Conference Schedule

Friday, October 2, 2026 

Glass Pavilion at Levering Hall

08:40 - 09:00 Registration & Breakfast

09:00 - 09:10 Welcome Remarks

09:10 - 10:00 Xunyu Zhou

10:00 - 10:30 Qi Feng

10:30 - 11:00 Coffee Break

11:00 - 11:30 Pietro Bini

11:30 - 12:15 Guanting Chen

12:15 - 12:30 Group Photo

12:30 - 14:00 Lunch Break (participants on their own)

14:00 - 14:30 Anran Hu

14:30 - 15:00 Valentin Tissot-Daguette

15:00 - 15:30 Coffee Break 

15:30 - 16:20 Ronnie Sircar

16:30 - 18:30 Poster Presentation & Networking

Saturday, October 3, 2026 

Glass Pavilion at Levering Hall

08:40 - 09:10 Registration & Breakfast

09:10 - 10:00 Erhan Bayraktar

10:00 - 10:30 Luhao Zhang

10:30 - 11:00 Coffee Break

11:00 - 11:30 Soroosh Shafiee

11:30 - 12:00 Yu An

12:00 - 14:00 Lunch Break (participants on their own)

14:00 - 14:30 Gökçe Dayanıklı

14:30 - 15:00 Nathan Sauldubois

15:00 - 15:30 Melih Iseri

15:30 - 16:00 Coffee Break 

16:00 - 16:50 Nizar Touzi

16:50 - 17:00 Closing Remarks

18:30 Conference Dinner

Poster Session on Friday, October 2, 2026 (Glass Pavilion)

The Generalized Volterra Signature - Munawar Ali (Florada State University)

Abstract: In this paper, we introduce the generalized Volterra signature of a path, extending the classical Volterra signature introduced in [Hager, Harang, Pelizzari & Tindel, 2026], which extends Lyons' path signature to the Volterra setting. We establish a new algebraic structure to encode the generalized Volterra signature. We also establish a generalized Chen's identity and the shuffle product of signature components for the geometric Volterra signature. The shuffle product for the Volterra signature is the first of its kind and allows us to formulate a universal approximation theorem for geometric Volterra signatures. We also provide efficient computational schemes for the generalized Volterra signature. This framework is useful for non-Markovian and memory-dependent problems. We show its effectiveness using both synthetic and real financial data.

An Event-Driven Limit-Order-Book Model for Binary Prediction Markets - Buyak Aydin (Princeton University)

Abstract: We develop an event-driven limit-order-book model for a binary prediction market. Sequentially arriving traders receive private signals and submit market orders, while a continuum of risk-averse liquidity providers with heterogeneous beliefs supplies the book. After observing each trade, liquidity providers update their beliefs and reconstruct the book through a static nonatomic aggregative equilibrium. The book is represented by a pair of finite measures on the continuous price interval, allowing both continuous depth and price atoms. The model links informed order flow, adverse selection, belief heterogeneity, and equilibrium book shape. We study when order size reveals private information, how signal precision and risk aversion affect liquidity and spreads, and whether successive trades cause prices and book depth to converge toward the full-information event probability.

A BSDE Approach to Continuous-Time Reinforcement Learning - Filippo Beretta (ETH Zürich)

Abstract: Entropy regularization has been widely adopted to improve the efficiency, stability, and convergence of reinforcement learning (RL) algorithms. The exploratory, or entropy-regularized, formulation of continuous-time and continuous-space RL has recently attracted considerable attention, motivated by settings in which agents interact with the environment at ultra-high frequency, such as high-frequency trading and autonomous driving. In this paper, we recast the exploratory formulation within a general stochastic control framework over Polish action spaces, thereby unifying classical and exploratory control problems in a single setting. Shannon entropy regularization, commonly used in exploratory RL, then arises as a special case, and we systematically investigate the use of alternative entropy functions. Our approach is based on BSDEs, which allows us to handle non-Markovian dynamics and coefficients with low regularity. Within this framework, we establish global convergence rates for a policy iteration algorithm when the control enters only the drift, and we show that the convergence is linear. In the entropy-regularized setting, we further analyze the convergence of value functions and policies as the temperature parameter tends to zero. We provide the first convergence rates for this zero-temperature limit, explicitly showing how the rate depends on the chosen entropy regularization.

