惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
大猫的无限游戏
大猫的无限游戏
博
博客园 - 聂微东
Jina AI
Jina AI
The Cloudflare Blog
V
Visual Studio Blog
博
博客园_首页
量
量子位
酷 壳 – CoolShell
酷 壳 – CoolShell
博
博客园 - 【当耐特】
爱范儿
爱范儿
博
博客园 - 三生石上(FineUI控件)
小众软件
小众软件
博
博客园 - 司徒正美
阮一峰的网络日志
阮一峰的网络日志
Last Week in AI
Last Week in AI
V
V2EX
钛媒体：引领未来商业与生活新知
钛媒体：引领未来商业与生活新知
博
博客园 - 叶小钗
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
WordPress大学
WordPress大学
宝玉的分享
宝玉的分享
T
Tailwind CSS Blog
博
博客园 - Franky

Proceedings of the AAAI Conference on Artificial Intelligence

Resource Efficient Sleep Staging via Multi-Level Masking and Prompt Learning AutoMalDesc: Large-Scale Script Analysis for Cyber Threat Research Modulation-Based Backdoors: Leveraging Amplitude and Frequency Patterns to Attack Speaker Recognition Learning Structurally Stabilized Representations for Lossless DNA Storage ViG-RAG: Video-aware Graph Retrieval-Augmented Generation via Temporal and Semantic Hybrid Reasoning Transferable Backdoor Attacks for Code Models via Sharpness-Aware Adversarial Perturbation Toward Multimodal Fake News Detection by Multi-perspective Rationale Generation and Verification RTMol: Rethinking Molecule-text Alignment in a Round-trip View Physical-regularized Hierarchical Generative Model for Metallic Glass Structural Generation and Energy Prediction Regressor-guided Diffusion Model for De Novo Peptide Sequencing with Explicit Mass Control RareAgents: Autonomous Multi-disciplinary Team for Rare Disease Diagnosis and Treatment Transferring Causal Driving Patterns for Generalizable Traffic Simulation with Diffusion-Based Distillation TRACE: Transformation-Aware Graph Refinement for Reaction Condition Prediction SIDE: Surrogate Conditional Data Extraction from Diffusion Models DyC-STG: Dynamic Causal Spatio-Temporal Graph Network for Real-time Data Credibility Analysis in IoT ProAR: Probabilistic Autoregressive Modeling for Molecular Dynamics Light but Sharp: SlimSTAD for Real-Time Action Detection from Sensor Data VFCionX: Bridging Large and Small Models for Robust Vulnerability-Fixing Commit Identification T2Agent: A Tool-augmented Multimodal Misinformation Detection Agent with Monte Carlo Tree Search Measuring What Matters: Scenario-Driven Evaluation for Trajectory Predictors in Autonomous Driving DensiCrafter: Physically-Constrained Generation and Fabrication of Self-Supporting Hollow Structures Topology-Enhanced and Label Correlation-Aware Model for Protein-Protein Interaction Prediction InteChar: A Unified Oracle Bone Character List for Ancient Chinese Language Modeling NucEL: Single-Nucleotide ELECTRA-Style Genomic Pre-training for Efficient and Interpretable Representations OR-R1: Automating Modeling and Solving of Operations Research Optimization Problem via Test-Time Reinforcement Learning Learning from Long-Term Engagement: Adaptive Tutoring Dialogue Planning for Personalized Education Toward Time-Continuous Data Inference in Sparse Urban CrowdSensing Multi-Horizon Time Series Forecasting of Non-Parametric CDFs with Deep Lattice Networks From Stimuli to Minds: Enhancing Psychological Reasoning in LLMs via Bilateral Reinforcement Learning Unveiling the Attribute Misbinding Threat in Identity-Preserving Models
MergeDNA: Context-Aware Genome Modeling with Dynamic Toke...
Siyuan Li, K · 2026-03-14 · via Proceedings of the AAAI Conference on Artificial Intelligence

Authors

  • Siyuan Li Zhejiang University Westlake University BioMap Research
  • Kai Yu Westlake University
  • Anna Wang Westlake University
  • Zicheng Liu Zhejiang University Westlake University BioMap Research
  • Chang Yu Westlake University
  • Jingbo Zhou Zhejiang University Westlake University
  • Qirong Yang BioMap Research
  • Yucheng Guo BioMap Research
  • Xiaoming Zhang BioMap Research
  • Stan Z. Li Westlake University

DOI:

https://doi.org/10.1609/aaai.v40i1.37032

Abstract

Modeling genomic sequences faces two unsolved challenges: the information density varies widely across different regions, while there is no clearly defined minimum vocabulary unit. Relying on either four primitive bases or independently designed DNA tokenizers, existing approaches with naive masked language modeling pre-training often fail to adapt to the varying complexities of genomic sequences. Leveraging Token Merging techniques, this paper introduces a hierarchical architecture that jointly optimizes a dynamic genomic tokenizer and latent Transformers with context-aware pre-training tasks. As for network structures, the tokenization module automatically chunks adjacent bases into words by stacking multiple layers of the differentiable token merging blocks with local-window constraints, then a Latent Encoder captures the global context of these merged words by full-attention blocks. Symmetrically employing a Latent Decoder and a Local Decoder, MergeDNA learns with two pre-training tasks: Merged Token Reconstruction simultaneously trains the dynamic tokenization module and adaptively filters important tokens, while Adaptive Masked Token Modeling learns to predict these filtered tokens to capture informative contents. Extensive experiments show that MergeDNA achieves superior performance on three popular DNA benchmarks and several multi-omics tasks with fine-tuning or zero-shot evaluation, outperforming typical tokenization methods and large-scale DNA foundation models.

How to Cite

Li, S., Yu, K., Wang, A., Liu, Z., Yu, C., Zhou, J., … Li, S. Z. (2026). MergeDNA: Context-Aware Genome Modeling with Dynamic Tokenization Through Token Merging. Proceedings of the AAAI Conference on Artificial Intelligence, 40(1), 668–676. https://doi.org/10.1609/aaai.v40i1.37032

Issue

Section

AAAI Technical Track on Application Domains I

此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。