# DAST Papers

DAST AI curated research paper collection across agents, robotics, AI for Science, safety, systems, hardware, and multimodal foundation models.

Canonical site: https://papers.dast.ai/
Generated: 2026-05-29T02:47:21+00:00
Total papers: 1063
Papers with interpretation videos: 241

## Use

This is a selective research collection, not a generic paper dump. It emphasizes reusable methods, durable benchmarks, workflow-changing systems, and high-spillover AI research.

Important URLs:
- Home and interactive search: https://papers.dast.ai/
- Machine-readable JSON index: https://papers.dast.ai/data/papers.json
- XML sitemap: https://papers.dast.ai/sitemap.xml
- Full paper list for AI assistants: https://papers.dast.ai/llms-full.txt

## Main Categories

- agents_and_autonomous_science: 305
- neuroscience_and_cognitive_science: 113
- chemistry_biology_and_lab_automation: 93
- reasoning_memory_and_inference_control: 82
- multimodal_foundation_models: 75
- reinforcement_learning: 73
- physics_and_ai_for_science: 40
- theory_robustness_and_core_ml: 39
- ai_hardware_and_accelerator_design: 36
- generative_modeling_and_diffusion: 29
- scientific_discovery_flagships: 29
- interpretability_and_mechanistic_analysis: 20
- jepa_and_predictive_world_models: 15
- robotics_and_embodied_intelligence: 14
- public_health_and_medical_operations: 12
- safety_governance_and_reliability: 12
- math_and_formal_reasoning: 11
- weather_climate_and_earth_systems: 8
- industrial_process_and_manufacturing_systems: 7
- software_engineering_and_coding_agents: 7
- energy_water_and_infrastructure_systems: 6
- biology_biomedicine_drug_discovery: 5
- generative_models_and_world_models: 4
- geospatial_remote_sensing_and_disaster_systems: 4
- projects_and_appendices: 4
- systems_hardware_and_codesign: 4
- life_science_foundation_models: 3
- agriculture_and_food_systems: 2
- ai_for_science: 2
- high_value_real_world_workflows: 2
- marine_fisheries_and_coastal_systems: 2
- supply_chain_logistics_and_operations: 2
- environment_and_ecological_systems: 1
- multimodal_generation_and_world_models: 1
- software_engineering: 1

## Recent Papers

- [Recursive Flow Matching](https://papers.dast.ai/papers/p35-recursive-flow-matching/) | id=P35_Recursive_Flow_Matching | theme=物理与 AI for Science | grade=突破级 | published=2026-05-26 | arxiv=2605.26535
- [PRISM: Position-encoded Regressive Inverse Spectral Model for Multilayer Thin-Film Design](https://papers.dast.ai/papers/p34-prism-thin-film-inverse-design/) | id=P34_PRISM_Thin_Film_Inverse_Design | theme=物理与 AI for Science | grade=突破级 | published=2026-05-26 | arxiv=2605.26502
- [DVAO: Dynamic Variance-adaptive Advantage Optimization for Multi-reward Reinforcement Learning](https://papers.dast.ai/papers/rl82-dvao-multi-reward-rl/) | id=RL82_DVAO_Multi_Reward_RL | theme=强化学习 | grade=突破级 | published=2026-05-25 | arxiv=2605.25604
- [Language Models Need Sleep](https://papers.dast.ai/papers/r81-language-models-need-sleep/) | id=R81_Language_Models_Need_Sleep | theme=推理、记忆与推理时控制 | grade=突破级 | published=2026-05-25 | arxiv=2605.26099
- [WBench: A Comprehensive Multi-turn Benchmark for Interactive Video World Model Evaluation](https://papers.dast.ai/papers/g23-wbench-interactive-video-world-model-eval/) | id=G23_WBench_Interactive_Video_World_Model_Eval | theme=生成建模与扩散 | grade=突破级 | published=2026-05-25 | arxiv=2605.25874
