Rongduo Han

Cihan Ruan, Lebin Zhou, Bingqing Zhao et al.

DNA-based storage has emerged as a promising approach to the global data crisis, offering molecular-scale density and millennial-scale stability at low maintenance cost. Over the past decade, substantial progress has been made in storing text, images, and files in DNA -- yet video remains an open challenge. The difficulty is not merely technical: effective video DNA storage requires co-designing compression and molecular encoding from the ground up, a challenge that sits at the intersection of two fields that have largely evolved independently. In this work, we present HELIX, the first end-to-end neural network jointly optimizing video compression and DNA encoding -- prior approaches treat the two stages independently, leaving biochemical constraints and compression objectives fundamentally misaligned. Our key insight: token-based representations naturally align with DNA's quaternary alphabet -- discrete semantic units map directly to ATCG bases. We introduce TK-SCONE (Token-Kronecker Structured Constraint-Optimized Neural Encoding), which achieves 1.91 bits per nucleotide through Kronecker-structured mixing that breaks spatial correlations and FSM-based mapping that guarantees biochemical constraints. Unlike two-stage approaches, HELIX learns token distributions simultaneously optimized for visual quality, prediction under masking, and DNA synthesis efficiency. This work demonstrates for the first time that learned compression and molecular storage converge naturally at token representations -- suggesting a new paradigm where neural video codecs are designed for biological substrates from the ground up.

Cihan Ruan, Lebin Zhou, Rongduo Han et al.

DNA storage has matured from concept to practical stage, yet its integration with neural compression pipelines remains inefficient. Early DNA encoders applied redundancy-heavy constraint layers atop raw binary data - workable but primitive. Recent neural codecs compress data into learned latent representations with rich statistical structure, yet still convert these latents to DNA via naive binary-to-quaternary transcoding, discarding the entropy model's optimization. This mismatch undermines compression efficiency and complicates the encoding stack. A plug-in module that collapses latent compression and DNA encoding into a single step. SCONE performs quaternary arithmetic coding directly on the latent space in DNA bases. Its Constraint-Aware Adaptive Coding module dynamically steers the entropy encoder's learned probability distribution to enforce biochemical constraints - Guanine-Cytosine (GC) balance and homopolymer suppression - deterministically during encoding, eliminating post-hoc correction. The design preserves full reversibility and exploits the hyperprior model's learned priors without modification. Experiments show SCONE achieves near-perfect constraint satisfaction with negligible computational overhead (<2% latency), establishing a latent-agnostic interface for end-to-end DNA-compatible learned codecs.

Zhennan Shen, Yanshu Li, Qingyu Yin et al.

On-policy distillation (OPD) is increasingly used to improve large language model reasoning, but its training dynamics remain poorly understood. We characterize the trajectory of OPD updates in parameter space and compare it with supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). A suite of parameter-space diagnostics consistently places OPD in a relaxed off-principal regime: compared with SFT, its updates affect fewer weights and avoid principal directions more strongly, while compared with RLVR, they remain less tightly constrained. Beyond this static localization, OPD exhibits subspace locking: its cumulative updates rapidly enter a narrow low-dimensional channel. Constraining training to the update subspace formed early in training preserves OPD performance but substantially degrades SFT, indicating that the locked subspace is functionally sufficient for OPD. Control experiments further show that sparsifying the update tokens and shifting rollout generation off-policy preserve the rank dynamics, whereas mixing the OPD objective with RLVR changes them. Overall, these results suggest that OPD is not merely an intermediate point between SFT and RLVR, but induces its own update geometry in parameter space.

Meiping Wang, Jian Zhong, Rongduo Han et al.

With the rapid development of mobile intelligent assistant technologies, multi-modal AI assistants have become essential interfaces for daily user interactions. However, current evaluation methods face challenges including high manual costs, inconsistent standards, and subjective bias. This paper proposes an automated multi-modal evaluation framework based on large language models and multi-agent collaboration. The framework employs a three-tier agent architecture consisting of interaction evaluation agents, semantic verification agents, and experience decision agents. Through supervised fine-tuning on the Qwen3-8B model, we achieve a significant evaluation matching accuracy with human experts. Experimental results on eight major intelligent agents demonstrate the framework's effectiveness in predicting users' satisfaction and identifying generation defects.