arXiv subjects

Tianzhe Zheng

Publications and source records attributed to Tianzhe Zheng.

3 recordsLinked to original sources

arXiv2026

ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B, PhAI-IDE-9B, and PhAI-IDE-4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity's scientific software a shared substrate for developing scientific intelligence. Code: https://github.com/aitofound/ScienceIDE

cs.CL↗

arXiv2022

3D-Patterned Inverse-Designed Mid-Infrared Metaoptics

Modern imaging systems can be enhanced in efficiency, compactness, and application through introduction of multilayer nanopatterned structures for manipulation of light based on its fundamental properties. High transmission efficiency multispectral imaging is surprisingly elusive due to the commonplace use of filter arrays which discard most of the incident light. Further, most cameras do not leverage the wealth of information in polarization and spatial degrees of freedom. Optical metamaterials can respond to these electromagnetic properties but have been explored primarily in single-layer geometries, limiting their performance and multifunctional capacity. Here we use advanced two-photon lithography to realize multilayer scattering structures that achieve highly nontrivial optical transformations intended to process light just before it reaches a focal plane array. Computationally optimized multispectral and polarimetric sorting devices are fabricated with submicron feature sizes and experimentally validated in the mid-infrared. A final structure shown in simulation redirects light based on its angular momentum. These devices demonstrate that with precise 3-dimensional nanopatterning, one can directly modify the scattering properties of a sensor array to create advanced imaging systems.

physics.optics↗

arXiv2021

Constraining continuous topology optimizations to discrete solutions for photonic applications

Photonic topology optimization is a technique used to find the electric permittivity distribution of a device that optimizes an electromagnetic figure-of-merit. Two common techniques are used: continuous density-based optimizations that optimize a grey-scale permittivity defined over a grid, and discrete level-set optimizations that optimize the shape of the material boundary of a device. More recently, continuous optimizations have been used to find an initial seed for a concluding level-set optimization since level-set techniques tend to benefit from a well-performing initial structure. However, continuous optimizations are not guaranteed to yield sufficient initial seeds for subsequent level-set optimizations, particularly for high-contrast structures, since they are not guaranteed to converge to solutions that resemble only two discrete materials. In this work, we present a method for constraining a continuous optimization such that it converges to a discrete solution. This is done by inserting a constrained sub-optimization at each iteration of an overall gradient-based optimization. This technique can be used purely on its own to optimize a device, or it can be used to provide a nearly discrete starting point for a level-set optimization.

physics.app-ph↗