We present a multi-agent system for automation of scientific research tasks, cmbagent. The system is formed by about 30 Large Language Model (LLM) agents and implements a Planning & Control strategy to orchestrate the agentic workflow, with no human-in-the-loop at any point. Each agent specializes in a different task (performing retrieval on scientific papers and codebases, writing code, interpreting results, critiquing the output of other agents) and the system is able to execute code locally. We successfully apply cmbagent to carry out a PhD level cosmology task (the measurement of cosmological parameters using supernova data) and evaluate its performance on two benchmark sets, finding superior performance over state-of-the-art LLMs. The source code is available on GitHub, demonstration videos are also available, and the system is deployed on HuggingFace and will be available on the cloud.
Link prediction infers missing or future relations between graph nodes, based on connection patterns. Scientific literature networks and knowledge graphs are typically large, sparse, and noisy, and often contain missing links between entities. We present an AI-driven hierarchical link prediction framework that integrates matrix factorization to infer hidden associations and steer discovery in complex material domains. Our method combines Hierarchical Nonnegative Matrix Factorization (HNMFk) and Boolean matrix factorization (BNMFk) with automatic model selection, as well as Logistic matrix factorization (LMF), we use to construct a three-level topic tree from a 46,862-document corpus focused on 73 transition-metal dichalcogenides (TMDs). These materials are studied in a variety of physics fields with many current and potential applications. An ensemble BNMFk + LMF approach fuses discrete interpretability with probabilistic scoring. The resulting HNMFk clusters map each material onto coherent topics like superconductivity, energy storage, and tribology. Also, missing or weakly connected links are highlight between topics and materials, suggesting novel hypotheses for cross-disciplinary exploration. We validate our method by removing publications about superconductivity in well-known superconductors, and show the model predicts associations with the superconducting TMD clusters. This shows the method finds hidden connections in a graph of material to latent topic associations built from scientific literature, especially useful when examining a diverse corpus of scientific documents covering the same class of phenomena or materials but originating from distinct communities and perspectives. The inferred links generating new hypotheses, produced by our method, are exposed through an interactive Streamlit dashboard, designed for human-in-the-loop scientific discovery.
The ability to extract structured information from unstructured sources-such as free-text documents and scientific literature-is critical for accelerating scientific discovery and knowledge synthesis. Large Language Models (LLMs) have demonstrated remarkable capabilities in various natural language processing tasks, including structured information extraction. However, their effectiveness often diminishes in specialized, domain-specific contexts that require nuanced understanding and expert-level domain knowledge. In addition, existing LLM-based approaches frequently exhibit poor transferability across tasks and domains, limiting their scalability and adaptability. To address these challenges, we introduce StructSense, a modular, task-agnostic, open-source framework for structured information extraction built on LLMs. StructSense is guided by domain-specific symbolic knowledge encoded in ontologies, enabling it to navigate complex domain content more effectively. It further incorporates agentic capabilities through self-evaluative judges that form a feedback loop for iterative refinement, and includes human-in-the-loop mechanisms to ensure quality and validation. We demonstrate that StructSense can overcome both the limitations of domain sensitivity and the lack of cross-task generalizability, as shown through its application to diverse neuroscience information extraction tasks.
Vision-based scientific foundation models hold significant promise for advancing scientific discovery and innovation. This potential stems from their ability to aggregate images from diverse sources such as varying physical groundings or data acquisition systems and to learn spatio-temporal correlations using transformer architectures. However, tokenizing and aggregating images can be compute-intensive, a challenge not fully addressed by current distributed methods. In this work, we introduce the Distributed Cross-Channel Hierarchical Aggregation (D-CHAG) approach designed for datasets with a large number of channels across image modalities. Our method is compatible with any model-parallel strategy and any type of vision transformer architecture, significantly improving computational efficiency. We evaluated D-CHAG on hyperspectral imaging and weather forecasting tasks. When integrated with tensor parallelism and model sharding, our approach achieved up to a 75% reduction in memory usage and more than doubled sustained throughput on up to 1,024 AMD GPUs on the Frontier Supercomputer.
