Selected work

Representative Papers

ChemActor: Enhancing Automated Extraction of Chemical Synthesis Actions with LLM-Generated Data

Jun 30, 2025

Automated extraction of chemical synthesis steps is hindered by textual ambiguity in scientific literature and the scarcity of high-quality annotated data. To address this, we propose ChemActor—a novel foundation model that achieves end-to-end, precise parsing of unstructured experimental text into machine-executable operation sequences. Methodologically, ChemActor integrates distribution-aware data filtering, a large-language-model-based multi-round iterative review mechanism, and a two-stage task learning paradigm (reaction → description → action), coupled with full-parameter fine-tuning and explicit modeling of machine-executable actions. On the Reaction-to-Description (R2D) and Description-to-Action (D2A) benchmarks—two core tasks for synthesis procedure understanding—ChemActor outperforms prior state-of-the-art methods by 10% absolute gain, substantially improving operational identification accuracy and procedural executability. This establishes ChemActor as a robust foundation model for automating organic synthesis workflows.

0 citationsRead paper

MGE-LDM: Joint Latent Diffusion for Simultaneous Music Generation and Source Extraction

May 29, 2025

This work introduces MGE-LDM—the first unified latent diffusion framework addressing the fragmentation among music generation, source completion, and query-driven source separation. Methodologically, it reformulates all three tasks as conditional inpainting in the latent space, enabled by multi-condition text guidance, cross-dataset heterogeneous alignment, and joint modeling of mixtures, submixes, and isolated sources—achieving fully instrument-agnostic, end-to-end training. Trained jointly on Slakh2100, MUSDB18, and MoisesDB, MGE-LDM supports zero-shot, text-driven separation of arbitrary instruments, controllable mixture generation, and missing-source completion. Experiments demonstrate substantial improvements in audio fidelity and functional consistency over staged baselines. To our knowledge, this is the first framework realizing, within a single model and architecture, instrument-agnostic integration of these three core music signal processing tasks.

0 citationsRead paper

Recent publications

Latest Papers

ChemActor: Enhancing Automated Extraction of Chemical Synthesis Actions with LLM-Generated Data

Jun 30, 2025

Automated extraction of chemical synthesis steps is hindered by textual ambiguity in scientific literature and the scarcity of high-quality annotated data. To address this, we propose ChemActor—a novel foundation model that achieves end-to-end, precise parsing of unstructured experimental text into machine-executable operation sequences. Methodologically, ChemActor integrates distribution-aware data filtering, a large-language-model-based multi-round iterative review mechanism, and a two-stage task learning paradigm (reaction → description → action), coupled with full-parameter fine-tuning and explicit modeling of machine-executable actions. On the Reaction-to-Description (R2D) and Description-to-Action (D2A) benchmarks—two core tasks for synthesis procedure understanding—ChemActor outperforms prior state-of-the-art methods by 10% absolute gain, substantially improving operational identification accuracy and procedural executability. This establishes ChemActor as a robust foundation model for automating organic synthesis workflows.

0 citationsRead paper

MGE-LDM: Joint Latent Diffusion for Simultaneous Music Generation and Source Extraction

May 29, 2025

This work introduces MGE-LDM—the first unified latent diffusion framework addressing the fragmentation among music generation, source completion, and query-driven source separation. Methodologically, it reformulates all three tasks as conditional inpainting in the latent space, enabled by multi-condition text guidance, cross-dataset heterogeneous alignment, and joint modeling of mixtures, submixes, and isolated sources—achieving fully instrument-agnostic, end-to-end training. Trained jointly on Slakh2100, MUSDB18, and MoisesDB, MGE-LDM supports zero-shot, text-driven separation of arbitrary instruments, controllable mixture generation, and missing-source completion. Experiments demonstrate substantial improvements in audio fidelity and functional consistency over staged baselines. To our knowledge, this is the first framework realizing, within a single model and architecture, instrument-agnostic integration of these three core music signal processing tasks.

0 citationsRead paper