Toward Ultra-Long-Horizon Agentic Science: Cognitive Accumulation for Machine Learning Engineering

The advancement of artificial intelligence toward agentic science is currently bottlenecked by the challenge of ultra-long-horizon autonomy, the ability to sustain strategic coherence and iterative correction over experimental cycles spanning days or weeks. While Large Language Models (LLMs) have demonstrated prowess in short-horizon reasoning, they are easily overwhelmed by execution details in the high-dimensional, delayed-feedback environments of real-world research, failing to consolidate sparse feedback into coherent long-term guidance. Here, we present ML-Master 2.0, an autonomous agent that masters ultra-long-horizon machine learning engineering (MLE) which is a representative microcosm of scientific discovery. By reframing context management as a process of cognitive accumulation, our approach introduces Hierarchical Cognitive Caching (HCC), a multi-tiered architecture inspired by computer systems that enables the structural differentiation of experience over time. By dynamically distilling transient execution traces into stable knowledge and cross-task wisdom, HCC allows agents to decouple immediate execution from long-term experimental strategy, effectively overcoming the scaling limits of static context windows. In evaluations on OpenAI's MLE-Bench under 24-hour budgets, ML-Master 2.0 achieves a state-of-the-art medal rate of 56.44%. Our findings demonstrate that ultra-long-horizon autonomy provides a scalable blueprint for AI capable of autonomous exploration beyond human-precedent complexities.

Bohrium + SciMaster: Building the Infrastructure and Ecosystem for Agentic Science at Scale

AI agents are emerging as a practical way to run multi-step scientific workflows that interleave reasoning with tool use and verification, pointing to a shift from isolated AI-assisted steps toward \emph{agentic science at scale}. This shift is increasingly feasible, as scientific tools and models can be invoked through stable interfaces and verified with recorded execution traces, and increasingly necessary, as AI accelerates scientific output and stresses the peer-review and publication pipeline, raising the bar for traceability and credible evaluation. However, scaling agentic science remains difficult: workflows are hard to observe and reproduce; many tools and laboratory systems are not agent-ready; execution is hard to trace and govern; and prototype AI Scientist systems are often bespoke, limiting reuse and systematic improvement from real workflow signals. We argue that scaling agentic science requires an infrastructure-and-ecosystem approach, instantiated in Bohrium+SciMaster. Bohrium acts as a managed, traceable hub for AI4S assets -- akin to a HuggingFace of AI for Science -- that turns diverse scientific data, software, compute, and laboratory systems into agent-ready capabilities. SciMaster orchestrates these capabilities into long-horizon scientific workflows, on which scientific agents can be composed and executed. Between infrastructure and orchestration, a \emph{scientific intelligence substrate} organizes reusable models, knowledge, and components into executable building blocks for workflow reasoning and action, enabling composition, auditability, and improvement through use. We demonstrate this stack with eleven representative master agents in real workflows, achieving orders-of-magnitude reductions in end-to-end scientific cycle time and generating execution-grounded signals from real workloads at multi-million scale.

PhysMaster: Building an Autonomous AI Physicist for Theoretical and Computational Physics Research

Advances in LLMs have produced agents with knowledge and operational capabilities comparable to human scientists, suggesting potential to assist, accelerate, and automate research. However, existing studies mainly evaluate such systems on well-defined benchmarks or general tasks like literature retrieval, limiting their end-to-end problem-solving ability in open scientific scenarios. This is particularly true in physics, which is abstract, mathematically intensive, and requires integrating analytical reasoning with code-based computation. To address this, we propose PhysMaster, an LLM-based agent functioning as an autonomous theoretical and computational physicist. PhysMaster couples absract reasoning with numerical computation and leverages LANDAU, the Layered Academic Data Universe, which preserves retrieved literature, curated prior knowledge, and validated methodological traces, enhancing decision reliability and stability. It also employs an adaptive exploration strategy balancing efficiency and open-ended exploration, enabling robust performance in ultra-long-horizon tasks. We evaluate PhysMaster on problems from high-energy theory, condensed matter theory to astrophysics, including: (i) acceleration, compressing labor-intensive research from months to hours; (ii) automation, autonomously executing hypothesis-driven loops ; and (iii) autonomous discovery, independently exploring open problems.

