Selected work

Representative Papers

Algorithm-driven Development: A Proactive Approach to Improving Software Quality and Reducing Defects

Aug 02, 2026

This work addresses the challenge of balancing software quality, testability, and maintainability under rapid iteration and frequent requirement changes. It proposes Algorithm-Driven Development (ADD), a novel approach that unifies requirements specification and technical design by using algorithm flowcharts as a single, coherent artifact. This integration enables end-to-end modeling of requirements, architecture, and testing. Leveraging this model, the system automatically generates high-coverage acceptance tests and incorporates continuous integration with code coverage feedback. Industrial adoption at Dassault Systèmes demonstrates that ADD achieves over 95% code coverage, substantially reduces defect density, and ensures a stable delivery cadence, outperforming conventional test-driven development and test-after approaches.

0 citationsRead paper

PGD-NO: A Neural Operator with Precomputed Geometry Decomposition for 3D Million-scale Physics Simulations

Jul 08, 2026

This work addresses the limited scalability of existing neural PDE solvers, which are hindered by high memory consumption and single-node bottlenecks that prevent application to meshes with tens of millions of nodes. The authors propose PGD-NO, the first neural operator architecture achieving linear memory scalability. By employing an iterative geometric decomposition algorithm to pre-extract geometric tokens, the method shifts geometric encoding to a deterministic preprocessing stage, thereby decoupling feature extraction from solution querying and circumventing single-GPU memory constraints. Integrated with an attention mechanism, PGD-NO enables efficient large-scale 3D physical field modeling while offering intrinsic interpretability. The approach attains state-of-the-art accuracy on multiple industrial benchmarks and successfully trains models exceeding ten million nodes, substantially surpassing the scale limits of current methods.

0 citationsRead paper

SceneTeract: Agentic Functional Affordances and VLM Grounding in 3D Scenes

Mar 31, 2026

This work addresses the challenge of automatically evaluating the functional usability of 3D scenes for embodied agents—i.e., whether a scene enables agents to perform meaningful interactive tasks—by introducing SceneTeract, a novel framework that decomposes high-level semantic activities into atomic action sequences. Integrating agent specifications with geometric and physical simulation, SceneTeract systematically verifies constraints such as reachability, spatial clearance, and navigability. As the first method dedicated to automated functional usability assessment for embodied agents, SceneTeract uncovers a systematic bias in current vision-language models (VLMs) between semantic plausibility and physical feasibility. Leveraging this insight, the framework employs the identified discrepancies as reward signals in reinforcement learning-based post-training distillation to enhance VLMs’ geometric reasoning capabilities. Experiments on synthetic indoor environments reveal numerous functional defects, effectively exposing and mitigating prevalent VLM misjudgments regarding scene usability.

0 citationsRead paper

Towards Solving Polynomial-Objective Integer Programming with Hypergraph Neural Networks

Mar 16, 2026

This work addresses polynomial integer programming problems involving high-order variable interactions—a class of optimization challenges significantly more difficult than linear integer programs due to their nonlinear structure. The paper introduces the first solution framework based on hypergraph neural networks, which constructs a high-order-term-aware hypergraph representation to uniformly capture complex dependencies among variables, high-order terms, and constraints. A dual-path hypergraph convolution mechanism is designed to separately aggregate variable–high-order-term and variable–constraint information for predicting high-quality initial solutions, subsequently refined via heuristic search. Experiments demonstrate that the proposed method substantially outperforms existing learning-based approaches and commercial solvers across multiple benchmarks, achieving notable advances in both solution quality and computational efficiency, and is applicable to general polynomial integer programming scenarios.

0 citationsRead paper

A Wachspress-based transfinite formulation for exactly enforcing Dirichlet boundary conditions on convex polygonal domains in physics-informed neural networks

Jan 05, 2026arXiv.org

This work addresses the challenge of accurately enforcing Dirichlet boundary conditions in physics-informed neural networks (PINNs) over convex polygonal domains. The authors propose a transfinite interpolation method based on Wachspress coordinates to exactly extend boundary functions into the interior, thereby constructing trial neural network functions that strictly satisfy the prescribed boundary conditions and possess bounded Laplacians. This approach represents the first integration of Wachspress coordinates with transfinite interpolation, overcoming limitations inherent in conventional approximate distance functions and offering applicability to arbitrary convex polygonal geometries. Numerical experiments on forward, inverse, and parametric Poisson boundary value problems demonstrate that the method achieves high accuracy, exhibits strong generalization capabilities, and enables a unified solution framework for parameterized convex domains.