- [Hierarchical Multi-agent Large Language Model Reasoning for Autonomous Heterogeneous Catalyst Discovery](https://papers.dast.ai/papers/c35-master-multiagent-catalyst-discovery/) | id=C35_MASTER_Multiagent_Catalyst_Discovery | theme=化学、生物与自动化实验室 | grade=突破级 | published=2026-05-25 | doi=10.1038/s41524-026-02139-1
- [SOFisher: reinforcement learning-guided experiment designs for spatial omics](https://papers.dast.ai/papers/c34-sofisher-rl-spatial-omics-design/) | id=C34_SOFisher_RL_Spatial_Omics_Design | theme=化学、生物与自动化实验室 | grade=突破级 | published=2026-05-25 | doi=10.1038/s41467-026-73404-6
- [A large-scale unified deep learning model for peptide mass spectrum interpretation trained on multimodal data](https://papers.dast.ai/papers/bio57-punifind-proteomics-mass-spectrum-model/) | id=BIO57_pUniFind_Proteomics_Mass_Spectrum_Model | theme=化学、生物与自动化实验室 | grade=突破级 | published=2026-05-25 | arxiv=2507.00087
- [De novo design of DNA origami with a generative diffusion model](https://papers.dast.ai/papers/bio56-dna-origami-diffusion-design/) | id=BIO56_DNA_Origami_Diffusion_Design | theme=化学、生物与自动化实验室 | grade=突破级 | published=2026-05-25 | doi=10.1038/s41467-026-73578-z
- [Personalize-then-Store: Benchmarking and Learning Personalized Memory for Long-horizon Agents](https://papers.dast.ai/papers/a323-permembench-personalized-memory/) | id=A323_PerMemBench_Personalized_Memory | theme=智能体与自主科学 | grade=突破级 | published=2026-05-25 | arxiv=2605.25535
- [Anticipate and Learn: Unleashing Idle-Time Compute in Proactive Agents](https://papers.dast.ai/papers/a321-proact-proactive-agents/) | id=A321_ProAct_Proactive_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-25 | arxiv=2605.25971
- [CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents](https://papers.dast.ai/papers/a320-cua-gym-verifiable-cua-rlvr/) | id=A320_CUA_Gym_Verifiable_CUA_RLVR | theme=智能体与自主科学 | grade=突破级 | published=2026-05-25 | arxiv=2605.25624
- [Claw-Anything: Benchmarking Always-On Personal Assistants with Broader Access to User's Digital World](https://papers.dast.ai/papers/a316-claw-anything-always-on-assistants/) | id=A316_Claw_Anything_Always_On_Assistants | theme=智能体与自主科学 | grade=突破级 | published=2026-05-25 | arxiv=2605.26086
- [Faithfulness Metrics Don't Measure Faithfulness: A Meta-Evaluation with Ground Truth](https://papers.dast.ai/papers/t41-bonafide-cot-faithfulness-meta-eval/) | id=T41_BonaFide_CoT_Faithfulness_Meta_Eval | theme=理论、鲁棒性与核心机器学习 | grade=突破级 | published=2026-05-24 | arxiv=2605.25052
- [SimuWoB: Simulating Real-World Mobile Apps for Fast and Faithful GUI Agent Benchmarking](https://papers.dast.ai/papers/a324-simuwob-mobile-gui-agent-benchmark/) | id=A324_SimuWoB_Mobile_GUI_Agent_Benchmark | theme=智能体与自主科学 | grade=突破级 | published=2026-05-24 | arxiv=2605.25160
- [Macaron-A2UI: A Model for Generative UI in Personal Agents](https://papers.dast.ai/papers/a313-macaron-a2ui-generative-ui-agents/) | id=A313_Macaron_A2UI_Generative_UI_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-24 | arxiv=2605.24830
- [Trainable neuromorphic spintronic hardware Via analog finite-difference gradient methods](https://papers.dast.ai/papers/hw35-trainable-neuromorphic-spintronic-hardware/) | id=HW35_Trainable_Neuromorphic_Spintronic_Hardware | theme=AI 硬件与加速器 | grade=突破级 | published=2026-05-23 | doi=10.1038/s41467-026-73624-w
