Selected work

Representative Papers

Retrieving and Refining Winning Noise Tickets for Diffusion-Based Motion Generation

Jul 07, 2026

Existing diffusion-based text-to-motion generation models often suffer from semantic drift in long-horizon and compositional motions, struggling to balance semantic fidelity with motion coherence. This work proposes WINRO, a novel framework that reveals—for the first time—the decisive role of initial noise in determining the semantics of generated motions. WINRO introduces a training-free, model-agnostic mechanism for noise retrieval and refinement: by retrieving the most text-aligned “winning noise ticket” and applying KL-regularized optimization—optionally enhanced with a single-step forward LoRA adapter—it substantially improves text-motion alignment. Experiments demonstrate that WINRO effectively enhances semantic consistency for both MDM and MotionLCM on HumanML3D, boosts temporal robustness on the MTT benchmark, and generalizes successfully to motion stylization and spatially constrained motion synthesis tasks.

0 citationsRead paper

ARMS: Anchor-Relational Motion Streaming for Seamless Solo-Social Motion Transitions

Jul 06, 2026

This work addresses the challenge of generating long-term, socially coherent human motion that naturally transitions between solitary and interactive states. The authors propose a unified causal generative framework that seamlessly integrates individual motion and interpersonal interaction within a single motion stream, enabling incremental long-horizon generation conditioned solely on historical context. Key innovations include a dynamically asymmetric representation to decouple individual temporal evolution from interpersonal alignment, a mode-aware relational gating mechanism for smooth transitions between generation modes, and a diffusion model operating in a causal latent space augmented with partner-referenced relative translation representations. Experiments demonstrate that the method outperforms existing interaction-centric baselines in terms of smoothness during solitary-to-social transitions and social coherence, while achieving competitive performance on standard human-human interaction benchmarks.

0 citationsRead paper

InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation

Jul 02, 2026

Existing methods for generating two-person interactive motions struggle to simultaneously preserve long-range temporal causality of individual actions and mutual coordination between agents: bidirectional denoising disrupts temporal causality, while autoregressive strategies suffer from coordination degradation due to temporal drift. This work proposes a block-causal latent diffusion framework that employs dual-stream causal diffusion Transformers to model each agent’s motion separately and introduces a unified multi-task attention mask to jointly capture diverse interaction patterns—including synchronization, response, leader-follower dynamics, and independence. By adopting block-level diffusion targets, the method avoids redundant encoding-decoding and enables mask-controlled interaction-type specification during inference. It achieves state-of-the-art performance on the InterHuman and Inter-X datasets, significantly improving text-motion alignment, motion realism, and long-term temporal coherence.

0 citationsRead paper

Escaping Iterative Parameter-Space Noise: Differentially Private Learning with a Hypernetwork

Jun 25, 2026

This work addresses the significant utility degradation in differentially private neural network training—such as DP-SGD—caused by repeated noise injection in high-dimensional parameter spaces. The authors propose a novel framework that leverages a hypernetwork pretrained on public data to map private data into target model parameters, injecting privacy-preserving noise only once in a low-dimensional data embedding space, thereby avoiding direct perturbation of high-dimensional parameters. By integrating embedding aggregation, a tailored perturbation mechanism, and LoRA fine-tuning, the method substantially improves model performance under a fixed privacy budget. Empirical results demonstrate clear advantages over DP-SGD and other public-data-based baselines, with notably better FID scores on both synthetic tasks and LoRA fine-tuning of diffusion models.

0 citationsRead paper

CV-Rules: Serializability Verification of Concurrency Control Protocols via Explicit Transaction Ordering

Jun 24, 2026

This work addresses the challenge of efficiently verifying whether concurrency control protocols satisfy serializability semantics. The authors propose CV-Rules, a novel method that decomposes serializability into two orthogonal conditions: causality (C-rules) and view consistency (V-rules), enabling explicit construction of transaction orders and overcoming limitations of traditional multi-version serialization graph techniques. Under bounded causal width, CV-Rules permits polynomial-time decidability and, for the first time, yields explicit transaction orders for several protocols. Implemented within the Lean theorem prover, the formal verification requires no additional axioms and successfully verifies Two-Phase Locking, MVTO, SSN, Aria, and SnapChain. The analysis further reveals that Aria’s unique-write constraint is unnecessary and demonstrates that SnapChain natively satisfies V-rules.