SDE-Synth: Stochastic Synthetic Data Generation for Pretraining Time-Series Foundation Models - Xiaotong Hu (Johns Hopkins University)

Abstract: Time-series foundation models (TSFMs) depend on pretraining corpora that expose the temporal mechanisms they must recognize, yet observational data provide limited control over rare stochastic events, nonstationarity, and cross-series dependence. We introduce SDE-Synth, a hierarchical stochastic-differential-equation-based generator for synthetic univariate and multivariate time series. It simulates configurable latent stochastic dynamics for univariate series and, for multivariate series, shared and variable-specific latent components with controllable cross-series dependence, then maps them to observed data through flexible observation and noise layers. Motivated by financial stylized facts, SDE-Synth targets structural coverage rather than market calibration. Across two TSFM architectures, it improves zero-shot forecasting across all FinStressTS mechanism families under synthetic-only pretraining and consistently improves held-out financial tasks when combined with real data, while GIFT-Eval transfer remains domain-dependent. These results support stochastic synthetic data as a targeted complement to real pretraining corpora.

Finite-State Mixed-Individual Mean Field Games With an Application to Epidemic Modeling - Christa Jia (University of Illinois Urbana-Champaign)

Abstract:  In recent years, finite-state mean field games (MFGs) have been used to study the mitigation of epidemics. However, existing models assume agents are entirely noncooperative and find a Nash equilibrium. This might fail to capture possible altruistic tendencies of the individuals. We first prove characterization of general form finite-state mean field control (MFC) problems with forward-backward ordinary differential equations (FBODEs) based on calculus of variations. Then, we propose a finite-state mixed-individual mean field game (MI-MFG) to model agents with both noncooperative and cooperative tendencies with an application to epidemic spread. We prove the FBODE characterization of the mixed-individual mean field Nash equilibrium (MI-MFNE) and its existence and uniqueness. We finalize by proposing an algorithm to find the MI-MFNE and conducting numerical experiments to investigate how individual altruism affects epidemic spread under different mitigation policies.

On Epidemic Mean Field Game Models with Social-Distancing and Vaccination - Huaning Liu (University of Illinois Urbana-Champaign)

Abstract: We present a finite-state epidemic mean field game (MFG) model and study two of its extensions to model epidemic spread with social-distancing and vaccination decisions of individuals. In the model, non-cooperative individuals choose their socialization and vaccination levels jointly as controls, and each individual's health state follows a continuous-time Markov chain of SIR type. We characterize the mean-field Nash equilibrium (MFNE) by a forward-backward ordinary differential equation (FBODE) system and establish its well-posedness. We calibrate the model parameters to real-world survey data and present numerical results. The first extension considers a more realistic linear vaccination cost, under which the equilibrium vaccination level becomes bang-bang with at-most one-jump. The second extension introduces a principal with budget dynamics, who chooses the social-distancing guidelines and the vaccination level for the population. This yields a Stackelberg MFG problem. We characterize the Stackelberg mean-field equilibrium by another FBODE system using optimal control techniques, and establish the corresponding well-posedness results.

AI for AI: Optimizing Additional Infrastructure to Power Artificial Intelligence Data Centers - Kyle Onghai (Princeton University)

Abstract: The twenty-first century’s transformative technology, artificial intelligence, is increasingly constrained by the twentieth century’s transformative technology, the electricity grid. Rapid growth in electricity demand from data centers is creating substantial uncertainty about the pace of generation infrastructure expansion and the resulting effects on wholesale electricity prices. To grasp this situation, we develop a framework linking

data-center load growth, available generation capacity, and market-clearing prices. We first analyze a deterministic model to show how differing estimates of demand and supply growth rates affect prices. We then model the buildout of new data centers and their associated electricity demand, together with additions to electricity supply, as stochastic processes, resulting in probabilistic distributions of supply, demand, and prices, rather than a single forecast. Finally, we formulate generation expansion as a stochastic control problem in which a revenue-maximizing investor dynamically chooses the intensity of supply-side investments. We derive the associated Hamilton–Jacobi–Bellman equation, characterize the optimal investment policy, and solve the model numerically. The framework allows investors to compare competing supply technologies—including natural gas, solar, wind, and nuclear—whose attractiveness depends greatly on their capital and operating costs and discrete scale of projects. The results highlight a central challenge of the data-center buildout: even when rapid demand growth increases the need for new generation, the uncertainties related to load forecasts, development execution risks, and value cannibalization from overbuilding capacity may weaken incentives to invest at the pace required to keep electricity prices stable.