Synthetic data has emerged as a powerful resource in life sciences, offering solutions for data scarcity, privacy protection and accessibility constraints. By creating artificial datasets that mirror the characteristics of real data, allows researchers to develop and validate computational methods in controlled environments. Despite its promise, the adoption of synthetic data in Life Sciences hinges on rigorous evaluation metrics designed to assess their fidelity and reliability. To explore the current landscape of synthetic data evaluation metrics in several Life Sciences domains, the ELIXIR Machine Learning Focus Group performed a systematic review of the scientific literature following the PRISMA guidelines. Six critical domains were examined to identify current practices for assessing synthetic data. Findings reveal that, while generation methods are rapidly evolving, systematic evaluation is often overlooked, limiting researchers ability to compare, validate, and trust synthetic datasets across different domains. This systematic review underscores the urgent need for robust, standardized evaluation approaches that not only bolster confidence in synthetic data but also guide its effective and responsible implementation. By laying the groundwork for establishing domain-specific yet interoperable standards, this scoping review paves the way for future initiatives aimed at enhancing the role of synthetic data in scientific discovery, clinical practice and beyond.
The rapid evolution of artificial intelligence has led to expectations of transformative scientific discovery, yet current systems remain fundamentally limited by their operational architectures, brittle reasoning mechanisms, and their separation from experimental reality. Building on earlier work, we contend that progress in AI-driven science now depends on closing three fundamental gaps -- the abstraction gap, the reasoning gap, and the reality gap -- rather than on model size/data/test time compute. Scientific reasoning demands internal representations that support simulation of actions and response, causal structures that distinguish correlation from mechanism, and continuous calibration. We define active inference AI systems for scientific discovery as those that (i) maintain long-lived research memories grounded in causal self-supervised foundation models, (ii) symbolic or neuro-symbolic planners equipped with Bayesian guardrails, (iii) grow persistent knowledge graphs where thinking generates novel conceptual nodes, reasoning establishes causal edges, and real-world interaction prunes false connections while strengthening verified pathways, and (iv) refine their internal representations through closed-loop interaction with both high-fidelity simulators and automated laboratories - an operational loop where mental simulation guides action and empirical surprise reshapes understanding. In essence, we outline an architecture where discovery arises from the interplay between internal models that enable counterfactual reasoning and external validation that grounds hypotheses in reality. It is also argued that the inherent ambiguity in feedback from simulations and experiments, and underlying uncertainties makes human judgment indispensable, not as a temporary scaffold but as a permanent architectural component.
Cellular automata (CA) provide a minimal formalism for investigating how simple local interactions generate rich spatiotemporal behavior in domains as diverse as traffic flow, ecology, tissue morphogenesis and crystal growth. However, automatically discovering the local update rules for a given phenomenon and using them for quantitative prediction remains challenging. Here we present AutomataGPT, a decoder-only transformer pretrained on around 1 million simulated trajectories that span 100 distinct two-dimensional binary deterministic CA rules on toroidal grids. When evaluated on previously unseen rules drawn from the same CA family, AutomataGPT attains 98.5% perfect one-step forecasts and reconstructs the governing update rule with up to 96% functional (application) accuracy and 82% exact rule-matrix match. These results demonstrate that large-scale pretraining over wider regions of rule space yields substantial generalization in both the forward (state forecasting) and inverse (rule inference) problems, without hand-crafted priors. By showing that transformer models can faithfully infer and execute CA dynamics from data alone, our work lays the groundwork for abstracting real-world dynamical phenomena into data-efficient CA surrogates, opening avenues in biology, tissue engineering, physics and AI-driven scientific discovery.