- [ECHO: Terminal Agents Learn World Models for Free](https://papers.dast.ai/papers/a325-echo-terminal-agent-world-models/) | id=A325_ECHO_Terminal_Agent_World_Models | theme=智能体与自主科学 | grade=突破级 | published=2026-05-23 | arxiv=2605.24517
- [SEAL: Synergistic Co-Evolution of Agents and Learning Environments](https://papers.dast.ai/papers/a317-seal-agent-environment-co-evolution/) | id=A317_SEAL_Agent_Environment_Co_Evolution | theme=智能体与自主科学 | grade=突破级 | published=2026-05-23 | arxiv=2605.24426
- [CoSPlay: Cooperative Self-Play at Test-Time with Self-Generated Code and Unit Test](https://papers.dast.ai/papers/se9-cosplay-test-time-code-self-play/) | id=SE9_CoSPlay_Test_Time_Code_Self_Play | theme=软件工程 | grade=突破级 | published=2026-05-22 | arxiv=2605.23491
- [Computational neural dynamics of goal-directed visual attention in macaques](https://papers.dast.ai/papers/n108-goal-directed-visual-attention-dynamics/) | id=N108_Goal_Directed_Visual_Attention_Dynamics | theme=神经科学与认知科学 | grade=突破级 | published=2026-05-22 | doi=10.1126/sciadv.aed9004
- [SCOPE: Simulating Cross-game Operations in Playable Environments for FPS World Models](https://papers.dast.ai/papers/g24-scope-cross-game-fps-world-models/) | id=G24_SCOPE_Cross_Game_FPS_World_Models | theme=生成建模与扩散 | grade=突破级 | published=2026-05-22 | arxiv=2605.23345
- [From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent Skills](https://papers.dast.ai/papers/a326-skilllens-model-generated-agent-skills/) | id=A326_SkillLens_Model_Generated_Agent_Skills | theme=智能体与自主科学 | grade=突破级 | published=2026-05-22 | arxiv=2605.23899
- [SkillEvolBench: Benchmarking the Evolution from Episodic Experience to Procedural Skills](https://papers.dast.ai/papers/a322-skillevolbench-procedural-skills/) | id=A322_SkillEvolBench_Procedural_Skills | theme=智能体与自主科学 | grade=突破级 | published=2026-05-22 | arxiv=2605.24117
- [QUEST: Training Frontier Deep Research Agents with Fully Synthetic Tasks](https://papers.dast.ai/papers/a315-quest-synthetic-deep-research-agents/) | id=A315_QUEST_Synthetic_Deep_Research_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-22 | arxiv=2605.24218
- [Foundation Protocol: A Coordination Layer for Agentic Society](https://papers.dast.ai/papers/a314-foundation-protocol-agentic-society/) | id=A314_Foundation_Protocol_Agentic_Society | theme=智能体与自主科学 | grade=突破级 | published=2026-05-22 | arxiv=2605.23218
- [SkillOpt: Executive Strategy for Self-Evolving Agent Skills](https://papers.dast.ai/papers/a312-skillopt-self-evolving-agent-skills/) | id=A312_SkillOpt_Self_Evolving_Agent_Skills | theme=智能体与自主科学 | grade=突破级 | published=2026-05-22 | arxiv=2605.23904
- [The Illusion of Reasoning: Exposing Evasive Data Contamination in LLMs via Zero-CoT Truncation](https://papers.dast.ai/papers/t42-zero-cot-contamination-probe/) | id=T42_Zero_CoT_Contamination_Probe | theme=理论、鲁棒性与核心机器学习 | grade=突破级 | published=2026-05-21 | arxiv=2605.21856
- [Forecasting Scientific Progress with Artificial Intelligence](https://papers.dast.ai/papers/s84-cusp-scientific-progress-forecasting/) | id=S84_CUSP_Scientific_Progress_Forecasting | theme=科学发现旗舰工作 | grade=突破级 | published=2026-05-21 | arxiv=2605.22681