Portfolio Choice with Competing Precautionary and Accumulation Goals - Ananya Parashar (Columbia University)

Abstract: We study optimal portfolio choice for a household managing two goals at once. A random-deadline precautionary goal, such as a medical emergency, must be paid in full whenever it arrives and is affordable; a fixed-deadline accumulation goal, such as a target retirement lifestyle, may be declined at its deadline. We show that precautionary saving crowds out growth-oriented investment, a coupling absent from either goal alone, and deadline pressure is amplified by the competing precautionary claim. We characterize the value function as the unique viscosity solution of an HJB equation, and derive the household's optimal terminal funding rule. The household funds the fixed-deadline goal exactly when the direct benefit at least offsets the resulting loss in random-deadline continuation value. In the reported numerical experiments, both flexibility and fungibility are most valuable at intermediate wealth, where committing in advance to fund forfeits the first and managing each goal in its own account forfeits the second. In a stylized two-goal example, separate accounts require roughly $6,000 more present-value initial wealth to match a pooled household holding $33,000.

Continuous Policy and Value Iteration for Classical Control - Renjie Shao (Florida State University)

Abstract: We introduce a continuous policy-value iteration algorithm for infinite-horizon classical stochastic control problems, updating value function and optimal control approximations simultaneously via continuous-time deterministic dynamics. We establish policy improvement and convergence to the optimal control under the Hamiltonian's monotonicity condition. To demonstrate its efficacy, we apply this framework to an incomplete market model where an agent manages an investment portfolio and continuous consumption to maximize discounted power utility. Leveraging the strict concavity of the Hamiltonian, our algorithm efficiently optimizes the value function and identifies the optimal strategy.

A Lifted-MDP Framework for Inverse Learning in LQ Mixed-Individual Mean-Field Games - Leyan Su (Johns Hopkins University)

Abstract: We formulate a lifted Markov decision process (MDP) for a linear–quadratic (LQ) mixed-individual mean-field game in which a mixing parameter λ governs the balance between individual and collective objectives. The lifted control problem is defined on the space of state distributions and admits a dynamic programming principle (DPP) and the associated Hamilton–Jacobi–Bellman (HJB) equation. Within this framework, we study the inverse problem of recovering λ from observed equilibrium behavior and establish its identifiability under an explicit observation regime and a nondegeneracy condition on the population mean. We then propose a model-based inverse reinforcement learning (RL) algorithm that uses an analytic LQ best-response solver for the inner forward problem and estimates λ through an objective combining an expert optimality-gap term with mean-field consistency. Using simulated expert datasets, we evaluate the method and quantify its parameter-recovery accuracy and cross-dataset variability. The consistency-regularized configurations recover the ground-truth parameter accurately across datasets, whereas removing the consistency term leads to larger recovery error and greater variability. As additional datasets are included in the cumulative evaluation, the empirical recovery mean and dataset-level dispersion settle into a stable regime. These results demonstrate accurate and stable recovery of the mixing parameter in the LQ setting and support the effectiveness of the proposed inverse-learning framework.

α-Potential Games for Decentralized Control of Heterogeneous Connected and Automated Vehicles - Zhexin Wang (Columbia University)

Abstract: Designing scalable control strategies for connected and automated vehicles (CAVs) requires accounting for strategic interactions among heterogeneous agents under decentralized information. We develop an α-potential game framework for finite-player CAV control with decentralized closed-loop policies, where minimizing a suitably constructed potential yields an approximate Nash equilibrium. To improve the equilibrium approximation, we introduce player-specific rescaling and optimize over the resulting family of potential functions. We establish existence and efficient computation of an optimal rescaling, and show that exact Nash equilibria can be recovered under the Kolmogorov cyclic condition. We further derive social-efficiency guarantees and identify a tradeoff between equilibrium approximation and social efficiency.  Numerical experiments distinguish weak mean-field and strong local-interaction regimes, demonstrate heterogeneous vehicle interactions and collision and obstacle avoidance, and illustrate diverse lane-changing and overtaking behaviors under different initial traffic configurations.  We further show that road-priority structures lead to distinct intersection-crossing equilibria, and that the proposed rescaling can substantially improve the approximation to a Nash equilibrium in asymmetric games.

