare, reinforcement learning, machine learning theory, optimization algorithms, and energy systems.

* Author names marked with an asterisk are listed in alphabetical order.

2026 Under Review

Scaling Clinician‑Grade Feature Generation from Clinical Notes with Multi‑Agent Language Models

Jiayi Wang, Jacqueline Jil Vallon, Neil Panjwani, Xi Ling, Margaret Redfield, Sushmita Vij, Sandy Srinivas, John Leppert, Mohsen Bayati, Mark K. Buyyounouski

PDF → Code →
2026 ALT ★ Elegant Paper Award

Reusing Samples in Variance Reduction

Yujia Jin*, Ishani Karmarkar*, Aaron Sidford*, Jiayi Wang*

PDF →
2026 EC

Endogenous Entry in Networked Markets with Production and Edge Capacity Constraints

Cheng Guo, Jiayi Wang, Ozan Candogan

SSRN →

Equation 7 · convex Cournot equilibrium
\begin{aligned} \max_{\mathbf{y}}\;\; & \sum_{m \in \mathcal{M}}\!\Big[\sum_{i \in \mathcal{F}}\Big(A_m - \tfrac{1}{2}B_m\!\!\sum_{j \in \mathcal{F},\,j\neq i}\!\!q_{jm}\Big)q_{im} - \sum_{i \in \mathcal{F}}\Big(B_m q_{im}^2 + \!\!\sum_{\delta \in \lambda(i,m)}\!\!C_\delta^T x_\delta\Big)\Big] - \sum_{i \in \mathcal{F}} f_i^C\!\Big(\!\sum_{m \in \mathcal{M}}\!q_{im}\!\Big) \\ \text{s.t.}\;\; & q_{im} = \!\!\sum_{\delta \in \lambda(i,m)}\!\! x_\delta && (4a) \\ & \sum_{\delta \ni i} x_\delta \leq Q_i^{\text{firm}} && (4b) \\ & x_\delta \geq 0 && (4c) \\ & \underline{Q}_{kl}^{\text{edge}} \leq \!\!\sum_{\delta \in \rho(k,l)}\!\! x_\delta - \!\!\sum_{\delta \in \rho(l,k)}\!\! x_\delta \leq \overline{Q}_{kl}^{\text{edge}} && (4g\text{-}4h) \end{aligned}

2024 NeurIPS

Truncated Variance‑Reduced Value Iteration

Yujia Jin*, Ishani Karmarkar*, Aaron Sidford*, Jiayi Wang*

PDF →

Table 1 · runtimes for ε-optimal policies

Algorithm	Runtime
Value Iteration	\widetilde{O}\bigl(\text{nnz}(\mathbf{P})(1-\gamma)^{-1}\bigr)
Empirical QVI	\widetilde{O}\bigl(\text{nnz}(\mathbf{P}) + \mathcal{A}_{\text{tot}}(1-\gamma)^{-3}\varepsilon^{-2}\bigr)
Randomized Primal‑Dual Method	\widetilde{O}\bigl(\text{nnz}(\mathbf{P}) + E\,\mathcal{A}_{\text{tot}}(1-\gamma)^{-4}\varepsilon^{-2}\bigr)
High Precision Variance‑Reduced Value Iteration	\widetilde{O}\bigl(\text{nnz}(\mathbf{P}) + \mathcal{A}_{\text{tot}}(1-\gamma)^{-3}\bigr)
Algorithm 4 (This Paper)	\widetilde{O}\bigl(\text{nnz}(\mathbf{P}) + \mathcal{A}_{\text{tot}}(1-\gamma)^{-2}\bigr)

2024 IEEE TEMPR

Risk‑Aware Security‑Constrained Unit Commitment

Daniel Bienstock*, Yury Dvorkin*, Cheng Guo*, Robert Mieth*, Jiayi Wang*

PDF →

Equation 11 · risk-aware SCUC problem
\begin{aligned} \min\;\; & \sum_{t \in \mathcal{T}^{\text{DA}}} \Big(\sum_{g \in \mathcal{G}}(h_{gt} + C_g^{\text{Start}} v_{gt} + C_g^{\text{Down}} w_{gt}) + \sum_{i \in \mathcal{N}} C^{\text{VOLL}} p_{it}^{\text{Unmet}}\Big) + \rho\,\hat{V}(\mathbf{y}) \\ \text{s.t.}\;\; & h_{gt} \geq C_{og}^{1} p_{gt} + C_{og}^{0} y_{gt} && (4) \\ & \sum_{g \in \mathcal{G}_i} p_{gt} + p_{it}^{\text{Unmet}} + \!\!\sum_{(j,i)\in\mathcal{L}}\!\! f_{jit} - \!\!\sum_{(i,j)\in\mathcal{L}}\!\! f_{ijt} = \bar{D}_{it} && (5) \\ & f_{ijt} = B_{ij}(\theta_{it} - \theta_{jt}) && (6) \\ & v_{gt} - w_{gt} = y_{gt} - y_{g,t-1} && (7) \\ & p_{gt} \leq P_g^{\max} y_{gt} && (8) \\ & p_{gt} - p_{g,t-1} \leq M_g y_{g,t-1} + P_g^{\min} v_{gt} && (9) \\ & \hat{V}(\mathbf{y}) = \max_{\boldsymbol{\omega}\in\Omega}\frac{1}{N^{\text{tp}}}\sum_{t \in \mathcal{T}^{\text{RT}}}\sum_{i \in \mathcal{N}} \lambda_{it}(\mathbf{y},\boldsymbol{\omega})(d_{it}^{\text{RT}}-\bar{D}_{it})^{+} \end{aligned}

Let’s build something.

Publications.

Scaling Clinician‑Grade Feature Generation from Clinical Notes with Multi‑Agent Language Models

Reusing Samples in Variance Reduction

Endogenous Entry in Networked Markets with Production and Edge Capacity Constraints

Truncated Variance‑Reduced Value Iteration

Risk‑Aware Security‑Constrained Unit Commitment