Learning GenAI via SOTA Papers

Title: MGDT: MLLM-Guided Diffusion Transformer with Relation-Adaptive Mixture-of-Experts for Multimodal Knowledge Graph Completion Source: http://arxiv.org/abs/2607.15592v1 Summary: This paper proposes a new architectural primitive for multimodal generative AI by combining MLLMs, Diffusion Transformers, and a Relation-Adaptive Mixture-of-Experts. This innovative architecture offers a foundational...

Title: TopoAgent: A Self-Evolving Topological Agent for Multimodal Scientific Reasoning Source: http://arxiv.org/abs/2607.14658v1 Summary: TopoAgent proposes a self-evolving topological agent framework, introducing novel agentic reasoning loops that leverage topological structures for multimodal scientific reasoning. This represents a foundational advancement in agent design, enabling more adaptiv...

Title: T^2MLR: Transformer with Temporal Middle-Layer Recurrence Source: http://arxiv.org/abs/2607.15178v1 Summary: This paper introduces T^2MLR, a novel Transformer architecture incorporating temporal recurrence directly within its middle layers. This design represents a new architectural primitive, enhancing Transformers' ability to process sequential information and potentially leading to s...

Title: Multi-Agent Collaborative Reasoning with Tool-Augmented Evidence for Urban Region Profiling Source: http://arxiv.org/abs/2607.13558v1 Summary: This paper introduces a novel framework for multi-agent collaborative reasoning, significantly advancing Agentic AI by enabling agents to work together and leverage external tools. This approach is foundational for developing sophisticated AI agents...

Title: Groc-PO: Grounded Context Preference Optimization for Truthful Multimodal LLMs Source: http://arxiv.org/abs/2607.13712v1 Summary: This work proposes a new preference optimization method, Groc-PO, specifically designed to enhance the truthfulness and grounding of Multimodal Large Language Models. This represents a significant breakthrough in improving the reliability and reducing hallucinati...

Title: Track, Rank, Crack: Epistemic Working Memory Scales Multi-Hop Reasoning in Language Agents Source: http://arxiv.org/abs/2607.12267v1 Summary: This work introduces a novel agentic reasoning framework centered on 'Epistemic Working Memory' and a 'Track, Rank, Crack' mechanism. It provides a significant breakthrough in scaling multi-hop reasoning capabilities, which is crucial...

Title: Atomic Units of X: The Compression Layer of Intelligence Source: http://arxiv.org/abs/2607.12634v1 Summary: This paper proposes a foundational concept of 'Atomic Units of X' as a compression layer, which could introduce new architectural primitives for intelligence itself. Such a fundamental theoretical framework has the potential to underpin the design of both future Generative AI...

Title: A Hybrid Mamba for Audio-Visual Navigation Source: http://arxiv.org/abs/2607.13110v1 Summary: This paper proposes a 'Hybrid Mamba' architecture, establishing a new architectural primitive or a significant evolution of the Mamba architecture for GenAI. This innovation promises substantial efficiency and capability gains across various domains, exemplified by its application to comple...

Title: Optimization Is Not All You Need Source: http://arxiv.org/abs/2607.11977v2 Summary: This work proposes a paradigm shift beyond traditional optimization-centric approaches in AI, suggesting new fundamental principles for learning and intelligence. By challenging a core tenet of modern AI, it could unlock significant reasoning and efficiency breakthroughs, impacting the very foundation of how...

Title: Mako: A Self-Evolving Agentic Operating System (SE-AOS) for Autonomous Web Exploitation Source: http://arxiv.org/abs/2607.11288v1 Summary: This work proposes a groundbreaking self-evolving agentic operating system (SE-AOS), presenting a new architectural primitive for agentic AI. This paradigm allows agents to autonomously manage and improve their foundational operational environment, signi...

Title: Agentic Routing: The Harness-Native Data Flywheel Source: http://arxiv.org/abs/2607.11399v1 Summary: This paper introduces a foundational step-level routing paradigm that treats multi-model coordination within agent execution harnesses as a dynamic, state-conditioned systems problem. It establishes a harness-native data flywheel where execution traces and environmental feedback automaticall...

Title: A Formal Hierarchical Architecture for Agentic Orchestration with Stack-Based Execution and Lazy Discovery Source: http://arxiv.org/abs/2607.11138v1 Summary: This paper presents a novel hierarchical, stack-based execution architecture for agentic orchestration that models complex, nested agent states as Pushdown Automata. By substituting flat tool registries with localized stack frames and...

