Selected work

Representative Papers

ATLAS: Atomic Translation&Language for Automated Structures

Aug 27, 2026

Large language models struggle to directly generate atomic structures satisfying global physical constraints such as periodicity and bond lengths, as conventional token-by-token generation lacks physical plausibility. This work proposes ATLAS, a framework that leverages component algebra to translate natural language into verifiable structure-building scripts. By introducing a checklist-based verification scoring mechanism and a JSON-formatted construction engine, the framework enables deterministic replay, rendering the LLM translation process both measurable and controllable. ATLAS successfully demonstrates the generation of fourteen structural classes spanning crystalline to amorphous networks, and can be directly applied to constructing training datasets for machine learning force fields.

1 citationsRead paper

Fully 3GPP-Compatible Long-Range Sensing for LEO-ISAC: A Window-Grid Processing Framework

Sep 24, 2026

This study addresses the OFDM sensing symbol misalignment and signal duration mismatch caused by long propagation delays in low Earth orbit integrated sensing and communication (LEO-ISAC) systems. To overcome these challenges, this work proposes a receiver-side windowed grid processing framework. By exploiting the geometric determinism of targets, the method performs purely receive-side signal processing, ensuring full compatibility with 3GPP standards without requiring any modifications to the transmitted waveform. Experimental results demonstrate that the proposed scheme achieves meter-level ranging accuracy at bistatic distances exceeding 640 kilometers. These findings indicate that the framework provides an efficient and practical solution for standardized long-range bistatic radar sensing within LEO-ISAC networks.

0 citationsRead paper

Multimodal deep learning from spectra for small-molecule structure identification: enhancing robustness with mixed-condition training

Sep 13, 2026

研究通过混合条件训练和多专家融合方法，提高小分子结构识别的鲁棒性，解决光谱缺失、降解或不匹配的问题。

0 citationsRead paper

URNet: A Unified Reparameterized Network for Efficient RGB-D Semantic Segmentation

Aug 06, 2026

This work addresses the limitations of existing RGB-D semantic segmentation methods, which typically employ dual-encoder architectures that suffer from inadequate depth representation, restricted cross-modal interaction, and computational redundancy. To overcome these issues, we propose URNet, a unified framework that processes both RGB and depth inputs through a single encoder to enable efficient multi-scale feature fusion. The core innovations include integrating reparameterized blocks (RepBlocks) with Linear Gated Attention (LGA) modules, allowing simultaneous feature extraction and cross-modal interaction within a unified architecture, as well as designing a lightweight, general-purpose Pyramid Merging Decoder (PMD). Extensive experiments demonstrate that URNet achieves state-of-the-art performance across multiple benchmarks while significantly improving inference efficiency.

0 citationsRead paper

Guideline-as-Oracle: Zero-Annotation Training of an Ophthalmic Telephone Triage Agent

Aug 05, 2026

This work addresses the scarcity of supervised signals in multi-turn ophthalmic telephone triage, caused by high expert annotation costs and clinical privacy constraints. To overcome this challenge, the authors propose the Guideline-as-Oracle (GAO) framework, which translates the American Academy of Ophthalmology guidelines into 70 operational rules to serve as the sole instance-level supervision for 3,000 training dialogues—eliminating the need for manual annotation. By introducing eight novel rule-to-dialogue construction strategies, a rule-based zero-annotation dialogue generation method, a label repair mechanism, and fine-tuning a 9B-parameter language model, GAO-Triage achieves a significant improvement on a reference set of 201 cases: expert agreement rises from 61.7% to 74.1%, and recall for urgent cases increases dramatically from 9.5% to 69.0%. The system outperforms all general-purpose baselines without relying on advanced reasoning techniques.

0 citationsRead paper

Recent publications

Latest Papers

Fully 3GPP-Compatible Long-Range Sensing for LEO-ISAC: A Window-Grid Processing Framework

Sep 24, 2026

This study addresses the OFDM sensing symbol misalignment and signal duration mismatch caused by long propagation delays in low Earth orbit integrated sensing and communication (LEO-ISAC) systems. To overcome these challenges, this work proposes a receiver-side windowed grid processing framework. By exploiting the geometric determinism of targets, the method performs purely receive-side signal processing, ensuring full compatibility with 3GPP standards without requiring any modifications to the transmitted waveform. Experimental results demonstrate that the proposed scheme achieves meter-level ranging accuracy at bistatic distances exceeding 640 kilometers. These findings indicate that the framework provides an efficient and practical solution for standardized long-range bistatic radar sensing within LEO-ISAC networks.

0 citationsRead paper

Multimodal deep learning from spectra for small-molecule structure identification: enhancing robustness with mixed-condition training

Sep 13, 2026

研究通过混合条件训练和多专家融合方法，提高小分子结构识别的鲁棒性，解决光谱缺失、降解或不匹配的问题。

0 citationsRead paper

ATLAS: Atomic Translation&Language for Automated Structures

Aug 27, 2026

Large language models struggle to directly generate atomic structures satisfying global physical constraints such as periodicity and bond lengths, as conventional token-by-token generation lacks physical plausibility. This work proposes ATLAS, a framework that leverages component algebra to translate natural language into verifiable structure-building scripts. By introducing a checklist-based verification scoring mechanism and a JSON-formatted construction engine, the framework enables deterministic replay, rendering the LLM translation process both measurable and controllable. ATLAS successfully demonstrates the generation of fourteen structural classes spanning crystalline to amorphous networks, and can be directly applied to constructing training datasets for machine learning force fields.

1 citationsRead paper

URNet: A Unified Reparameterized Network for Efficient RGB-D Semantic Segmentation

Aug 06, 2026

This work addresses the limitations of existing RGB-D semantic segmentation methods, which typically employ dual-encoder architectures that suffer from inadequate depth representation, restricted cross-modal interaction, and computational redundancy. To overcome these issues, we propose URNet, a unified framework that processes both RGB and depth inputs through a single encoder to enable efficient multi-scale feature fusion. The core innovations include integrating reparameterized blocks (RepBlocks) with Linear Gated Attention (LGA) modules, allowing simultaneous feature extraction and cross-modal interaction within a unified architecture, as well as designing a lightweight, general-purpose Pyramid Merging Decoder (PMD). Extensive experiments demonstrate that URNet achieves state-of-the-art performance across multiple benchmarks while significantly improving inference efficiency.

0 citationsRead paper

Guideline-as-Oracle: Zero-Annotation Training of an Ophthalmic Telephone Triage Agent

Aug 05, 2026

This work addresses the scarcity of supervised signals in multi-turn ophthalmic telephone triage, caused by high expert annotation costs and clinical privacy constraints. To overcome this challenge, the authors propose the Guideline-as-Oracle (GAO) framework, which translates the American Academy of Ophthalmology guidelines into 70 operational rules to serve as the sole instance-level supervision for 3,000 training dialogues—eliminating the need for manual annotation. By introducing eight novel rule-to-dialogue construction strategies, a rule-based zero-annotation dialogue generation method, a label repair mechanism, and fine-tuning a 9B-parameter language model, GAO-Triage achieves a significant improvement on a reference set of 201 cases: expert agreement rises from 61.7% to 74.1%, and recall for urgent cases increases dramatically from 9.5% to 69.0%. The system outperforms all general-purpose baselines without relying on advanced reasoning techniques.

0 citationsRead paper