P. Melchior

Matt L. Wiemann, Lindsay M. Smith, Peter Melchior et al.

Frontier LLMs now perform strongly across a wide range of physics evaluations, but it is hard to disentangle genuine reasoning from recall of established science. We introduce DiscoverPhysics, an interactive benchmark that asks a LLM agent to discover the laws of motion of a simulated world whose physics deliberately deviates from our own. We construct 22 worlds governed by, among others, screened and fractional-power gravity, multi-species couplings, hidden dark-matter-like particles, non-coordinate-free physics, and time-varying interactions. Each world is generated on demand by an N-body simulator, for which the agent proposes several rounds of experiments, observes raw trajectory data, and ultimately submits both a natural-language explanation of the world's physics and a Python implementation of the inferred law. Because solving a world requires the agent to design informative experiments and revise its hypotheses, the benchmark probes long-horizon reasoning over an experimental history. We evaluate submissions along two complementary axes: trajectory MSE on held-out particles and an LLM-judged explanation score following an expert-written rubric assessing conceptual understanding of each world. Across eleven frontier models, we find that the strongest agents pass only half of the worlds and consistently fail on those where latent structure must be uncovered. Open-source models lag substantially behind commercial models, both in their ability to design informative experiments and in extracting conclusions from the data. We further find that good predictive accuracy does not guarantee high explanation quality and that conceptual understanding depends on hypothesis refinement through well-chosen experiments.

Christian Kragh Jespersen, Miles Cranmer, Peter Melchior et al. · cambridge, princeton

Efficiently mapping baryonic properties onto dark matter is a major challenge in astrophysics. Although semi-analytic models (SAMs) and hydrodynamical simulations have made impressive advances in reproducing galaxy observables across cosmologically significant volumes, these methods still require significant computation times, representing a barrier to many applications. Graph Neural Networks (GNNs) have recently proven to be the natural choice for learning physical relations. Among the most inherently graph-like structures found in astrophysics are the dark matter merger trees that encode the evolution of dark matter halos. In this paper we introduce a new, graph-based emulator framework, $\texttt{Mangrove}$, and show that it emulates the galactic stellar mass, cold gas mass and metallicity, instantaneous and time-averaged star formation rate, and black hole mass -- as predicted by a SAM -- with root mean squared error up to two times lower than other methods across a $(75 Mpc/h)^3$ simulation box in 40 seconds, 4 orders of magnitude faster than the SAM. We show that $\texttt{Mangrove}$ allows for quantification of the dependence of galaxy properties on merger history. We compare our results to the current state of the art in the field and show significant improvements for all target properties. $\texttt{Mangrove}$ is publicly available.

D. Huppenkothen, M. Ntampaka, M. Ho et al. · princeton

Machine learning has rapidly become a tool of choice for the astronomical community. It is being applied across a wide range of wavelengths and problems, from the classification of transients to neural network emulators of cosmological simulations, and is shifting paradigms about how we generate and report scientific results. At the same time, this class of method comes with its own set of best practices, challenges, and drawbacks, which, at present, are often reported on incompletely in the astrophysical literature. With this paper, we aim to provide a primer to the astronomical community, including authors, reviewers, and editors, on how to implement machine learning models and report their results in a way that ensures the accuracy of the results, reproducibility of the findings, and usefulness of the method.

Jeff Shen, Peter Melchior

Machine learning techniques can automatically identify outliers in massive datasets, much faster and more reproducible than human inspection ever could. But finding such outliers immediately leads to the question: which features render this input anomalous? We propose a new feature attribution method, Inverse Multiscale Occlusion, that is specifically designed for outliers, for which we have little knowledge of the type of features we want to identify and expect that the model performance is questionable because anomalous test data likely exceed the limits of the training data. We demonstrate our method on outliers detected in galaxy spectra from the Dark Energy Survey Instrument and find its results to be much more interpretable than alternative attribution approaches.

Matt L. Sampson, Peter Melchior

Latent ODE models provide flexible descriptions of dynamic systems, but they can struggle with extrapolation and predicting complicated non-linear dynamics. The latent ODE approach implicitly relies on encoders to identify unknown system parameters and initial conditions, whereas the evaluation times are known and directly provided to the ODE solver. This dichotomy can be exploited by encouraging time-independent latent representations. By replacing the common variational penalty in latent space with an $\ell_2$ penalty on the path length of each system, the models learn data representations that can easily be distinguished from those of systems with different configurations. This results in faster training, smaller models, more accurate interpolation and long-time extrapolation compared to the baseline ODE models with GRU, RNN, and LSTM encoder/decoders on tests with damped harmonic oscillator, self-gravitating fluid, and predator-prey systems. We also demonstrate superior results for simulation-based inference of the Lotka-Volterra parameters and initial conditions by using the latents as data summaries for a conditional normalizing flow. Our change to the training loss is agnostic to the specific recognition network used by the decoder and can therefore easily be adopted by other latent ODE models.

Angelina Yan, Matt L. Sampson, Peter Melchior

12-lead ECGs with high sampling frequency are the clinical gold standard for arrhythmia detection, but their short-term, spot-check nature often misses intermittent events. Wearable ECGs enable long-term monitoring but suffer from irregular, lower sampling frequencies due to battery constraints, making morphology analysis challenging. We present an end-to-end classification pipeline to address these issues. We train a latent ODE to model continuous ECG waveforms and create robust feature vectors from high-frequency single-channel signals. We construct three latent vectors per waveform via downsampling the initial 360 Hz ECG to 90 Hz and 45 Hz. We then use a gradient boosted tree to classify these vectors and test robustness across frequencies. Performance shows minimal degradation, with macro-averaged AUC-ROC values of 0.984, 0.978, and 0.976 at 360 Hz, 90 Hz, and 45 Hz, respectively, suggesting a way to sidestep the trade-off between signal fidelity and battery life. This enables smaller wearables, promoting long-term monitoring of cardiac health.