- [ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention](https://papers.dast.ai/papers/r79-thriftattention-fp4-long-context/) | id=R79_ThriftAttention_FP4_Long_Context | theme=推理、记忆与推理时控制 | grade=突破级 | published=2026-05-21 | arxiv=2605.23081
- [Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention](https://papers.dast.ai/papers/r78-gated-deltanet-2/) | id=R78_Gated_DeltaNet_2 | theme=推理、记忆与推理时控制 | grade=突破级 | published=2026-05-21 | arxiv=2605.22791
- [Learning to Think in Physics: Breaking Shortcut Learning in Scientific Diffusion via Representation Alignment](https://papers.dast.ai/papers/p33-repa-p-physics-diffusion-alignment/) | id=P33_REPA_P_Physics_Diffusion_Alignment | theme=物理与 AI for Science | grade=突破级 | published=2026-05-21 | arxiv=2605.20780
- [Reward magnitude determines reinforcement learning efficiency](https://papers.dast.ai/papers/n98-reward-magnitude-rl-efficiency/) | id=N98_Reward_Magnitude_RL_Efficiency | theme=神经科学与认知科学 | grade=突破级 | published=2026-05-21 | doi=10.1126/science.aeb0813
- [VGenST-Bench: A Benchmark for Spatio-Temporal Reasoning via Active Video Synthesis](https://papers.dast.ai/papers/mm69-vgenst-bench-active-video-synthesis-eval/) | id=MM69_VGenST_Bench_Active_Video_Synthesis_Eval | theme=多模态基础模型 | grade=突破级 | published=2026-05-21 | arxiv=2605.22570
- [Advancing Mathematics Research with AI-Driven Formal Proof Search](https://papers.dast.ai/papers/m52-ai-formal-proof-search/) | id=M52_AI_Formal_Proof_Search | theme=数学与形式推理 | grade=颠覆级 | published=2026-05-21 | arxiv=2605.22763
- [Therm-FM: Foundation Model is ALL YOU NEED for 3D-ICs Thermal Simulation](https://papers.dast.ai/papers/hw34-therm-fm-3dic-thermal/) | id=HW34_Therm_FM_3DIC_Thermal | theme=系统、硬件与协同设计 | grade=突破级 | published=2026-05-21 | arxiv=2605.22663
- [SpectralEarth-FM: Bringing Hyperspectral Imagery into Multimodal Earth Observation Pretraining](https://papers.dast.ai/papers/geo4-spectralearth-fm-hyperspectral-eo/) | id=GEO4_SpectralEarth_FM_Hyperspectral_EO | theme=地理空间、遥感与灾害系统 | grade=突破级 | published=2026-05-21 | arxiv=2605.21075
- [WorldKV: Efficient World Memory with World Retrieval and Compression](https://papers.dast.ai/papers/g20-worldkv-world-memory/) | id=G20_WorldKV_World_Memory | theme=生成建模与扩散 | grade=突破级 | published=2026-05-21 | arxiv=2605.22718
- [Rethinking Cross-Layer Information Routing in Diffusion Transformers](https://papers.dast.ai/papers/g19-dar-diffusion-transformer-routing/) | id=G19_DAR_Diffusion_Transformer_Routing | theme=生成建模与扩散 | grade=突破级 | published=2026-05-21 | arxiv=2605.20708
- [SciCore-Mol: Augmenting Large Language Models with Pluggable Molecular Cognition Modules](https://papers.dast.ai/papers/c33-scicore-mol-pluggable-cognition/) | id=C33_SciCore_Mol_Pluggable_Cognition | theme=化学、生物与自动化实验室 | grade=突破级 | published=2026-05-21 | arxiv=2605.22287
- [Maestro: Reinforcement Learning to Orchestrate Hierarchical Model-Skill Ensembles](https://papers.dast.ai/papers/a308-maestro-model-skill-orchestration/) | id=A308_Maestro_Model_Skill_Orchestration | theme=智能体与自主科学 | grade=突破级 | published=2026-05-21 | arxiv=2605.22177
- [Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning](https://papers.dast.ai/papers/a307-spreadsheet-rl-excel-agents/) | id=A307_Spreadsheet_RL_Excel_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-21 | arxiv=2605.22642