Large language model (LLM) agents have demonstrated remarkable potential in advancing scientific discovery. However, their capability in the fundamental yet crucial task of reproducing code from research papers, especially in the NLP domain, remains underexplored. This task includes unique complex reasoning challenges in the intellectual synthesis of abstract concepts and the comprehension of code repositories with interdependent files. Motivated by this gap, we present LMR-BENCH, a benchmark designed to systematically evaluate the capability of LLM agents on code reproduction from Language Modeling Research. It consists of 28 code reproduction tasks derived from 23 research papers published in top-tier NLP venues over the past five years, spanning nine fundamental categories. Models are provided with a research paper, a code repository containing one or more masked functions, and instructions for implementing these functions. We conduct extensive experiments in standard prompting and LLM agent settings with state-of-the-art LLMs, evaluating the accuracy of unit tests and performing LLM-based evaluation of code correctness. Experimental results reveal that even the most advanced models still exhibit persistent limitations in scientific reasoning and code synthesis, highlighting critical gaps in LLM agents' ability to autonomously reproduce scientific research
Computer graphics, often associated with films, games, and visual effects, has long been a powerful tool for addressing scientific challenges--from its origins in 3D visualization for medical imaging to its role in modern computational modeling and simulation. This course explores the deep and evolving relationship between computer graphics and science, highlighting past achievements, ongoing contributions, and open questions that remain. We show how core methods, such as geometric reasoning and physical modeling, provide inductive biases that help address challenges in both fields, especially in data-scarce settings. To that end, we aim to reframe graphics as a modeling language for science by bridging vocabulary gaps between the two communities. Designed for both newcomers and experts, Graphics4Science invites the graphics community to engage with science, tackle high-impact problems where graphics expertise can make a difference, and contribute to the future of scientific discovery. Additional details are available on the course website: https://graphics4science.github.io
Foundation models (FMs) are catalyzing a transformative shift in materials science (MatSci) by enabling scalable, general-purpose, and multimodal AI systems for scientific discovery. Unlike traditional machine learning models, which are typically narrow in scope and require task-specific engineering, FMs offer cross-domain generalization and exhibit emergent capabilities. Their versatility is especially well-suited to materials science, where research challenges span diverse data types and scales. This survey provides a comprehensive overview of foundation models, agentic systems, datasets, and computational tools supporting this growing field. We introduce a task-driven taxonomy encompassing six broad application areas: data extraction, interpretation and Q\&A; atomistic simulation; property prediction; materials structure, design and discovery; process planning, discovery, and optimization; and multiscale modeling. We discuss recent advances in both unimodal and multimodal FMs, as well as emerging large language model (LLM) agents. Furthermore, we review standardized datasets, open-source tools, and autonomous experimental platforms that collectively fuel the development and integration of FMs into research workflows. We assess the early successes of foundation models and identify persistent limitations, including challenges in generalizability, interpretability, data imbalance, safety concerns, and limited multimodal fusion. Finally, we articulate future research directions centered on scalable pretraining, continual learning, data governance, and trustworthiness.
Interpretable machine learning is rapidly becoming a crucial tool for scientific discovery. Among existing approaches, variational autoencoders (VAEs) have shown promise in extracting the hidden physical features of some input data, with no supervision nor prior knowledge of the system at study. Yet, the ability of VAEs to create meaningful, interpretable representations relies on their accurate approximation of the underlying probability distribution of their input. When dealing with quantum data, VAEs must hence account for its intrinsic randomness and complex correlations. While VAEs have been previously applied to quantum data, they have often neglected its probabilistic nature, hindering the extraction of meaningful physical descriptors. Here, we demonstrate that two key modifications enable VAEs to learn physically meaningful latent representations: a decoder capable of faithfully reproduce quantum states and a probabilistic loss tailored to this task. Using benchmark quantum spin models, we identify regimes where standard methods fail while the representations learned by our approach remain meaningful and interpretable. Applied to experimental data from Rydberg atom arrays, the model autonomously uncovers the phase structure without access to prior labels, Hamiltonian details, or knowledge of relevant order parameters, highlighting its potential as an unsupervised and interpretable tool for the study of quantum systems.
Recent progress in autonomous code generation has fueled excitement around AI agents capable of accelerating scientific discovery by running experiments. However, there is currently no benchmark that evaluates whether such agents can implement scientific ideas when given varied amounts of code as a starting point, interpolating between reproduction (running code) and from-scratch replication (fully re-implementing and running code). We introduce AutoExperiment, a benchmark that evaluates AI agents' ability to implement and run machine learning experiments based on natural language descriptions in research papers. In each task, agents are given a research paper, a codebase with key functions masked out, and a command to run the experiment. The goal is to generate the missing code, execute the experiment in a sandboxed environment, and reproduce the results. AutoExperiment scales in difficulty by varying the number of missing functions $n$, ranging from partial reproduction to full replication. We evaluate state-of-the-art agents and find that performance degrades rapidly as $n$ increases. Agents that can dynamically interact with the environment (e.g. to debug their code) can outperform agents in fixed "agentless" harnesses, and there exists a significant gap between single-shot and multi-trial success rates (Pass@1 vs. Pass@5), motivating verifier approaches to our benchmark. Our findings highlight critical challenges in long-horizon code generation, context retrieval, and autonomous experiment execution, establishing AutoExperiment as a new benchmark for evaluating progress in AI-driven scientific experimentation. Our data and code are open-sourced at https://github.com/j1mk1m/AutoExperiment .
