The NSF AI Institute for Artificial Intelligence and Fundamental Interactions

As AI-powered systems increasingly mediate consequential decision-making, their explainability is critical for end-users to take informed and accountable actions. Explanations in human-human interactions are socially-situated. AI systems are often socio-organizationally embedded. However, Explainable AI (XAI) approaches have been predominantly algorithm-centered. We take a developmental step towards socially-situated XAI by introducing and exploring Social Transparency (ST), a sociotechnically informed perspective that incorporates the socio-organizational context into explaining AI-mediated decision-making. To explore ST conceptually, we conducted interviews with 29 AI users and practitioners grounded in a speculative design scenario. We suggested constitutive design elements of ST and developed a conceptual framework to unpack ST’s effect and implications at the technical, decision-making, and organizational level. The framework showcases how ST can potentially calibrate trust in AI, improve decision-making, facilitate organizational collective actions, and cultivate holistic explainability. Our work contributes to the discourse of Human-Centered XAI by expanding the design space of XAI.

Nuclear physics deals with complex systems, large datasets, and complicated correlations between parameters, which makes the field suitable for the application of machine learning techniques. Machine learning can help classify and analyze data, find hidden correlations, and assist in the design of new experiments and detectors. This Colloquium explains how this will lead to advances in nuclear theory, experimental methods and data acquisition, and accelerator technology.

What does it mean for a generative AI model to be explainable? The emergent discipline of explainable AI (XAI) has made great strides in helping people understand discriminative models. Less attention has been paid to generative models that produce artifacts, rather than decisions, as output. Meanwhile, generative AI (GenAI) technologies are maturing and being applied to application domains such as software engineering. Using scenario-based design and question-driven XAI design approaches, we explore users’ explainability needs for GenAI in three software engineering use cases: natural language to code, code translation, and code auto-completion. We conducted 9 workshops with 43 software engineers in which real examples from state-of-the-art generative AI models were used to elicit users’ explainability needs. Drawing from prior work, we also propose 4 types of XAI features for GenAI for code and gathered additional design ideas from participants. Our work explores explainability needs for GenAI for code and demonstrates how human-centered approaches can drive the technical development of XAI in novel domains.

A major challenge of AI plus science lies in its inherent incompatibility: Today’s AI is primarily based on connectionism, while science depends on symbolism. To bridge the two worlds, we propose a framework to seamlessly synergize Kolmogorov-Arnold networks (KANs) and science. The framework highlights KANs’ usage for three aspects of scientific discovery: identifying relevant features, revealing modular structures, and discovering symbolic formulas. The synergy is bidirectional: science to KAN (incorporating scientific knowledge into KANs), and KAN to science (extracting scientific insights from KANs). We highlight major new functionalities in : (1) MultKAN, KANs with multiplication nodes, (2) kanpiler, a KAN compiler that compiles symbolic formulas into KANs; (3) tree converter, convert KANs (or any neural networks) into tree graphs. Based on these tools, we demonstrate KANs’ capability to discover various types of physical laws, including conserved quantities, Lagrangians, symmetries, and constitutive laws.

We describe the outcome of a data challenge conducted as part of the Dark Machines (https://www.darkmachines.org) initiative and the Les Houches 2019 workshop on Physics at TeV colliders. The challenged aims to detect signals of new physics at the Large Hadron Collider (LHC) using unsupervised machine learning algorithms. First, we propose how an anomaly score could be implemented to define model-independent signal regions in LHC searches. We define and describe a large benchmark dataset, consisting of &gt;1 billion simulated LHC events corresponding to 10\, fb^{-1} <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"> <mml:mrow> <mml:mn>10</mml:mn> <mml:mspace width="0.167em"/> <mml:mi>f</mml:mi> <mml:msup> <mml:mi>b</mml:mi> <mml:mrow> <mml:mo>−</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:msup> </mml:mrow> </mml:math> of proton-proton collisions at a center-of-mass energy of 13 TeV. We then review a wide range of anomaly detection and density estimation algorithms, developed in the context of the data challenge, and we measure their performance in a set of realistic analysis environments. We draw a number of useful conclusions that will aid the development of unsupervised new physics searches during the third run of the LHC, and provide our benchmark dataset for future studies at https://www.phenoMLdata.org. Code to reproduce the analysis is provided at https://github.com/bostdiek/DarkMachines-UnsupervisedChallenge.