- [ACC: Compiling Agent Trajectories for Long-Context Training](https://papers.dast.ai/papers/a306-acc-agent-context-compilation/) | id=A306_ACC_Agent_Context_Compilation | theme=智能体与自主科学 | grade=突破级 | published=2026-05-21 | arxiv=2605.21850
- [AgentCo-op: Retrieval-Based Synthesis of Interoperable Multi-Agent Workflows](https://papers.dast.ai/papers/a304-agentcoop-interoperable-workflows/) | id=A304_AgentCoop_Interoperable_Workflows | theme=智能体与自主科学 | grade=突破级 | published=2026-05-21 | arxiv=2605.20425
- [Surviving the Unseen: Predictive Defense for Novel Multi-Turn Multimodal Attacks](https://papers.dast.ai/papers/t37-triad-multimodal-attack-defense/) | id=T37_TRIAD_Multimodal_Attack_Defense | theme=安全、治理与可靠性 | grade=突破级 | published=2026-05-20 | arxiv=2605.18988
- [RoboJailBench: Benchmarking Adversarial Attacks and Defenses in Embodied Robotic Agents](https://papers.dast.ai/papers/t36-robojailbench/) | id=T36_RoboJailBench | theme=安全、治理与可靠性 | grade=突破级 | published=2026-05-20 | arxiv=2605.19328
- [Robotics-Inspired Guardrails for Foundation Models in Socially Sensitive Domains](https://papers.dast.ai/papers/t35-robotics-inspired-guardrails/) | id=T35_Robotics_Inspired_Guardrails | theme=安全、治理与可靠性 | grade=突破级 | published=2026-05-20 | arxiv=2605.19940
- [SciAtlas: A Large-Scale Knowledge Graph for Automated Scientific Research](https://papers.dast.ai/papers/s85-sciatlas-automated-scientific-research-kg/) | id=S85_SciAtlas_Automated_Scientific_Research_KG | theme=科学发现旗舰工作 | grade=突破级 | published=2026-05-20 | arxiv=2605.22878
- [Imaging hidden objects with consumer LiDAR via motion-induced sampling](https://papers.dast.ai/papers/rl83-consumer-lidar-hidden-object-imaging/) | id=RL83_Consumer_LiDAR_Hidden_Object_Imaging | theme=机器人与具身智能 | grade=突破级 | published=2026-05-20 | doi=10.1038/s41586-026-10502-x
- [DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards](https://papers.dast.ai/papers/rl80-delta-rlvr-credit-assignment/) | id=RL80_DelTA_RLVR_Credit_Assignment | theme=强化学习 | grade=突破级 | published=2026-05-20 | arxiv=2605.21467
- [Probing Embodied LLMs: When Higher Observation Fidelity Hurts Problem Solving](https://papers.dast.ai/papers/rl77-embodied-llm-observation-fidelity/) | id=RL77_Embodied_LLM_Observation_Fidelity | theme=机器人与具身智能 | grade=突破级 | published=2026-05-20 | arxiv=2605.20072
- [Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning](https://papers.dast.ai/papers/r80-equilibrium-reasoners-attractors/) | id=R80_Equilibrium_Reasoners_Attractors | theme=推理、记忆与推理时控制 | grade=突破级 | published=2026-05-20 | arxiv=2605.21488
- [Neural representation of action symbols in primate frontal cortex](https://papers.dast.ai/papers/n94-primate-action-symbols-frontal-cortex/) | id=N94_Primate_Action_Symbols_Frontal_Cortex | theme=神经科学与认知科学 | grade=突破级 | published=2026-05-20 | doi=10.1038/s41586-026-10297-x
- [A critical initialization for biological neural networks](https://papers.dast.ai/papers/n93-critical-initialization-biological-networks/) | id=N93_Critical_Initialization_Biological_Networks | theme=神经科学与认知科学 | grade=突破级 | published=2026-05-20 | doi=10.1038/s41586-026-10528-1
