L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Lesson 1 · AI in Astronomy and Physics

Scanning the Sky: AI and the Data Deluge

Modern telescopes generate more data in a single night than astronomers could analyze in a lifetime — so AI does the sorting.
How does machine learning turn petabytes of starlight into scientific discovery?

The Vera C. Rubin Observatory's Legacy Survey of Space and Time (LSST) is projected to photograph the entire southern sky every few nights, producing roughly 15 terabytes of raw image data per night. No human team could flag every transient event — supernovae, asteroid approaches, variable stars — before they fade. The observatory's alert-brokering pipeline uses machine-learning classifiers to tag millions of candidate events within 60 seconds of each image being taken, routing only the most scientifically promising alerts to human astronomers.

Why Astronomy Needs AI Now

The history of astronomy is a history of instrument improvements producing more light than human eyes could process. Photographic plates replaced direct observation; digital CCDs replaced plates. Each leap multiplied the data volume by orders of magnitude. The current transition to wide-field survey telescopes — Rubin, the Square Kilometre Array (SKA), Euclid — creates a data regime where the bottleneck is no longer collection but interpretation.

A single Rubin image contains roughly 3,200 megapixels covering 9.6 square degrees of sky. Differencing that image against a reference frame to find transients, then classifying each transient candidate by type, shape, variability, and host-galaxy context, is a task that scales to millions of events per night. Machine learning — particularly convolutional neural networks trained on labeled historical datasets — handles this classification pipeline automatically.

Beyond transient detection, AI enables photometric redshift estimation (inferring how far away a galaxy is from its color alone), morphological classification (spiral vs. elliptical vs. merging), and source separation in crowded stellar fields. Each of these tasks would require years of expert human labor per survey; AI completes them in hours.

Real Case — Zwicky Transient Facility

The Zwicky Transient Facility (ZTF) at Palomar Observatory, operating since 2018, uses a real-time machine-learning broker called ANTARES and the community broker ALeRCE to classify roughly one million alerts per night. In 2019, ZTF's AI pipeline identified the closest known Type Ia supernova in a decade — SN 2019ehk — within hours of first light, enabling rapid spectroscopic follow-up that would have been impossible with manual scanning.

Convolutional Neural Networks in Telescope Pipelines

The standard architecture for image-based astronomical classification is a convolutional neural network (CNN) trained on labeled postage-stamp images of known source types. The network learns spatial features — the elongation of a galaxy, the point-spread function of a star, the light-curve shape of a supernova — and maps them to probability scores across classes.

Training data comes from previous surveys and simulations. A key challenge is class imbalance: genuine supernovae are outnumbered by artefacts, variable stars, and active galactic nuclei by factors of thousands. Techniques like synthetic minority oversampling, cost-sensitive learning, and transfer from simulation-trained models address this imbalance without requiring astronomers to manually label millions of examples.

Another challenge is domain shift: a CNN trained on one telescope's images (with its specific point-spread function, pixel scale, and filter set) may perform poorly on a different instrument. The field has responded with domain-adaptation techniques and instrument-agnostic feature representations.

Key Terms
TransientAn astronomical event that changes brightness or position over time — supernovae, gamma-ray bursts, variable stars, near-Earth objects.
Alert BrokerA software system that receives telescope event streams, classifies objects in near-real time, and distributes prioritized alerts to the community.
Photometric RedshiftAn estimate of a galaxy's distance derived from multi-band imaging brightness rather than a full spectrum; AI dramatically improves its accuracy.
Domain ShiftDegradation in model performance when the instrument, observing conditions, or data format differ from the training set.
Scale Check

During its 10-year LSST survey, Rubin Observatory is expected to catalog approximately 20 billion galaxies and 17 billion stars. Classifying even 1% of those objects manually at one minute per object would require over 300,000 person-years of labor. AI makes the science possible — not just faster.

Gravitational-Wave Astronomy

LIGO and Virgo detect gravitational waves by measuring interference patterns in 4-km laser arms — but the raw data is dominated by instrumental noise from seismic events, scattered light, and electronic glitches called glitches. Machine-learning classifiers trained on thousands of labeled noise events can identify and veto glitches in near-real time, improving the sensitivity of the entire detector network.

In 2017, the detection of GW170817 — the first neutron-star merger observed in both gravitational waves and light — triggered a global electromagnetic follow-up campaign. The rapid ML-assisted vetting of that signal as astrophysical rather than instrumental was essential to issuing the alert fast enough for telescopes to capture the kilonova before it faded.

