Selected work

Representative Papers

Scientific Discovery under Validation Congestion via Multi-Fidelity Pairwise Rankings

Oct 01, 2026

This study addresses the verification bottleneck in scientific discovery arising from the disparity between an overabundance of candidate designs and scarce experimental resources. To overcome this challenge, we propose PRISMS, a framework that replaces data-dependent regression models with multi-fidelity expert pairwise ranking, thereby eliminating reliance on absolute score prediction. Furthermore, it dynamically upgrades query fidelity using a Fisher information criterion and integrates active learning strategies to efficiently screen high-potential designs. In drug screening tasks, PRISMS significantly improves recall rates while reducing the number of required experimental rounds, achieving an approximate 18.8% improvement in task hypervolume over baseline methods.

0 citationsRead paper

Rolling intrinsic for battery valuation in day-ahead and intraday markets

Oct 02, 2025

This study addresses the profit optimization problem for battery energy storage systems (BESS) participating in Central European wholesale electricity markets—specifically EPEX SPOT’s day-ahead auction and continuous intraday market. Methodologically, we propose a multi-market coordinated bidding framework based on a rolling endogenous value model that explicitly incorporates bid–ask spreads to capture market liquidity constraints. Crucially, we relax the rigid daily cycling limit while retaining annual total charge/discharge volume constraints, enabling inter-temporal value capture. Empirical results show that this strategy increases revenue by 18–32% over single-market approaches, confirming that cycling constraint relaxation unlocks additional arbitrage opportunities. Our key contributions are: (i) the first extension of the rolling endogenous value approach to coupled multi-market bidding; and (ii) a dynamic cycling management mechanism that jointly optimizes economic returns and battery lifetime. The framework provides a scalable decision-making paradigm for BESS operation in complex, multi-session electricity markets.

0 citationsRead paper

Towards a fast and robust deep hedging approach

Apr 23, 2025

Pricing and hedging option portfolios under complex stochastic volatility models (e.g., Heston, SABR) suffer from low computational efficiency, poor robustness, and slow adaptation to changing market conditions. Method: We propose a deep hedging framework based on low-dimensional model embedding. First, a neural network learns a universal hedge policy representation across diverse model paths. Second, an embedding-driven fast recalibration mechanism enables efficient parameter updates without full network retraining. Third, we systematically characterize the PnL distribution over the model parameter space—enabling model-agnostic hedge construction. Results: Experiments demonstrate substantial improvements in training efficiency and cross-model/market-state generalization. The framework achieves robust PnL performance on Monte Carlo–generated complex synthetic data, providing a scalable simulation foundation and practical paradigm for model-agnostic hedging.

0 citationsRead paper

Deep Hedging of Green PPAs in Electricity Markets

Mar 17, 2025

Green PPAs face three interrelated challenges in evolving electricity markets: exposure to volatile electricity prices, non-tradable weather-related risks, and market incompleteness exacerbated by the “cannibalization effect.” This paper introduces the first systematic application of deep hedging to Green PPA risk management, proposing a dynamic hedging framework that integrates deep reinforcement learning with stochastic control to end-to-end optimize risk-sensitive hedging policies. The model jointly leverages historical electricity prices, multi-source meteorological data, and market supply-demand features to train a differentiable hedging policy network. Evaluated under multiple risk metrics—including CVaR, MSE, and maximum drawdown—the approach significantly outperforms conventional static and dynamic hedging benchmarks: it reduces hedging costs by 18–32%, compresses risk exposure by over 40%, and overcomes the fundamental hedging bottleneck imposed by non-tradable weather factors.

0 citationsRead paper

Recent publications

Latest Papers

Scientific Discovery under Validation Congestion via Multi-Fidelity Pairwise Rankings

Oct 01, 2026

This study addresses the verification bottleneck in scientific discovery arising from the disparity between an overabundance of candidate designs and scarce experimental resources. To overcome this challenge, we propose PRISMS, a framework that replaces data-dependent regression models with multi-fidelity expert pairwise ranking, thereby eliminating reliance on absolute score prediction. Furthermore, it dynamically upgrades query fidelity using a Fisher information criterion and integrates active learning strategies to efficiently screen high-potential designs. In drug screening tasks, PRISMS significantly improves recall rates while reducing the number of required experimental rounds, achieving an approximate 18.8% improvement in task hypervolume over baseline methods.

0 citationsRead paper

Rolling intrinsic for battery valuation in day-ahead and intraday markets

Oct 02, 2025

This study addresses the profit optimization problem for battery energy storage systems (BESS) participating in Central European wholesale electricity markets—specifically EPEX SPOT’s day-ahead auction and continuous intraday market. Methodologically, we propose a multi-market coordinated bidding framework based on a rolling endogenous value model that explicitly incorporates bid–ask spreads to capture market liquidity constraints. Crucially, we relax the rigid daily cycling limit while retaining annual total charge/discharge volume constraints, enabling inter-temporal value capture. Empirical results show that this strategy increases revenue by 18–32% over single-market approaches, confirming that cycling constraint relaxation unlocks additional arbitrage opportunities. Our key contributions are: (i) the first extension of the rolling endogenous value approach to coupled multi-market bidding; and (ii) a dynamic cycling management mechanism that jointly optimizes economic returns and battery lifetime. The framework provides a scalable decision-making paradigm for BESS operation in complex, multi-session electricity markets.

0 citationsRead paper

Towards a fast and robust deep hedging approach

Apr 23, 2025

Pricing and hedging option portfolios under complex stochastic volatility models (e.g., Heston, SABR) suffer from low computational efficiency, poor robustness, and slow adaptation to changing market conditions. Method: We propose a deep hedging framework based on low-dimensional model embedding. First, a neural network learns a universal hedge policy representation across diverse model paths. Second, an embedding-driven fast recalibration mechanism enables efficient parameter updates without full network retraining. Third, we systematically characterize the PnL distribution over the model parameter space—enabling model-agnostic hedge construction. Results: Experiments demonstrate substantial improvements in training efficiency and cross-model/market-state generalization. The framework achieves robust PnL performance on Monte Carlo–generated complex synthetic data, providing a scalable simulation foundation and practical paradigm for model-agnostic hedging.

0 citationsRead paper

Deep Hedging of Green PPAs in Electricity Markets

Mar 17, 2025

Green PPAs face three interrelated challenges in evolving electricity markets: exposure to volatile electricity prices, non-tradable weather-related risks, and market incompleteness exacerbated by the “cannibalization effect.” This paper introduces the first systematic application of deep hedging to Green PPA risk management, proposing a dynamic hedging framework that integrates deep reinforcement learning with stochastic control to end-to-end optimize risk-sensitive hedging policies. The model jointly leverages historical electricity prices, multi-source meteorological data, and market supply-demand features to train a differentiable hedging policy network. Evaluated under multiple risk metrics—including CVaR, MSE, and maximum drawdown—the approach significantly outperforms conventional static and dynamic hedging benchmarks: it reduces hedging costs by 18–32%, compresses risk exposure by over 40%, and overcomes the fundamental hedging bottleneck imposed by non-tradable weather factors.

0 citationsRead paper