Generative AI, especially through large language models (LLMs), is transforming how technical knowledge can be accessed, reused, and extended. PETSc, a widely used numerical library for high-performance scientific computing, has accumulated a rich but fragmented knowledge base over its three decades of development, spanning source code, documentation, mailing lists, GitLab issues, Discord conversations, technical papers, and more. Much of this knowledge remains informal and inaccessible to users and new developers. To activate and utilize this knowledge base more effectively, the PETSc team has begun building an LLM-powered system that combines PETSc content with custom LLM tools -- including retrieval-augmented generation (RAG), reranking algorithms, and chatbots -- to assist users, support developers, and propose updates to formal documentation. This paper presents initial experiences designing and evaluating these tools, focusing on system architecture, using RAG and reranking for PETSc-specific information, evaluation methodologies for various LLMs and embedding models, and user interface design. Leveraging the Argonne Leadership Computing Facility resources, we analyze how LLM responses can enhance the development and use of numerical software, with an initial focus on scalable Krylov solvers. Our goal is to establish an extensible framework for knowledge-centered AI in scientific software, enabling scalable support, enriched documentation, and enhanced workflows for research and development. We conclude by outlining directions for expanding this system into a robust, evolving platform that advances software ecosystems to accelerate scientific discovery.
Neural networks are a powerful tool for learning patterns from data. However, they do not respect known scientific laws, nor can they reveal novel scientific insights due to their black-box nature. In contrast, scientific reasoning distills biological or physical principles from observations and controlled experiments, and quantitatively interprets them with process-based models made of mathematical equations. Yet, process-based models rely on numerous free parameters that must be set in an ad-hoc manner, and thus often fit observations poorly in cross-scale predictions. While prior work has embedded process-based models in conventional neural networks, discovering interpretable relationships between parameters in process-based models and input features is still a grand challenge for scientific discovery. We thus propose Scientifically-Interpretable Reasoning Network (ScIReN), a fully-transparent framework that combines interpretable neural and process-based reasoning. An interpretable encoder predicts scientifically-meaningful latent parameters, which are then passed through a differentiable process-based decoder to predict labeled output variables. ScIReN also uses a novel hard-sigmoid constraint layer to restrict latent parameters to meaningful ranges defined by scientific prior knowledge, further enhancing its interpretability. While the embedded process-based model enforces established scientific knowledge, the encoder reveals new scientific mechanisms and relationships hidden in conventional black-box models. We apply ScIReN on two tasks: simulating the flow of organic carbon through soils, and modeling ecosystem respiration from plants. In both tasks, ScIReN outperforms black-box networks in predictive accuracy while providing substantial scientific interpretability -- it can infer latent scientific mechanisms and their relationships with input features.
This paper presents an evolvable conditional diffusion method such that black-box, non-differentiable multi-physics models, as are common in domains like computational fluid dynamics and electromagnetics, can be effectively used for guiding the generative process to facilitate autonomous scientific discovery. We formulate the guidance as an optimization problem where one optimizes for a desired fitness function through updates to the descriptive statistic for the denoising distribution, and derive an evolution-guided approach from first principles through the lens of probabilistic evolution. Interestingly, the final derived update algorithm is analogous to the update as per common gradient-based guided diffusion models, but without ever having to compute any derivatives. We validate our proposed evolvable diffusion algorithm in two AI for Science scenarios: the automated design of fluidic topology and meta-surface. Results demonstrate that this method effectively generates designs that better satisfy specific optimization objectives without reliance on differentiable proxies, providing an effective means of guidance-based diffusion that can capitalize on the wealth of black-box, non-differentiable multi-physics numerical models common across Science.
As the issue of global climate change becomes increasingly severe, the demand for research in climate science continues to grow. Natural language processing technologies, represented by Large Language Models (LLMs), have been widely applied to climate change-specific research, providing essential information support for decision-makers and the public. Some studies have improved model performance on relevant tasks by constructing climate change-related instruction data and instruction-tuning LLMs. However, current research remains inadequate in efficiently producing large volumes of high-precision instruction data for climate change, which limits further development of climate change LLMs. This study introduces an automated method for constructing instruction data. The method generates instructions using facts and background knowledge from documents and enhances the diversity of the instruction data through web scraping and the collection of seed instructions. Using this method, we constructed a climate change instruction dataset, named ClimateChat-Corpus, which was used to fine-tune open-source LLMs, resulting in an LLM named ClimateChat. Evaluation results show that ClimateChat significantly improves performance on climate change question-and-answer tasks. Additionally, we evaluated the impact of different base models and instruction data on LLM performance and demonstrated its capability to adapt to a wide range of climate change scientific discovery tasks, emphasizing the importance of selecting an appropriate base model for instruction tuning. This research provides valuable references and empirical support for constructing climate change instruction data and training climate change-specific LLMs.