Veena Krishnaraj, Adrian E. Bayer, Christian Kragh Jespersen et al.

Machine learning enables powerful cosmological inference but typically requires many high-fidelity simulations covering many cosmological models. Transfer learning offers a way to reduce the simulation cost by reusing knowledge across models. We show that pre-training on the standard model of cosmology, $Λ$CDM, and fine-tuning on various beyond-$Λ$CDM scenarios -- including massive neutrinos, modified gravity, and primordial non-Gaussianities -- can enable inference with significantly fewer beyond-$Λ$CDM simulations. However, we also show that negative transfer can occur when strong physical degeneracies exist between $Λ$CDM and beyond-$Λ$CDM parameters. We consider various transfer architectures, finding that including bottleneck structures provides the best performance. Our findings illustrate the opportunities and pitfalls of foundation-model approaches in physics: pre-training can accelerate inference, but may also hinder learning new physics.

Christian K. Jespersen, Peter Melchior, David N. Spergel et al.

Galaxies are often modelled as composites of separable components with distinct spectral signatures, implying that different wavelength ranges are only weakly correlated. They are not. We present a data-driven model which exploits subtle correlations between physical processes to accurately predict infrared (IR) WISE photometry from a neural summary of optical SDSS spectra. The model achieves accuracies of $χ^2_N \approx 1$ for all photometric bands in WISE, as well as good colors. We are also able to tightly constrain typically IR-derived properties, e.g. the bolometric luminosities of AGN and dust parameters such as $\mathrm{q_{PAH}}$. We find that current SED-fitting methods are incapable of making comparable predictions, and that model misspecification often leads to correlated biases in star-formation rates and AGN luminosities. To help improve SED models, we determine what features of the optical spectrum are responsible for our improved predictions, and identify several lines (CaII, SrII, FeI, [OII] and H$α$), which point to the complex chronology of star formation and chemical enrichment being incorrectly modelled.

Tianshu Wang, Peter Melchior

Resource allocation problems are often approached with linear programming techniques. But many concrete allocation problems in the experimental and observational sciences cannot or should not be expressed in the form of linear objective functions. Even if the objective is linear, its parameters may not be known beforehand because they depend on the results of the experiment for which the allocation is to be determined. To address these challenges, we present a bipartite Graph Neural Network architecture for trainable resource allocation strategies. Items of value and constraints form the two sets of graph nodes, which are connected by edges corresponding to possible allocations. The GNN is trained on simulations or past problem occurrences to maximize any user-supplied, scientifically motivated objective function, augmented by an infeasibility penalty. The amount of feasibility violation can be tuned in relation to any available slack in the system. We apply this method to optimize the astronomical target selection strategy for the highly multiplexed Subaru Prime Focus Spectrograph instrument, where it shows superior results to direct gradient descent optimization and extends the capabilities of the currently employed solver which uses linear objective functions. The development of this method enables fast adjustment and deployment of allocation strategies, statistical analyses of allocation patterns, and fully differentiable, science-driven solutions for resource allocation problems.

Miles Cranmer, Peter Melchior, Brian Nord

We present an approach for maximizing a global utility function by learning how to allocate resources in an unsupervised way. We expect interactions between allocation targets to be important and therefore propose to learn the reward structure for near-optimal allocation policies with a GNN. By relaxing the resource constraint, we can employ gradient-based optimization in contrast to more standard evolutionary algorithms. Our algorithm is motivated by a problem in modern astronomy, where one needs to select-based on limited initial information-among $10^9$ galaxies those whose detailed measurement will lead to optimal inference of the composition of the universe. Our technique presents a way of flexibly learning an allocation strategy by only requiring forward simulators for the physics of interest and the measurement process. We anticipate that our technique will also find applications in a range of resource allocation problems.

S. Mucesh, W. G. Hartley, A. Palmese et al.

We demonstrate that highly accurate joint redshift-stellar mass probability distribution functions (PDFs) can be obtained using the Random Forest (RF) machine learning (ML) algorithm, even with few photometric bands available. As an example, we use the Dark Energy Survey (DES), combined with the COSMOS2015 catalogue for redshifts and stellar masses. We build two ML models: one containing deep photometry in the $griz$ bands, and the second reflecting the photometric scatter present in the main DES survey, with carefully constructed representative training data in each case. We validate our joint PDFs for $10,699$ test galaxies by utilizing the copula probability integral transform and the Kendall distribution function, and their univariate counterparts to validate the marginals. Benchmarked against a basic set-up of the template-fitting code BAGPIPES, our ML-based method outperforms template fitting on all of our predefined performance metrics. In addition to accuracy, the RF is extremely fast, able to compute joint PDFs for a million galaxies in just under $6$ min with consumer computer hardware. Such speed enables PDFs to be derived in real time within analysis codes, solving potential storage issues. As part of this work we have developed GALPRO, a highly intuitive and efficient Python package to rapidly generate multivariate PDFs on-the-fly. GALPRO is documented and available for researchers to use in their cosmology and galaxy evolution studies.

Francois Lanusse, Peter Melchior, Fred Moolekamp

We present a Bayesian machine learning architecture that combines a physically motivated parametrization and an analytic error model for the likelihood with a deep generative model providing a powerful data-driven prior for complex signals. This combination yields an interpretable and differentiable generative model, allows the incorporation of prior knowledge, and can be utilized for observations with different data quality without having to retrain the deep network. We demonstrate our approach with an example of astronomical source separation in current imaging data, yielding a physical and interpretable model of astronomical scenes.