Naifan Zhang

Zhiyuan Peng, Naifan Zhang, Yuanbo Tang et al.

DNA sequences encode critical genetic information, yet their variable length and discrete nature impede direct utilization in deep learning models. Existing DNA representation schemes convert sequences into numerical vectors but fail to capture structural features of local subsequences and often suffer from limited interpretability and poor generalization on small datasets. To address these limitations, we propose Dy-mer, an interpretable and robust DNA representation scheme based on dictionary learning. Dy-mer formulates an optimization problem in tensor format, which ensures computational efficiency in batch processing. Our scheme reconstructs DNA sequences as concatenations of dynamic-length subsequences (dymers) through a convolution operation and simultaneously optimize a learnable dymer dictionary and sparse representations. Our method achieves state-of-the-art performance in downstream tasks such as DNA promoter classification and motif detection. Experiments further show that the learned dymers match known DNA motifs and clustering using Dy-mer yields semantically meaningful phylogenetic trees. These results demonstrate that the proposed approach achieves both strong predictive performance and high interpretability, making it well suited for biological research applications.

Yuanbo Tang, Yan Tang, Naifan Zhang et al.

Mixture-of-Experts based large language models (MoE LLMs) have shown significant promise in multitask adaptability by dynamically routing inputs to specialized experts. Despite their success, the collaborative mechanisms among experts are still not well understood, limiting both the interpretability and optimization of these models. In this paper, we focus on two critical issues: (1) identifying expert collaboration patterns, and (2) optimizing MoE LLMs through expert pruning. To address the first issue, we propose a hierarchical sparse dictionary learning (HSDL) method that uncovers the collaboration patterns among experts. For the second issue, we introduce the Contribution-Aware Expert Pruning (CAEP) algorithm, which effectively prunes low-contribution experts. Our extensive experiments demonstrate that expert collaboration patterns are closely linked to specific input types and exhibit semantic significance across various tasks. Moreover, pruning experiments show that our approach improves overall performance by 2.5\% on average, outperforming existing methods. These findings offer valuable insights into enhancing the efficiency and interpretability of MoE LLMs, offering a clearer understanding of expert interactions and improving model optimization.

Naifan Zhang, Ruihan Sun, Jinwei Su et al.

Subjective multi-turn dialogue tasks, such as emotional support, require conversational policies that adapt to evolving user states and optimize long-horizon interaction quality. However, reinforcement learning (RL) for such settings remains challenging due to the absence of reliable process supervision. Outcome-only training collapses credit assignment across turns into a single trajectory-level reward, while naïve turn-level group sampling incurs prohibitive rollout costs in interactive environments. We propose a critic-free and efficient RL algorithm named MAPO that leverages dense process feedback from a judge model and propagates long-horizon effects through Monte Carlo returns. To stabilize optimization, we introduce a mixed advantage estimator that combines turn-level normalization with batch-level normalization, enabling fine-grained yet scalable credit assignment. Across multiple subjective dialogue benchmarks, including EMPA, EmoBench, and EQ-Bench, and model scales ranging from 7B to 32B, our method consistently improves both training stability and final performance over outcome-only GRPO and single-level normalization baselines. On EMPA, we improve rates by up to 9 points and increase dialogue scores by as much as +43.2 over the 7B base model. Despite training only on EMPA-style environments, our approach generalizes well, yielding consistent improvements on unseen emotional-intelligence benchmarks, including up to +4 points on EmoBench and +3.5 on EQ-Bench. Together, these results demonstrate that dense process supervision combined with mixed-level normalization enables effective and scalable RL for subjective, open-ended multi-turn dialogue.