- [Vision Harnessing Agent for Open Ad-hoc Segmentation](https://papers.dast.ai/papers/mm66-vasa-open-adhoc-segmentation/) | id=MM66_VASA_Open_Adhoc_Segmentation | theme=多模态基础模型 | grade=突破级 | published=2026-05-20 | arxiv=2605.19410
- [RankE: End-to-End Post-Training for Discrete Text-to-Image Generation with Decoder Co-Evolution](https://papers.dast.ai/papers/g26-ranke-discrete-t2i-decoder-co-evolution/) | id=G26_RankE_Discrete_T2I_Decoder_Co_Evolution | theme=生成建模与扩散 | grade=突破级 | published=2026-05-20 | arxiv=2605.21195
- [Lens: Rethinking Training Efficiency for Foundational Text-to-Image Models](https://papers.dast.ai/papers/g25-lens-efficient-foundational-t2i/) | id=G25_Lens_Efficient_Foundational_T2I | theme=生成建模与扩散 | grade=突破级 | published=2026-05-20 | arxiv=2605.21573
- [PhysX-Omni: Unified Simulation-Ready Physical 3D Generation for Rigid, Deformable, and Articulated Objects](https://papers.dast.ai/papers/g21-physx-omni-physical-3d/) | id=G21_PhysX_Omni_Physical_3D | theme=生成建模与扩散 | grade=突破级 | published=2026-05-20 | arxiv=2605.21572
- [Mem-π: Adaptive Memory through Learning When and What to Generate](https://papers.dast.ai/papers/a327-mempi-adaptive-generative-memory/) | id=A327_MemPi_Adaptive_Generative_Memory | theme=智能体与自主科学 | grade=突破级 | published=2026-05-20 | arxiv=2605.21463
- [CutVerse: A Compositional GUI Agents Benchmark for Media Post-Production Editing](https://papers.dast.ai/papers/a299-cutverse-gui-agents/) | id=A299_CutVerse_GUI_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-20 | arxiv=2605.19484
- [Learning to Hand Off: Provably Convergent Workflow Learning under Interface Constraints](https://papers.dast.ai/papers/a298-icq-workflow-handoff/) | id=A298_ICQ_Workflow_Handoff | theme=智能体与自主科学 | grade=突破级 | published=2026-05-20 | arxiv=2605.19140
- [DecisionBench: A Benchmark for Emergent Delegation in Long-Horizon Agentic Workflows](https://papers.dast.ai/papers/a297-decisionbench-delegation/) | id=A297_DecisionBench_Delegation | theme=智能体与自主科学 | grade=突破级 | published=2026-05-20 | arxiv=2605.19099
- [EngiAI: A Multi-Agent Framework and Benchmark Suite for LLM-Driven Engineering Design](https://papers.dast.ai/papers/a296-engiai-engineering-design-agents/) | id=A296_EngiAI_Engineering_Design_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-20 | arxiv=2605.19743
- [Accelerating scientific discovery with Co-Scientist](https://papers.dast.ai/papers/s83-coscientist-hypothesis-generation/) | id=S83_CoScientist_Hypothesis_Generation | theme=科学发现旗舰工作 | grade=范式级 | published=2026-05-19 | doi=10.1038/s41586-026-10644-y
- [AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration](https://papers.dast.ai/papers/s82-autoresearchclaw/) | id=S82_AutoResearchClaw | theme=科学发现旗舰工作 | grade=突破级 | published=2026-05-19 | arxiv=2605.20025
- [An AI system to help scientists write expert-level empirical software](https://papers.dast.ai/papers/s81-era-empirical-research-assistance/) | id=S81_ERA_Empirical_Research_Assistance | theme=科学发现旗舰工作 | grade=颠覆级 | published=2026-05-19 | doi=10.1038/s41586-026-10658-6
- [A multi-agent system for automating scientific discovery](https://papers.dast.ai/papers/s80-robin-multiagent-discovery/) | id=S80_Robin_MultiAgent_Discovery | theme=科学发现旗舰工作 | grade=范式级 | published=2026-05-19 | doi=10.1038/s41586-026-10652-y