Title: LOGOS: A Living Logic for AI Agent Teams That Evolve With Humans Source: http://arxiv.org/abs/2607.10878v1 Summary: LOGOS introduces a pluggable layer for self-evolution and governance within multi-agent frameworks, enabling verifiable human-agent loop engineering. This framework is foundational for controlling, auditing, and safely evolving AI agent teams at scale, addressing critical ques...

Title: Agentic-DPO: From Imitation to Agentic Policy Optimization on Expert Trajectories Source: http://arxiv.org/abs/2607.10601v1 Summary: Agentic-DPO proposes a lightweight offline policy optimization method that transforms expert trajectories into state-conditioned preference supervision for DPO-style training. This approach represents a significant breakthrough in efficiently training robust L...

Title: Beyond Euclidean Clipping: Overcoming Exploration Collapse in LLM RL via Riemannian Isometric Policy Optimization Source: http://arxiv.org/abs/2607.10169v1 Summary: This paper introduces Riemannian Isometric Policy Optimization (RIPO), a foundational mathematical advancement that corrects the geometric mismatch in standard PPO-Clip's Euclidean-based policy distance measurements. By ensu...

Title: ARMOR: Stabilizing On-Policy LLM RL with Off-Policy Anchor Samples Source: http://arxiv.org/abs/2607.10481v1 Summary: This paper presents ARMOR, a novel post-training reinforcement learning framework that addresses the persistent issue of training instability and over-optimization in reasoning LLMs. By introducing active anchor rollouts from reference policies paired with a mixed optimizati...

Title: Shared Selective Persistent Memory for Agentic LLM Systems Source: http://arxiv.org/abs/2607.09493v1 Summary: This work introduces a foundational shared selective persistent memory architecture that solves the critical context-bloat and token-inefficiency bottlenecks in multi-session LLM agents. By isolating reusable context parameters and decoupling runtime data, the framework achieves dra...

Title: A Sovereign, Open-Source Foundation Model for German and English Source: http://arxiv.org/abs/2607.09424v1 Summary: This paper presents a novel hybrid MoE Mamba-Transformer architecture at scale, representing a significant architectural milestone for foundation model design. The hybrid model maintains a near-constant inference cache during context growth, demonstrating a major throughput an...

Title: Compete Then Collaborate: Frontier AI Teachers Build a Verifiable Curriculum to Improve a Coding Student Beyond Imitation Source: http://arxiv.org/abs/2607.08255v1 Summary: This work delivers a foundational breakthrough in LLM post-training by demonstrating that traditional supervised fine-tuning on teacher-generated data can degrade capable student models, whereas reinforcement learning wi...

Title: Cognitive-structured Multimodal Agent for Multimodal Understanding, Generation, and Editing Source: http://arxiv.org/abs/2607.08497v1 Summary: This paper introduces a novel architectural framework for multimodal agents that replaces monolithic context-window scaling with an Episodic Visual Memory and selective reactivation system to prevent visual token explosion. By decoupling perception,...

Title: RL Post-Training Builds Compositional Reasoning Strategies Source: http://arxiv.org/abs/2607.07646v1 Summary: This paper provides foundational insights into how reinforcement learning post-training enables models to transition from simple skills to complex, multi-step compositional reasoning strategies. By demonstrating how RL systematically organizes and compresses primitive actions into s...

Title: From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents Source: http://arxiv.org/abs/2607.07321v1 Summary: This work introduces a novel framework for self-evolving agents by enabling them to autonomously synthesize low-level, atomic actions into reusable, higher-order Standard Operating Procedures (SOPs). This dynamic tool-optimization...

Title: Information Gain-based Rollout Policy Optimization: An Adaptive Tree-Structured Rollout Approach for Multi-Turn LLM Agents Source: http://arxiv.org/abs/2607.06223v1 Summary: This paper introduces a novel policy optimization framework (IGRPO) that dynamically allocates rollout budgets based on node-level information gain during tree-structured exploration. By unifying adaptive search-tree ex...

Title: Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade Source: http://arxiv.org/abs/2607.06503v1 Summary: This paper introduces a novel framework that uses internal activation probes to detect and early-abort doomed agent trajectories, saving up to 47% of inference compute. This represents a significant efficiency breakthrough for agentic loops, addre...

Title: Weak-to-Strong Generalization via Direct On-Policy Distillation Source: http://arxiv.org/abs/2607.05394v1 Summary: This work introduces a novel post-training paradigm that transfers the reinforcement learning policy shift of a smaller, cheaper weak model as an implicit reward signal to a stronger target model. By bypassing the need for explicit reward modeling or expensive on-policy rollout...