ML-Agent: Reinforcing LLM Agents for Autonomous Machine Learning Engineering

The emergence of large language model (LLM)-based agents has significantly advanced the development of autonomous machine learning (ML) engineering. However, most existing approaches rely heavily on manual prompt engineering, failing to adapt and optimize based on diverse experimental experiences. Focusing on this, for the first time, we explore the paradigm of learning-based agentic ML, where an LLM agent learns through interactive experimentation on ML tasks using online reinforcement learning (RL). To realize this, we propose a novel agentic ML training framework with three key components: (1) exploration-enriched fine-tuning, which enables LLM agents to generate diverse actions for enhanced RL exploration; (2) step-wise RL, which enables training on a single action step, accelerating experience collection and improving training efficiency; (3) an agentic ML-specific reward module, which unifies varied ML feedback signals into consistent rewards for RL optimization. Leveraging this framework, we train ML-Agent, driven by a 7B-sized Qwen-2.5 LLM for autonomous ML. Remarkably, despite being trained on merely 9 ML tasks, our 7B-sized ML-Agent outperforms the 671B-sized DeepSeek-R1 agent. Furthermore, it achieves continuous performance improvements and demonstrates exceptional cross-task generalization capabilities.

Synthesizing Post-Training Data for LLMs through Multi-Agent Simulation

Post-training is essential for enabling large language models (LLMs) to follow human instructions. Inspired by the recent success of using LLMs to simulate human society, we leverage multi-agent simulation to automatically generate diverse text-based scenarios, capturing a wide range of real-world human needs. We propose MATRIX, a multi-agent simulator that creates realistic and scalable scenarios. Leveraging these outputs, we introduce a novel scenario-driven instruction generator MATRIX-Gen for controllable and highly realistic data synthesis. Extensive experiments demonstrate that our framework effectively generates both general and domain-specific data. Notably, on AlpacaEval 2 and Arena-Hard benchmarks, Llama-3-8B-Base, post-trained on datasets synthesized by MATRIX-Gen with just 20K instruction-response pairs, outperforms Meta's Llama-3-8B-Instruct model, which was trained on over 10M pairs; see our project at https://github.com/ShuoTang123/MATRIX-Gen.

Hypergraph Node Classification With Graph Neural Networks

Hypergraphs, with hyperedges connecting more than two nodes, are key for modelling higher-order interactions in real-world data. The success of graph neural networks (GNNs) reveals the capability of neural networks to process data with pairwise interactions. This inspires the usage of neural networks for data with higher-order interactions, thereby leading to the development of hypergraph neural networks (HyperGNNs). GNNs and HyperGNNs are typically considered distinct since they are designed for data on different geometric topologies. However, in this paper, we theoretically demonstrate that, in the context of node classification, most HyperGNNs can be approximated using a GNN with a weighted clique expansion of the hypergraph. This leads to WCE-GNN, a simple and efficient framework comprising a GNN and a weighted clique expansion (WCE), for hypergraph node classification. Experiments on nine real-world hypergraph node classification benchmarks showcase that WCE-GNN demonstrates not only higher classification accuracy compared to state-of-the-art HyperGNNs, but also superior memory and runtime efficiency.

Hypergraph Transformer for Semi-Supervised Classification

Hypergraphs play a pivotal role in the modelling of data featuring higher-order relations involving more than two entities. Hypergraph neural networks emerge as a powerful tool for processing hypergraph-structured data, delivering remarkable performance across various tasks, e.g., hypergraph node classification. However, these models struggle to capture global structural information due to their reliance on local message passing. To address this challenge, we propose a novel hypergraph learning framework, HyperGraph Transformer (HyperGT). HyperGT uses a Transformer-based neural network architecture to effectively consider global correlations among all nodes and hyperedges. To incorporate local structural information, HyperGT has two distinct designs: i) a positional encoding based on the hypergraph incidence matrix, offering valuable insights into node-node and hyperedge-hyperedge interactions; and ii) a hypergraph structure regularization in the loss function, capturing connectivities between nodes and hyperedges. Through these designs, HyperGT achieves comprehensive hypergraph representation learning by effectively incorporating global interactions while preserving local connectivity patterns. Extensive experiments conducted on real-world hypergraph node classification tasks showcase that HyperGT consistently outperforms existing methods, establishing new state-of-the-art benchmarks. Ablation studies affirm the effectiveness of the individual designs of our model.