Lesson 1 Quiz

AI and the Astronomical Data Deluge — 4 questions
Approximately how much raw image data does the Vera C. Rubin Observatory's LSST survey produce per night?
Correct. The LSST pipeline is projected to produce roughly 15 terabytes per night, making automated AI classification essential.
Not quite. The projected figure is approximately 15 terabytes per night — already far beyond manual analysis capacity.
What is a "transient" in the context of astronomical surveys?
Correct. Transients include supernovae, gamma-ray bursts, variable stars, and near-Earth objects — all requiring rapid detection and classification.
Not quite. A transient is an astronomical event — like a supernova or variable star — that changes brightness or position over time.
What was significant about the gravitational-wave event GW170817 in 2017?
Correct. GW170817 was the first multi-messenger event combining gravitational waves and electromagnetic light, with ML-assisted vetting enabling rapid alert distribution.
Not quite. GW170817 was the first neutron-star merger detected in both gravitational waves and across the electromagnetic spectrum — a landmark multi-messenger event.
Which challenge in astronomical AI refers to a model performing poorly when applied to data from a different telescope than the one it was trained on?
Correct. Domain shift occurs when differing instrument characteristics (pixel scale, point-spread function, filter set) cause model performance to degrade outside the training context.
Not quite. This is called domain shift — when the data distribution encountered at deployment differs from the training distribution, here due to instrument differences.

Lab 1 — Alert Classification

Explore how AI brokers triage millions of nightly telescope alerts

Your Task

You are analyzing the design of an AI alert-brokering system for a wide-field survey telescope. Discuss with the AI assistant how the classification pipeline works, what data features the model uses, and how astronomers decide which alerts deserve follow-up.

Suggested opening: "Walk me through how a CNN-based alert broker classifies a supernova candidate in real time. What features does it use, and how does it handle class imbalance?"
AI Lab Assistant
Alert Classification
Welcome to Lab 1. I'm here to help you explore how AI-powered alert brokers classify transient events from wide-field telescopes like ZTF and the future Rubin Observatory. Ask me anything about the classification pipeline, training data, or how astronomers use these systems in practice.
Lesson 2 · AI in Astronomy and Physics

Exoplanets and the Transit Signal

Finding Earth-sized worlds inside noisy stellar light curves is a needle-in-a-haystack problem that AI has fundamentally transformed.
How did a neural network find two new exoplanets in Kepler data that human analysts had missed?

In December 2017, Google Brain researcher Christopher Shallue and astronomer Andrew Vanderburg announced they had trained a neural network on 15,000 labeled Kepler light curves and applied it to 670 previously examined star systems. The model found two previously undetected exoplanets — Kepler-90i and Kepler-80g — hiding in signals that automated and human searches had classified as noise or threshold crossings. Kepler-90i gave its star the distinction of hosting eight confirmed planets, tying our own solar system for the most in any known system at the time.

The Transit Method and Its Noise Problem

When a planet passes in front of its host star, it blocks a tiny fraction of the star's light. For an Earth-sized planet orbiting a Sun-like star, this "transit depth" is roughly 84 parts per million — a signal smaller than typical stellar variability and instrumental noise. The Kepler space telescope, operating from 2009 to 2018, collected photometric time series for over 150,000 stars, producing about 14 billion data points.

NASA's official planet-search pipeline (Transiting Planet Search, or TPS) uses a box-least-squares algorithm that looks for periodic, symmetric dips. It flags "threshold crossing events" (TCEs) — roughly 35,000 from the full Kepler dataset — which then require human review. But with limited time and analyst capacity, many TCEs in multi-planet systems or at low signal-to-noise were deprioritized. That is exactly the gap Shallue and Vanderburg's neural network exploited.

Real Case — Kepler-90i Discovery

Shallue and Vanderburg's 1D convolutional neural network, trained on "global" light-curve views (the full phase-folded dip) and "local" views (a zoomed window around the transit), achieved 96% accuracy on a held-out test set. Applied to 670 candidate multi-planet systems, it flagged Kepler-90i (an Earth-sized rocky planet with an orbital period of 14.4 days) and Kepler-80g. Both were confirmed by follow-up analysis. The model and training data were released as open source in 2018.

TESS and Ongoing AI Discovery

The Transiting Exoplanet Survey Satellite (TESS), launched in 2018, surveys nearly the entire sky in 27-day sectors, generating thousands of planet candidates per sector. Because TESS targets brighter, nearer stars than Kepler did, its candidates are more amenable to ground-based confirmation — but the volume still demands AI triage.