Agile maneuvers such as sprinting and high-speed turning in the wild are challenging for legged robots. We present an end-to-end learned controller that achieves record agility for the MIT Mini Cheetah, sustaining speeds up to 3.9 m/s. This system runs and turns fast on natural terrains like grass, ice, and gravel and responds robustly to disturbances. Our controller is a neural network trained in simulation via reinforcement learning and transferred to the real world. The two key components are (i) an adaptive curriculum on velocity commands and (ii) an online system identification strategy for sim-to-real transfer leveraged from prior work.

Agile maneuvers such as sprinting and high-speed turning in the wild are challenging for legged robots. We present an end-to-end learned controller that achieves record agility for the MIT Mini Cheetah, sustaining speeds up to 3.9 m/s. This system runs and turns fast on natural terrains like grass, ice, and gravel and responds robustly to disturbances. Our controller is a neural network trained in simulation via reinforcement learning and transferred to the real world. The two key components are (i) an adaptive curriculum on velocity commands and (ii) an online system identification strategy for sim-to-real transfer. Videos of the robot’s behaviors are available at https://agility.csail.mit.edu/ .

ABSTRACT In this paper, we construct the circular velocity curve of the Milky Way out to ∼30 kpc, providing an updated model of the dark matter density profile. We derive precise parallaxes for 120 309 stars with a data-driven model, using APOGEE DR17 spectra combined with GaiaDR3, 2MASS, and WISE photometry. At outer galactic radii up to 30 kpc, we find a significantly faster decline in the circular velocity curve compared to the inner parts. This decline is better fit with a cored Einasto profile with a slope parameter $0.91^{+0.04}_{-0.05}$ than a generalized Navarro–Frenk–White (NFW) profile. The virial mass of the best-fitting dark matter halo profile is only $1.81^{+0.06}_{-0.05}\times 10^{11}$ M⊙, significantly lower than what a generalized NFW profile delivers. We present a study of the potential systematics, affecting mainly large radii. Such a low mass for the Galaxy is driven by the functional forms tested, given that it probes beyond our measurements. It is found to be in tension with mass measurements from globular clusters, dwarf satellites, and streams. Our best-fitting profile also lowers the expected dark matter annihilation signal flux from the galactic centre by more than an order of magnitude, compared to an NFW profile-fit. In future work, we will explore profiles with more flexible functional forms to more fully leverage the circular velocity curve and observationally constrain the properties of the Milky Way’s dark matter halo.

Abstract We present the discovery of the Type II supernova SN 2023ixf in M101 and follow-up photometric and spectroscopic observations, respectively, in the first month and week of its evolution. Our discovery was made within a day of estimated first light, and the following light curve is characterized by a rapid rise (≈5 days) to a luminous peak ( M V ≈ − 18.2 mag) and plateau ( M V ≈ − 17.6 mag) extending to 30 days with a fast decline rate of ≈0.03 mag day −1 . During the rising phase, U − V color shows blueward evolution, followed by redward evolution in the plateau phase. Prominent flash features of hydrogen, helium, carbon, and nitrogen dominate the spectra up to ≈5 days after first light, with a transition to a higher ionization state in the first ≈2 days. Both the U − V color and flash ionization states suggest a rise in the temperature, indicative of a delayed shock breakout inside dense circumstellar material (CSM). From the timescales of CSM interaction, we estimate its compact radial extent of ∼(3–7) × 10 14 cm. We then construct numerical light-curve models based on both continuous and eruptive mass-loss scenarios shortly before explosion. For the continuous mass-loss scenario, we infer a range of mass-loss history with 0.1–1.0 M ⊙ yr −1 in the final 2−1 yr before explosion, with a potentially decreasing mass loss of 0.01–0.1 M ⊙ yr −1 in ∼0.7–0.4 yr toward the explosion. For the eruptive mass-loss scenario, we favor eruptions releasing 0.3–1 M ⊙ of the envelope at about a year before explosion, which result in CSM with mass and extent similar to the continuous scenario. We discuss the implications of the available multiwavelength constraints obtained thus far on the progenitor candidate and SN 2023ixf to our variable CSM models.

