Institution profile

Basis

Industry research
Research library13linked papers
Opportunities0open roles
Selected work

Representative Papers

Simulation-Free Learning of Population Dynamics with Wasserstein Lagrangian Residuals

Oct 02, 2026

This study addresses the inability of Wasserstein gradient flows to capture conservative dynamics and the prohibitive training costs of existing Lagrangian methods that rely on numerical simulations. To overcome these limitations, this work proposes Double-Stitch, a simulation-free framework. By leveraging the Clebsch variational principle, the method derives velocity-gradient-free equations of motion and achieves efficient learning of Lagrangian mechanics in Wasserstein space by penalizing equation residuals along learned trajectories. Theoretically, vanishing residuals guarantee that the equations of motion are satisfied, thereby entirely eliminating numerical solving steps during training. Experiments demonstrate that Double-Stitch matches or surpasses baseline performance on synthetic, single-cell, and ocean eddy datasets while accelerating training by 4 to 14 times.

0 citationsRead paper

Slaying the Hydra: Interaction-Aware Circuit Discovery in Language Models

Oct 02, 2026

This study addresses the imprecise behavioral localization in mechanistic interpretability of large language models caused by neglecting component interaction effects. To this end, it proposes the WISE causal estimator and JuntaLearner, a gradient-based circuit discovery method. The core innovation lies in pioneering set-effect estimation integrated with a witness mechanism, which effectively circumvents combinatorial enumeration explosion, alongside the introduction of the CRS metric for comprehensively evaluating the critical roles of small circuits. By synergizing causal inference, witness variables, and gradient optimization techniques, the proposed approach achieves scalable, high-precision circuit discovery. Experimental results demonstrate that JuntaLearner significantly outperforms baselines in average CRS across diverse tasks and varying model scales.

0 citationsRead paper

EMPIRIC: Experiment-Driven Learning of Residual World Models for Robot Planning

Sep 28, 2026

This study addresses the challenge of robots interacting with objects governed by mechanisms unmodeled in standard physics engines. To this end, it proposes an interpretable and reusable residual world model that leverages program synthesis to generate code extending the physics engine, thereby compensating for missing dynamics. By integrating Bayesian inference with active experimental design, the framework achieves data-efficient mechanism learning and motion planning. In simulation, it efficiently completes complex tasks, while on a real robot, it successfully infers latent physical parameters such as wind disturbances and domino masses. These capabilities enable precise prediction and manipulation control, demonstrating the framework’s effectiveness in bridging the gap between simulated and real-world physical interactions.

0 citationsRead paper

Wasserstein Residuals: Learning Gradient Flows from Population Dynamics

Jul 06, 2026

This work addresses the reconstruction of population dynamics governed by Wasserstein gradient flows from sparse observational data. The authors propose a particle-based "stitching" method that bypasses the traditional Jordan–Kinderlehrer–Otto (JKO) time discretization, instead enforcing the continuity equation via a non-negative residual loss and integrating a data-fidelity divergence into a unified optimization objective. By eliminating the need for costly optimal transport computations and fixed time steps, the approach is simulation-agnostic and robust to irregular or widely spaced observation intervals. Evaluated on multiple trajectory inference benchmarks, the method achieves state-of-the-art performance, demonstrating particular superiority in regimes with highly sparse observations or large temporal gaps between measurements.

0 citationsRead paper
Recent publications

Latest Papers

Simulation-Free Learning of Population Dynamics with Wasserstein Lagrangian Residuals

Oct 02, 2026

This study addresses the inability of Wasserstein gradient flows to capture conservative dynamics and the prohibitive training costs of existing Lagrangian methods that rely on numerical simulations. To overcome these limitations, this work proposes Double-Stitch, a simulation-free framework. By leveraging the Clebsch variational principle, the method derives velocity-gradient-free equations of motion and achieves efficient learning of Lagrangian mechanics in Wasserstein space by penalizing equation residuals along learned trajectories. Theoretically, vanishing residuals guarantee that the equations of motion are satisfied, thereby entirely eliminating numerical solving steps during training. Experiments demonstrate that Double-Stitch matches or surpasses baseline performance on synthetic, single-cell, and ocean eddy datasets while accelerating training by 4 to 14 times.

0 citationsRead paper

Slaying the Hydra: Interaction-Aware Circuit Discovery in Language Models

Oct 02, 2026

This study addresses the imprecise behavioral localization in mechanistic interpretability of large language models caused by neglecting component interaction effects. To this end, it proposes the WISE causal estimator and JuntaLearner, a gradient-based circuit discovery method. The core innovation lies in pioneering set-effect estimation integrated with a witness mechanism, which effectively circumvents combinatorial enumeration explosion, alongside the introduction of the CRS metric for comprehensively evaluating the critical roles of small circuits. By synergizing causal inference, witness variables, and gradient optimization techniques, the proposed approach achieves scalable, high-precision circuit discovery. Experimental results demonstrate that JuntaLearner significantly outperforms baselines in average CRS across diverse tasks and varying model scales.

0 citationsRead paper

EMPIRIC: Experiment-Driven Learning of Residual World Models for Robot Planning

Sep 28, 2026

This study addresses the challenge of robots interacting with objects governed by mechanisms unmodeled in standard physics engines. To this end, it proposes an interpretable and reusable residual world model that leverages program synthesis to generate code extending the physics engine, thereby compensating for missing dynamics. By integrating Bayesian inference with active experimental design, the framework achieves data-efficient mechanism learning and motion planning. In simulation, it efficiently completes complex tasks, while on a real robot, it successfully infers latent physical parameters such as wind disturbances and domino masses. These capabilities enable precise prediction and manipulation control, demonstrating the framework’s effectiveness in bridging the gap between simulated and real-world physical interactions.

0 citationsRead paper

Wasserstein Residuals: Learning Gradient Flows from Population Dynamics

Jul 06, 2026

This work addresses the reconstruction of population dynamics governed by Wasserstein gradient flows from sparse observational data. The authors propose a particle-based "stitching" method that bypasses the traditional Jordan–Kinderlehrer–Otto (JKO) time discretization, instead enforcing the continuity equation via a non-negative residual loss and integrating a data-fidelity divergence into a unified optimization objective. By eliminating the need for costly optimal transport computations and fixed time steps, the approach is simulation-agnostic and robust to irregular or widely spaced observation intervals. Evaluated on multiple trajectory inference benchmarks, the method achieves state-of-the-art performance, demonstrating particular superiority in regimes with highly sparse observations or large temporal gaps between measurements.

0 citationsRead paper