0 citationsRead paper

Recent publications

Latest Papers

Algorithm-driven Development: A Proactive Approach to Improving Software Quality and Reducing Defects

Aug 02, 2026

This work addresses the challenge of balancing software quality, testability, and maintainability under rapid iteration and frequent requirement changes. It proposes Algorithm-Driven Development (ADD), a novel approach that unifies requirements specification and technical design by using algorithm flowcharts as a single, coherent artifact. This integration enables end-to-end modeling of requirements, architecture, and testing. Leveraging this model, the system automatically generates high-coverage acceptance tests and incorporates continuous integration with code coverage feedback. Industrial adoption at Dassault Systèmes demonstrates that ADD achieves over 95% code coverage, substantially reduces defect density, and ensures a stable delivery cadence, outperforming conventional test-driven development and test-after approaches.

0 citationsRead paper

PGD-NO: A Neural Operator with Precomputed Geometry Decomposition for 3D Million-scale Physics Simulations

Jul 08, 2026

This work addresses the limited scalability of existing neural PDE solvers, which are hindered by high memory consumption and single-node bottlenecks that prevent application to meshes with tens of millions of nodes. The authors propose PGD-NO, the first neural operator architecture achieving linear memory scalability. By employing an iterative geometric decomposition algorithm to pre-extract geometric tokens, the method shifts geometric encoding to a deterministic preprocessing stage, thereby decoupling feature extraction from solution querying and circumventing single-GPU memory constraints. Integrated with an attention mechanism, PGD-NO enables efficient large-scale 3D physical field modeling while offering intrinsic interpretability. The approach attains state-of-the-art accuracy on multiple industrial benchmarks and successfully trains models exceeding ten million nodes, substantially surpassing the scale limits of current methods.

0 citationsRead paper

SceneTeract: Agentic Functional Affordances and VLM Grounding in 3D Scenes

Mar 31, 2026

This work addresses the challenge of automatically evaluating the functional usability of 3D scenes for embodied agents—i.e., whether a scene enables agents to perform meaningful interactive tasks—by introducing SceneTeract, a novel framework that decomposes high-level semantic activities into atomic action sequences. Integrating agent specifications with geometric and physical simulation, SceneTeract systematically verifies constraints such as reachability, spatial clearance, and navigability. As the first method dedicated to automated functional usability assessment for embodied agents, SceneTeract uncovers a systematic bias in current vision-language models (VLMs) between semantic plausibility and physical feasibility. Leveraging this insight, the framework employs the identified discrepancies as reward signals in reinforcement learning-based post-training distillation to enhance VLMs’ geometric reasoning capabilities. Experiments on synthetic indoor environments reveal numerous functional defects, effectively exposing and mitigating prevalent VLM misjudgments regarding scene usability.

0 citationsRead paper

Towards Solving Polynomial-Objective Integer Programming with Hypergraph Neural Networks

Mar 16, 2026

This work addresses polynomial integer programming problems involving high-order variable interactions—a class of optimization challenges significantly more difficult than linear integer programs due to their nonlinear structure. The paper introduces the first solution framework based on hypergraph neural networks, which constructs a high-order-term-aware hypergraph representation to uniformly capture complex dependencies among variables, high-order terms, and constraints. A dual-path hypergraph convolution mechanism is designed to separately aggregate variable–high-order-term and variable–constraint information for predicting high-quality initial solutions, subsequently refined via heuristic search. Experiments demonstrate that the proposed method substantially outperforms existing learning-based approaches and commercial solvers across multiple benchmarks, achieving notable advances in both solution quality and computational efficiency, and is applicable to general polynomial integer programming scenarios.

0 citationsRead paper

A Wachspress-based transfinite formulation for exactly enforcing Dirichlet boundary conditions on convex polygonal domains in physics-informed neural networks

Jan 05, 2026arXiv.org

This work addresses the challenge of accurately enforcing Dirichlet boundary conditions in physics-informed neural networks (PINNs) over convex polygonal domains. The authors propose a transfinite interpolation method based on Wachspress coordinates to exactly extend boundary functions into the interior, thereby constructing trial neural network functions that strictly satisfy the prescribed boundary conditions and possess bounded Laplacians. This approach represents the first integration of Wachspress coordinates with transfinite interpolation, overcoming limitations inherent in conventional approximate distance functions and offering applicability to arbitrary convex polygonal geometries. Numerical experiments on forward, inverse, and parametric Poisson boundary value problems demonstrate that the method achieves high accuracy, exhibits strong generalization capabilities, and enables a unified solution framework for parameterized convex domains.

0 citationsRead paper