- [GoLongRL: Capability-Oriented Long Context Reinforcement Learning with Multitask Alignment](https://papers.dast.ai/papers/r76-golongrl-long-context-rl/) | id=R76_GoLongRL_Long_Context_RL | theme=推理、记忆与推理时控制 | grade=突破级 | published=2026-05-19 | arxiv=2605.19577
- [FPED: A Functional-Network Prior-Guided Mixture-of-Experts Framework for Interpretable Brain Decoding](https://papers.dast.ai/papers/n88-fped-interpretable-brain-decoding/) | id=N88_FPED_Interpretable_Brain_Decoding | theme=神经科学与认知科学 | grade=突破级 | published=2026-05-19 | arxiv=2605.19279
- [ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning](https://papers.dast.ai/papers/mm68-paravt-parallel-video-tool-rl/) | id=MM68_ParaVT_Parallel_Video_Tool_RL | theme=多模态基础模型 | grade=突破级 | published=2026-05-19 | arxiv=2605.20342
- [Streamlined optical training of large-scale modern deep learning architectures with direct feedback alignment](https://papers.dast.ai/papers/hw33-optical-dfa-training/) | id=HW33_Optical_DFA_Training | theme=AI 硬件与加速器 | grade=突破级 | published=2026-05-19
- [PEEK: Context Map as an Orientation Cache for Long-Context LLM Agents](https://papers.dast.ai/papers/a302-peek-context-map-agents/) | id=A302_PEEK_Context_Map_Agents | theme=智能体与自主科学 | grade=突破级 | published=2026-05-19 | arxiv=2605.19932
- [OpenComputer: Verifiable Software Worlds for Computer-Use Agents](https://papers.dast.ai/papers/a301-opencomputer-software-worlds/) | id=A301_OpenComputer_Software_Worlds | theme=智能体与自主科学 | grade=突破级 | published=2026-05-19 | arxiv=2605.19769
- [Agent Security is a Systems Problem](https://papers.dast.ai/papers/t39-agent-security-systems/) | id=T39_Agent_Security_Systems | theme=理论、鲁棒性与核心机器学习 | grade=突破级 | published=2026-05-18 | arxiv=2605.18991
- [Deterministic Event-Graph Substrates as World Models for Counterfactual Reasoning](https://papers.dast.ai/papers/t17-eventgraph-world-models/) | id=T17_EventGraph_World_Models | theme=理论、鲁棒性与核心机器学习 | grade=突破级 | published=2026-05-18 | arxiv=2605.15967
- [AgentStop: Terminating Local AI Agents Early to Save Energy in Consumer Devices](https://papers.dast.ai/papers/sys13-agentstop-local-agents/) | id=SYS13_AgentStop_Local_Agents | theme=系统、硬件与协同设计 | grade=突破级 | published=2026-05-18 | arxiv=2605.15206
- [TopoEvo: A Topology-Aware Self-Evolving Multi-Agent Framework for Root Cause Analysis in Microservices](https://papers.dast.ai/papers/sys12-topoevo-rca-agents/) | id=SYS12_TopoEvo_RCA_Agents | theme=软件工程与编程智能体 | grade=突破级 | published=2026-05-18 | arxiv=2605.15611
- [STAR: A Stage-attributed Triage and Repair framework for RCA Agents in Microservices](https://papers.dast.ai/papers/sys11-star-rca-agents/) | id=SYS11_STAR_RCA_Agents | theme=软件工程与编程智能体 | grade=突破级 | published=2026-05-18 | arxiv=2605.15581
- [ESI-Bench: Towards Embodied Spatial Intelligence that Closes the Perception-Action Loop](https://papers.dast.ai/papers/rl79-esi-bench-spatial-intelligence/) | id=RL79_ESI_Bench_Spatial_Intelligence | theme=机器人与具身智能 | grade=突破级 | published=2026-05-18 | arxiv=2605.18746
- [Bridging the interpretability gap for medical artificial intelligence models using class-association manifold learning](https://papers.dast.ai/papers/ph6-caml-medical-ai-interpretability/) | id=PH6_CAML_Medical_AI_Interpretability | theme=公共卫生与医疗运营 | grade=突破级 | published=2026-05-18 | doi=10.1038/s41551-026-01676-w