0 citationsRead paper

Recent publications

Latest Papers

Retrieving and Refining Winning Noise Tickets for Diffusion-Based Motion Generation

Jul 07, 2026

Existing diffusion-based text-to-motion generation models often suffer from semantic drift in long-horizon and compositional motions, struggling to balance semantic fidelity with motion coherence. This work proposes WINRO, a novel framework that reveals—for the first time—the decisive role of initial noise in determining the semantics of generated motions. WINRO introduces a training-free, model-agnostic mechanism for noise retrieval and refinement: by retrieving the most text-aligned “winning noise ticket” and applying KL-regularized optimization—optionally enhanced with a single-step forward LoRA adapter—it substantially improves text-motion alignment. Experiments demonstrate that WINRO effectively enhances semantic consistency for both MDM and MotionLCM on HumanML3D, boosts temporal robustness on the MTT benchmark, and generalizes successfully to motion stylization and spatially constrained motion synthesis tasks.

0 citationsRead paper

ARMS: Anchor-Relational Motion Streaming for Seamless Solo-Social Motion Transitions

Jul 06, 2026

This work addresses the challenge of generating long-term, socially coherent human motion that naturally transitions between solitary and interactive states. The authors propose a unified causal generative framework that seamlessly integrates individual motion and interpersonal interaction within a single motion stream, enabling incremental long-horizon generation conditioned solely on historical context. Key innovations include a dynamically asymmetric representation to decouple individual temporal evolution from interpersonal alignment, a mode-aware relational gating mechanism for smooth transitions between generation modes, and a diffusion model operating in a causal latent space augmented with partner-referenced relative translation representations. Experiments demonstrate that the method outperforms existing interaction-centric baselines in terms of smoothness during solitary-to-social transitions and social coherence, while achieving competitive performance on standard human-human interaction benchmarks.

0 citationsRead paper

InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation

Jul 02, 2026

Existing methods for generating two-person interactive motions struggle to simultaneously preserve long-range temporal causality of individual actions and mutual coordination between agents: bidirectional denoising disrupts temporal causality, while autoregressive strategies suffer from coordination degradation due to temporal drift. This work proposes a block-causal latent diffusion framework that employs dual-stream causal diffusion Transformers to model each agent’s motion separately and introduces a unified multi-task attention mask to jointly capture diverse interaction patterns—including synchronization, response, leader-follower dynamics, and independence. By adopting block-level diffusion targets, the method avoids redundant encoding-decoding and enables mask-controlled interaction-type specification during inference. It achieves state-of-the-art performance on the InterHuman and Inter-X datasets, significantly improving text-motion alignment, motion realism, and long-term temporal coherence.

0 citationsRead paper

Escaping Iterative Parameter-Space Noise: Differentially Private Learning with a Hypernetwork

Jun 25, 2026

This work addresses the significant utility degradation in differentially private neural network training—such as DP-SGD—caused by repeated noise injection in high-dimensional parameter spaces. The authors propose a novel framework that leverages a hypernetwork pretrained on public data to map private data into target model parameters, injecting privacy-preserving noise only once in a low-dimensional data embedding space, thereby avoiding direct perturbation of high-dimensional parameters. By integrating embedding aggregation, a tailored perturbation mechanism, and LoRA fine-tuning, the method substantially improves model performance under a fixed privacy budget. Empirical results demonstrate clear advantages over DP-SGD and other public-data-based baselines, with notably better FID scores on both synthetic tasks and LoRA fine-tuning of diffusion models.

0 citationsRead paper

CV-Rules: Serializability Verification of Concurrency Control Protocols via Explicit Transaction Ordering

Jun 24, 2026

This work addresses the challenge of efficiently verifying whether concurrency control protocols satisfy serializability semantics. The authors propose CV-Rules, a novel method that decomposes serializability into two orthogonal conditions: causality (C-rules) and view consistency (V-rules), enabling explicit construction of transaction orders and overcoming limitations of traditional multi-version serialization graph techniques. Under bounded causal width, CV-Rules permits polynomial-time decidability and, for the first time, yields explicit transaction orders for several protocols. Implemented within the Lean theorem prover, the formal verification requires no additional axioms and successfully verifies Two-Phase Locking, MVTO, SSN, Aria, and SnapChain. The analysis further reveals that Aria’s unique-write constraint is unnecessary and demonstrates that SnapChain natively satisfies V-rules.

0 citationsRead paper