Abstract We present the optical spectroscopic evolution of SN 2023ixf seen in subnight cadence spectra from 1.18 to 15 days after explosion. We identify high-ionization emission features, signatures of interaction with material surrounding the progenitor star, that fade over the first 7 days, with rapid evolution between spectra observed within the same night. We compare the emission lines present and their relative strength to those of other supernovae with early interaction, finding a close match to SN 2020pni and SN 2017ahn in the first spectrum and SN 2014G at later epochs. To physically interpret our observations, we compare them to CMFGEN models with confined, dense circumstellar material around a red supergiant (RSG) progenitor from the literature. We find that very few models reproduce the blended N iii ( λλ 4634.0,4640.6)/C iii ( λλ 4647.5,4650.0) emission lines observed in the first few spectra and their rapid disappearance thereafter, making this a unique diagnostic. From the best models, we find a mass-loss rate of 10 −3 –10 −2 M ⊙ yr −1 , which far exceeds the mass-loss rate for any steady wind, especially for an RSG in the initial mass range of the detected progenitor. These mass-loss rates are, however, similar to rates inferred for other supernovae with early circumstellar interaction. Using the phase when the narrow emission features disappear, we calculate an outer dense radius of circumstellar material R CSM,out ≈ 5 × 10 14 cm, and a mean circumstellar material density of ρ = 5.6 × 10 −14 g cm −3 . This is consistent with the lower limit on the outer radius of the circumstellar material we calculate from the peak H α emission flux, R CSM,out ≳ 9 × 10 13 cm.

Abstract A fundamental requirement for reionizing the Universe is that a sufficient fraction of the ionizing photons emitted by galaxies successfully escapes into the intergalactic medium. However, due to the scarcity of high-redshift observational data, the sources driving reionization remain uncertain. In this work, we calculate the ionizing escape fractions (fesc) of reionization-era galaxies from the state-of-the-art thesan simulations, which combine an accurate radiation-hydrodynamic solver (arepo-rt) with the well-tested IllustrisTNG galaxy formation model to self-consistently simulate both small-scale galaxy physics and large-scale reionization throughout a large patch of the universe ($L_\text{box} = 95.5\, \text{cMpc}$). This allows the formation of numerous massive haloes ($M_\text{halo} \gtrsim 10^{10}\, {\text{M}_{\odot }}$), which are often statistically underrepresented in previous studies but are believed to be important to achieving rapid reionization. We find that low-mass galaxies ($M_\text{stars} \lesssim 10^7\, {\text{M}_{\odot }}$) are the main drivers of reionization above z ≳ 7, while high-mass galaxies ($M_\text{stars} \gtrsim 10^8\, {\text{M}_{\odot }}$) dominate the escaped ionizing photon budget at lower redshifts. We find a strong dependence of fesc on the effective star formation rate (SFR) surface density defined as the SFR per gas mass per escape area, i.e. $\bar{\Sigma }_\text{SFR} = \text{SFR}/M_\text{gas}/R_{200}^2$. The variation in halo escape fractions decreases for higher mass haloes, which can be understood from the more settled galactic structure, SFR stability, and fraction of sightlines within each halo significantly contributing to the escaped flux. Dust is capable of reducing the escape fractions of massive galaxies, but the impact on the global fesc depends on the dust model. Finally, active galactic nuclei are unimportant for reionization in thesan and their escape fractions are lower than stellar ones due to being located near the centres of galaxy gravitational potential wells.

Abstract We present the densely sampled early light curve of the Type II supernova (SN) 2023ixf, first observed within hours of explosion in the nearby Pinwheel Galaxy (Messier 101; 6.7 Mpc). Comparing these data to recently updated models of shock-cooling emission, we find that the progenitor likely had a radius of 410 ± 10 R ⊙ . Our estimate is model dependent but consistent with a red supergiant. These models provide a good fit to the data starting about 1 day after the explosion, despite the fact that the classification spectrum shows signatures of circumstellar material around SN 2023ixf during that time. Photometry during the first day after the explosion, provided almost entirely by amateur astronomers, does not agree with the shock-cooling models or a simple power-law rise fit to data after 1 day. We consider the possible causes of this discrepancy, including precursor activity from the progenitor star, circumstellar interaction, and emission from the shock before or after it breaks out of the stellar surface. The very low luminosity (−11 mag &gt; M &gt; −14 mag) and short duration of the initial excess lead us to prefer a scenario related to prolonged emission from the SN shock traveling through the progenitor system.

