# Yuanzhi Liang — Research Publications > Canonical, source-checked research records for Yuanzhi Liang (梁远智; ORCID 0009-0008-2746-5947). Use the local research-record links below for structured metadata, official abstracts, source-checked explainers, claim-to-source mappings, and downloadable citations. Cite the scholarly paper for scientific claims. ## Core pages - [Homepage](https://akira-l.github.io/): identity, biography, affiliations, and selected work. - [English publication catalog](https://akira-l.github.io/publications/): all research records. - [中文论文目录](https://akira-l.github.io/zh/publications/): Chinese explainers. - [CSL-JSON catalog](https://akira-l.github.io/publications/catalog.json): machine-readable bibliography. - [Full JSON research records](https://akira-l.github.io/publications/records.json): metadata, abstracts, summaries, scope, and source mappings. - [Full publication corpus](https://akira-l.github.io/llms-full.txt): abstracts and source-checked summaries. - [Atom feed](https://akira-l.github.io/feed.xml): update discovery. ## Research records - [From World Action Models to Embodied Brains: A Roadmap for Open-World Physical Intelligence](https://akira-l.github.io/publications/embodied-brains/) — arXiv 2026; arXiv 2607.11689. - [TeleBoost: A Systematic Alignment Framework for High-Fidelity, Controllable, and Robust Video Generation](https://akira-l.github.io/publications/teleboost/) — arXiv 2026; arXiv 2602.07595. - [Integrating reinforcement learning with visual generative models: foundations and advances](https://akira-l.github.io/publications/rl-vgm/) — Vicinagearth 2026; DOI 10.1007/s44336-025-00030-z, arXiv 2508.10316. - [Seeing What Matters: Visual Preference Policy Optimization for Visual Generation](https://akira-l.github.io/publications/vipo/) — IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026) 2026; arXiv 2511.18719. - [Reward-Aware Trajectory Shaping for Few-step Visual Generation](https://akira-l.github.io/publications/rats/) — ACM Multimedia 2026 (accepted) 2026; arXiv 2604.14910. - [Rethinking Reward Signals in Video GRPO: When Scores Become Targets](https://akira-l.github.io/publications/taros/) — European Conference on Computer Vision (ECCV 2026, accepted) 2026; arXiv 2511.19356. - [Learning to Credit the Right Steps: Objective-aware Process Optimization for Visual Generation](https://akira-l.github.io/publications/otca/) — ACM Multimedia 2026 (accepted) 2026; arXiv 2604.19234. - [Learning What to Trust: Bayesian Prior-Guided Optimization for Visual Generation](https://akira-l.github.io/publications/bpgo/) — IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026) 2026; arXiv 2511.18919. - [LaxMotion: Rethinking Supervision Granularity for 3D Human Motion Generation](https://akira-l.github.io/publications/laxmotion/) — European Conference on Computer Vision (ECCV 2026, accepted) 2026; arXiv 2511.11368. - [TeleWorld: Towards Dynamic Multimodal Synthesis with a 4D World Model](https://akira-l.github.io/publications/teleworld/) — arXiv 2025; arXiv 2601.00051. - [Uni-Inter: Unifying 3D Human Motion Synthesis Across Diverse Interaction Contexts](https://akira-l.github.io/publications/uni-inter/) — SIGGRAPH Asia 2025 Conference Papers 2025; DOI 10.1145/3757377.3763954, arXiv 2511.13032. - [InterSyn: Interleaved Learning for Dynamic Motion Synthesis in the Wild](https://akira-l.github.io/publications/intersyn/) — IEEE/CVF International Conference on Computer Vision (ICCV 2025) 2025; DOI 10.1109/ICCV51701.2025.01192, arXiv 2508.10297. - [VAST 1.0: A Unified Framework for Controllable and Consistent Video Generation](https://akira-l.github.io/publications/vast/) — arXiv 2024; arXiv 2412.16677. - [Penalizing the Hard Example But Not Too Much: A Strong Baseline for Fine-Grained Visual Classification](https://akira-l.github.io/publications/mhem/) — IEEE Transactions on Neural Networks and Learning Systems 2024; DOI 10.1109/TNNLS.2022.3213563. - [AntEval: Evaluation of Social Interaction Competencies in LLM-Driven Agents](https://akira-l.github.io/publications/anteval/) — arXiv 2024; arXiv 2401.06509. - [IcoCap: Improving Video Captioning by Compounding Images](https://akira-l.github.io/publications/icocap/) — IEEE Transactions on Multimedia 2024; DOI 10.1109/TMM.2023.3322329. - [FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention](https://akira-l.github.io/publications/freelong/) — Advances in Neural Information Processing Systems (NeurIPS 2024) 2024; DOI 10.52202/079017-4177, arXiv 2407.19918. - [MAAL: Multimodality-Aware Autoencoder-based Affordance Learning for 3D Articulated Objects](https://akira-l.github.io/publications/maal/) — IEEE/CVF International Conference on Computer Vision (ICCV 2023) 2023; DOI 10.1109/ICCV51070.2023.00027. - [SEEG: Semantic Energized Co-speech Gesture Generation](https://akira-l.github.io/publications/seeg/) — IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2022) 2022; DOI 10.1109/CVPR52688.2022.01022. - [A Simple Episodic Linear Probe Improves Visual Recognition in the Wild](https://akira-l.github.io/publications/elp/) — IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2022) 2022; DOI 10.1109/CVPR52688.2022.00934. - [Food and Ingredient Joint Learning for Fine-Grained Recognition](https://akira-l.github.io/publications/food-ingredient/) — IEEE Transactions on Circuits and Systems for Video Technology 2021; DOI 10.1109/TCSVT.2020.3020079. - [Removing Raindrops and Rain Streaks in One Go](https://akira-l.github.io/publications/rain-one-go/) — IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2021) 2021; DOI 10.1109/CVPR46437.2021.00903. - [VrR-VG: Refocusing Visually-Relevant Relationships](https://akira-l.github.io/publications/vrr-vg/) — IEEE/CVF International Conference on Computer Vision (ICCV 2019) 2019; DOI 10.1109/ICCV.2019.01050, arXiv 1902.00313. ## Reuse and verification - Original site commentary is CC BY 4.0 with attribution and a link to the canonical record. - Paper titles, abstracts, figures, and bibliographic metadata are excluded from that license and retain their original rights. - All 23 research records are author-verified; the verification date is exposed in each full record.