Heavy-Tailed Flow Matching via Random Clocks - Zhouhao Yang (Johns Hopkins University)

Abstract: Heavy-tailed data arise in many domains where rare events carry disproportionate importance, such as imbalanced image datasets, financial returns, and weather extremes. Standard diffusion and flow-matching models typically begin from Gaussian noise or Gaussian source distributions, which yield tractable training targets but provide a poor inductive match for heavy-tailed data. We propose Heavy-Tailed Flow Matching via Random Clocks (HTFM), a framework that portrays heavy-tailed sources as mixtures of clock-conditioned Gaussian sources. Conditioning on a given clock path, the source distribution and flow are Gaussian; marginalizing over the clock gives a Gaussian scale mixture covering Gaussian, α-stable, and Student-t families. To make the clock-conditioned vector field practical, we encode the path-valued clock using truncated logsignature features, allowing the velocity field to adapt to the realized conditional space with negligible overhead. Empirically, on 2D imbalanced  α-stable mixtures, CIFAR10-LT, and HRRR weather fields, HTFM improves mode coverage, sample quality, and tail-statistic recovery over Gaussian flow matching and competitive heavy-tailed baselines, while retaining the low-NFE sampling advantage of flow matching. Moreover, the random-clock formulation further provides a practical tail-control interface: by varying only the clock law or tail parameter, the same architecture can calibrate the "heaviness" of generated tails across different distribution families.

Kullback-Leibler Mirror-Prox for Measure-Valued Variational Inequalities and Mean-Field Equilibria - Ziqing Zhang (University of Michigan)

Abstract: We study the computation of static mean-field equilibria on a compact state space by formulating the equilibrium condition as a variational inequality over probability measures. We propose an entropic variant of Korpelevich's extragradient algorithm---the Kullback--Leibler Mirror-Prox method---in which Euclidean projections are replaced by relative-entropy proximal steps. Each half-step is an exponential reweighting of the measure, implemented on a finite state-space discretization. Under Lasry--Lions monotonicity and continuity assumptions, we prove convergence of mesh-refined ergodic averages and obtain finite-iteration Minty-residual and approximate-equilibrium bounds that jointly quantify iteration and discretization errors. Under strong monotonicity, we derive metric convergence rates for the last, best, and averaged iterates. We also develop a KL-type Tikhonov regularization that selects the equilibrium minimizing relative entropy with respect to a reference measure. The framework applies to potential and nonpotential cost operators and does not require differentiability or convexity of the cost in the individual state.

The Convergence Problem for Ergodic Mean Field Games - Kaiwen Zhang (Princeton University)

Abstract: The convergence of Nash equilibria of large N-player games to mean field game equilibria has been extensively studied over the past decades, particularly for finite horizon games. For infinite horizon ergodic games, however, this convergence problem remains less well understood and is known to fail in general. We address this question for open-loop equilibria in the displacement monotone setting and derive explicit convergence rates. Specifically, we consider ergodic games with drift and volatility control, as well as common noise, without requiring time-homogeneous coefficients. Our approach is based on a version of the Pontryagin maximum principle adapted for ergodic control problems and the resulting infinite horizon McKean-Vlasov Forward Backward Stochastic Differential Equations (FBSDEs), for which we prove well-posedness. Solutions to these FBSDEs are shown to correspond to particular equilibria of feedback nature, for which we prove quantitative convergence using uniform in time propagation of chaos estimates. In contrast to the finite horizon setting, the particular equilibria found using FBSDEs are not the unique ones, and we analyze when the convergence result can be extended for general equilibria.

Nickel Squeeze: Modelling Extreme Commodity Futures Price Distortions - Iosif Zimbidis (Princeton University)

Abstract: The March 2022 nickel squeeze on the London Metal Exchange demonstrates how a concentrated short position held by traders who cannot deliver can lead to extreme distortions in futures prices. As nickel prices rose, large short positions faced increasing margin calls. Covering these positions required buying futures, which pushed prices even higher in an increasingly illiquid market. We apply our structural framework, which decomposes the observed futures price into a fundamental price and a correction generated by traders without delivery capability. In contrast to the April 2020 WTI episode, where forced liquidation created downward price pressure, the nickel squeeze involved forced short covering and therefore an upward price distortion. The model captures how market composition, position constraints, and endogenous feedback can drive futures prices away from fundamentals, providing a unified mechanism for extreme price distortions in both directions.