Ground-truth labeling is an important activity in machine learning. Many studies have examined how crowdworkers apply labels to records in machine learning datasets. However, there have been few studies that have examined the work of domain experts when their knowledge and expertise are needed to apply labels.

Over the past five years, modern machine learning has been quietly revolutionizing particle physics. Old methodology is being outdated and entirely new ways of thinking about data are becoming commonplace. This article will review some aspects of the natural synergy between modern machine learning and particle physics, focusing on applications at the Large Hadron Collider (LHC). A sampling of examples is given, from signal/background discrimination tasks using supervised learning to direct data-driven approaches. Some comments on persistent challenges and possible future directions for the field are included at the end.

Abstract We present ultraviolet/optical/near-infrared observations and modeling of Type II supernovae (SNe II) whose early time ( δ t &lt; 2 days) spectra show transient, narrow emission lines from shock ionization of confined ( r &lt; 10 15 cm) circumstellar material (CSM). The observed electron-scattering broadened line profiles (i.e., IIn-like) of H i , He i/ii , C iv , and N iii/iv/v from the CSM persist on a characteristic timescale ( t IIn ) that marks a transition to a lower-density CSM and the emergence of Doppler-broadened features from the fast-moving SN ejecta. Our sample, the largest to date, consists of 39 SNe with early time IIn-like features in addition to 35 “comparison” SNe with no evidence of early time IIn-like features, all with ultraviolet observations. The total sample includes 50 unpublished objects with a total of 474 previously unpublished spectra and 50 multiband light curves, collected primarily through the Young Supernova Experiment and Global Supernova Project collaborations. For all sample objects, we find a significant correlation between peak ultraviolet brightness and both t IIn and the rise time, as well as evidence for enhanced peak luminosities in SNe II with IIn-like features. We quantify mass-loss rates and CSM density for the sample through the matching of peak multiband absolute magnitudes, rise times, t IIn , and optical SN spectra with a grid of radiation hydrodynamics and non-local thermodynamic equilibrium radiative-transfer simulations. For our grid of models, all with the same underlying explosion, there is a trend between the duration of the electron-scattering broadened line profiles and inferred mass-loss rate: <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:msub> <mml:mrow> <mml:mi>t</mml:mi> </mml:mrow> <mml:mrow> <mml:mi>IIn</mml:mi> </mml:mrow> </mml:msub> <mml:mo>≈</mml:mo> <mml:mn>3.8</mml:mn> <mml:mo stretchy="false">[</mml:mo> <mml:mover accent="true"> <mml:mrow> <mml:mi>M</mml:mi> </mml:mrow> <mml:mrow> <mml:mo>̇</mml:mo> </mml:mrow> </mml:mover> <mml:mrow> <mml:mo stretchy="true">/</mml:mo> </mml:mrow> </mml:math> (0.01 M ⊙ yr −1 )] days.

A bstract We show that the strong CP problem is solved in a large class of compactifications of string theory. The Peccei-Quinn mechanism solves the strong CP problem if the CP-breaking effects of the ultraviolet completion of gravity and of QCD are small compared to the CP-preserving axion potential generated by low-energy QCD instantons. We characterize both classes of effects. To understand quantum gravitational effects, we consider an ensemble of flux compactifications of type IIB string theory on orientifolds of Calabi-Yau hypersurfaces in the geometric regime, taking a simple model of QCD on D7-branes. We show that the D-brane instanton contribution to the neutron electric dipole moment falls exponentially in N 4 , with N the number of axions. In particular, this contribution is negligible in all models in our ensemble with N &gt; 17. We interpret this result as a consequence of large N effects in the geometry that create hierarchies in instanton actions and also suppress the ultraviolet cutoff. We also compute the CP breaking due to high-energy instantons in QCD. In the absence of vectorlike pairs, we find contributions to the neutron electric dipole moment that are not excluded, but that could be accessible to future experiments if the scale of supersymmetry breaking is sufficiently low. The existence of vectorlike pairs can lead to a larger dipole moment. Finally, we show that a significant fraction of models are allowed by standard cosmological and astrophysical constraints.