Large Language Models (LLMs) are rapidly advancing across diverse domains, yet their application in theoretical physics research is not yet mature. This position paper argues that LLM agents can potentially help accelerate theoretical, computational, and applied physics when properly integrated with domain knowledge and toolbox. We analyze current LLM capabilities for physics -- from mathematical reasoning to code generation -- identifying critical gaps in physical intuition, constraint satisfaction, and reliable reasoning. We envision future physics-specialized LLMs that could handle multimodal data, propose testable hypotheses, and design experiments. Realizing this vision requires addressing fundamental challenges: ensuring physical consistency, and developing robust verification methods. We call for collaborative efforts between physics and AI communities to help advance scientific discovery in physics.
Kolmogorov-Arnold Networks (KANs) uniquely combine high accuracy with interpretability, making them valuable for scientific modeling. However, it is unclear a priori how deep a network needs to be for any given task, and deeper KANs can be difficult to optimize. Here we introduce multi-exit KANs, where each layer includes its own prediction branch, enabling the network to make accurate predictions at multiple depths simultaneously. This architecture provides deep supervision that improves training while discovering the right level of model complexity for each task. Multi-exit KANs consistently outperform standard, single-exit versions on synthetic functions, dynamical systems, and real-world datasets. Remarkably, the best predictions often come from earlier, simpler exits, revealing that these networks naturally identify smaller, more parsimonious and interpretable models without sacrificing accuracy. To automate this discovery, we develop a differentiable "learning to exit" algorithm that balances contributions from exits during training. Our approach offers scientists a practical way to achieve both high performance and interpretability, addressing a fundamental challenge in machine learning for scientific discovery.
Many critical applications, from autonomous scientific discovery to personalized medicine, demand systems that can both strategically acquire the most informative data and instantaneously perform inference based upon it. While amortized methods for Bayesian inference and experimental design offer part of the solution, neither approach is optimal in the most general and challenging task, where new data needs to be collected for instant inference. To tackle this issue, we introduce the Amortized Active Learning and Inference Engine (ALINE), a unified framework for amortized Bayesian inference and active data acquisition. ALINE leverages a transformer architecture trained via reinforcement learning with a reward based on self-estimated information gain provided by its own integrated inference component. This allows it to strategically query informative data points while simultaneously refining its predictions. Moreover, ALINE can selectively direct its querying strategy towards specific subsets of model parameters or designated predictive tasks, optimizing for posterior estimation, data prediction, or a mixture thereof. Empirical results on regression-based active learning, classical Bayesian experimental design benchmarks, and a psychometric model with selectively targeted parameters demonstrate that ALINE delivers both instant and accurate inference along with efficient selection of informative points.
Despite long-standing efforts in accelerating scientific discovery with AI, building AI co-scientists remains challenging due to limited high-quality data for training and evaluation. To tackle this data scarcity issue, we present AutoSDT, an automatic pipeline that collects high-quality coding tasks in real-world data-driven discovery workflows. AutoSDT leverages the coding capabilities and parametric knowledge of LLMs to search for diverse sources, select ecologically valid tasks, and synthesize accurate task instructions and code solutions. Using our pipeline, we construct AutoSDT-5K, a dataset of 5,404 coding tasks for data-driven discovery that covers four scientific disciplines and 756 unique Python packages. To the best of our knowledge, AutoSDT-5K is the only automatically collected and the largest open dataset for data-driven scientific discovery. Expert feedback on a subset of 256 tasks shows the effectiveness of AutoSDT: 93% of the collected tasks are ecologically valid, and 92.2% of the synthesized programs are functionally correct. Trained on AutoSDT-5K, the Qwen2.5-Coder-Instruct LLM series, dubbed AutoSDT-Coder, show substantial improvement on two challenging data-driven discovery benchmarks, ScienceAgentBench and DiscoveryBench. Most notably, AutoSDT-Coder-32B reaches the same level of performance as GPT-4o on ScienceAgentBench with a success rate of 7.8%, doubling the performance of its base model. On DiscoveryBench, it lifts the hypothesis matching score to 8.1, bringing a 17.4% relative improvement and closing the gap between open-weight models and GPT-4o.