A bstract Dark Yang-Mills sectors, which are ubiquitous in the string landscape, may be reheated above their critical temperature and subsequently go through a confining first-order phase transition that produces stochastic gravitational waves in the early universe. Taking into account constraints from lattice and from Yang-Mills (center and Weyl) symmetries, we use a phenomenological model to construct an effective potential of the semi quark-gluon plasma phase, from which we compute the gravitational wave signal produced during confinement for numerous gauge groups. The signal is maximized when the dark sector dominates the energy density of the universe at the time of the phase transition. In that case, we find that it is within reach of the next-to-next generation of experiments (BBO, DECIGO) for a range of dark confinement scales near the weak scale.

Abstract The field of particle physics is at the crossroads. The discovery of a Higgs-like boson completed the Standard Model (SM), but the lacking observation of convincing resonances Beyond the SM (BSM) offers no guidance for the future of particle physics. On the other hand, the motivation for New Physics has not diminished and is, in fact, reinforced by several striking anomalous results in many experiments. Here we summarise the status of the most significant anomalies, including the most recent results for the flavour anomalies, the multi-lepton anomalies at the LHC, the Higgs-like excess at around 96 GeV, and anomalies in neutrino physics, astrophysics, cosmology, and cosmic rays. While the LHC promises up to 4 $$\hbox {ab}^{-1}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msup> <mml:mtext>ab</mml:mtext> <mml:mrow> <mml:mo>-</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:msup> </mml:math> of integrated luminosity and far-reaching physics programmes to unveil BSM physics, we consider the possibility that the latter could be tested with present data, but that systemic shortcomings of the experiments and their search strategies may preclude their discovery for several reasons, including: final states consisting in soft particles only, associated production processes, QCD-like final states, close-by SM resonances, and SUSY scenarios where no missing energy is produced. New search strategies could help to unveil the hidden BSM signatures, devised by making use of the CERN open data as a new testing ground. We discuss the CERN open data with its policies, challenges, and potential usefulness for the community. We showcase the example of the CMS collaboration, which is the only collaboration regularly releasing some of its data. We find it important to stress that individuals using public data for their own research does not imply competition with experimental efforts, but rather provides unique opportunities to give guidance for further BSM searches by the collaborations. Wide access to open data is paramount to fully exploit the LHCs potential.

A bstract Anomaly detection relies on designing a score to determine whether a particular event is uncharacteristic of a given background distribution. One way to define a score is to use autoencoders, which rely on the ability to reconstruct certain types of data (background) but not others (signals). In this paper, we study some challenges associated with variational autoencoders, such as the dependence on hyperparameters and the metric used, in the context of anomalous signal (top and W ) jets in a QCD background. We find that the hyperparameter choices strongly affect the network performance and that the optimal parameters for one signal are non-optimal for another. In exploring the networks, we uncover a connection between the latent space of a variational autoencoder trained using mean-squared-error and the optimal transport distances within the dataset. We then show that optimal transport distances to representative events in the background dataset can be used directly for anomaly detection, with performance comparable to the autoencoders. Whether using autoencoders or optimal transport distances for anomaly detection, we find that the choices that best represent the background are not necessarily best for signal identification. These challenges with unsupervised anomaly detection bolster the case for additional exploration of semi-supervised or alternative approaches.

We present a technique for translating a black-box machine-learned classifier operating on a high-dimensional input space into a small set of human-interpretable observables that can be combined to make the same classification decisions. We iteratively select these observables from a large space of high-level discriminants by finding those with the highest decision similarity relative to the black box, quantified via a metric we introduce that evaluates the relative ordering of pairs of inputs. Successive iterations focus only on the subset of input pairs that are misordered by the current set of observables. This method enables simplification of the machine-learning strategy, interpretation of the results in terms of well-understood physical concepts, validation of the physical model, and the potential for new insights into the nature of the problem itself. As a demonstration, we apply our approach to the benchmark task of jet classification in collider physics, where a convolutional neural network acting on calorimeter jet images outperforms a set of six well-known jet substructure observables. Our method maps the convolutional neural network into a set of observables called energy flow polynomials, and it closes the performance gap by identifying a class of observables with an interesting physical interpretation that has been previously overlooked in the jet substructure literature.