The SHERLOCK pipeline and the TFOP (TESS Follow-up Observing Program) use machine-learning vetters to rank TESS Objects of Interest (TOIs) by their likelihood of being genuine planets versus astrophysical false positives (eclipsing binaries, background stars) or instrumental systematics. By 2024, more than 400 TESS planets had been confirmed, with AI-assisted vetting playing a central role in prioritizing ground-based telescope time.

Atmospheric Characterization

Once a planet is confirmed, astronomers use transmission spectroscopy — measuring which wavelengths of starlight are absorbed as the planet transits — to probe its atmosphere. The James Webb Space Telescope (JWST), operational since 2022, produces spectra of extraordinary detail. AI tools including ExoTransmit and Bayesian atmospheric retrieval codes like POSEIDON and petitRADTRANS fit millions of model spectra to observations to infer temperature-pressure profiles, molecular abundances, and cloud properties.

In 2023, JWST's first confirmed detection of carbon dioxide in the atmosphere of WASP-39b was followed by detections of sulfur dioxide, water, and other molecules — with ML-assisted retrieval helping disentangle overlapping spectral features at the instrument's noise floor.

Key Terms
Transit DepthThe fractional decrease in stellar brightness caused by a planet crossing the star's disk — as small as 84 ppm for an Earth-Sun analog.
Threshold Crossing Event (TCE)A statistically significant periodic dip in a Kepler or TESS light curve that triggers further review; not all TCEs are planets.
Transmission SpectroscopyMeasuring wavelength-dependent transit depths to infer the composition and structure of a planet's atmosphere.
Atmospheric RetrievalBayesian or ML-based inference of atmospheric parameters from observed spectra by fitting large libraries of forward models.
Perspective

As of 2024, more than 5,600 exoplanets have been confirmed, with thousands more awaiting confirmation. The rate of discovery has accelerated precisely as AI pipeline tools have become standard practice in transit photometry, radial velocity analysis, and direct imaging. The question is no longer whether we can find planets, but whether any of them harbor biosignatures — and AI will likely be central to that answer too.

Lesson 2 Quiz

Exoplanets and the Transit Signal — 4 questions
In December 2017, researchers Shallue and Vanderburg used a neural network on Kepler data to discover two new exoplanets. What was notable about Kepler-90i specifically?
Correct. Kepler-90i brought its star's confirmed planet count to eight, matching our own solar system — the first time any star had been found with so many planets.
Not quite. Kepler-90i was notable for bringing the Kepler-90 system to eight confirmed planets, tying our solar system for the most planets known around a single star.
What is a "Threshold Crossing Event" (TCE) in the context of Kepler data analysis?
Correct. TCEs are candidates flagged by the automated pipeline — not all are planets. The Kepler dataset produced about 35,000 TCEs requiring vetting.
Not quite. A TCE is a statistically significant periodic brightness dip flagged by the pipeline as potentially being a planet transit — but it must then be vetted to confirm it is real.
JWST's 2023 observations of WASP-39b used what technique to detect molecules like CO₂ and sulfur dioxide in its atmosphere?
Correct. Transmission spectroscopy measures how different wavelengths of starlight are absorbed as the planet transits, revealing atmospheric composition.
Not quite. JWST used transmission spectroscopy — measuring wavelength-dependent changes in transit depth as molecules in the planet's atmosphere absorb specific wavelengths of starlight.
What is "atmospheric retrieval" in exoplanet science?
Correct. Retrieval codes use Bayesian or ML methods to fit millions of forward-model spectra to observations, inferring temperature, pressure, and molecular abundances.
Not quite. Atmospheric retrieval means fitting large libraries of theoretical spectra to observed data to infer the properties — temperature, pressure, chemistry — of an exoplanet's atmosphere.

Lab 2 — Exoplanet Transit Analysis

Investigate how neural networks find planets hiding in stellar light curves

Your Task

You are a research student working with a TESS light curve vetting pipeline. Discuss with the AI assistant how the pipeline distinguishes genuine planet transits from false positives, what the "global" and "local" view architecture means, and how to interpret a vetter's confidence score.

Suggested opening: "Explain the difference between a genuine planet transit and an eclipsing binary false positive. How does the neural network tell them apart from the light curve alone?"
AI Lab Assistant
Transit Vetting
Welcome to Lab 2. I'm your guide through exoplanet transit analysis and AI vetting pipelines. We can explore how neural networks distinguish planet signals from false positives, how training data is built from Kepler and TESS archives, and what confidence scores actually mean in practice. What would you like to explore?
Lesson 3 · AI in Astronomy and Physics

Imaging Black Holes and Dark Matter Mapping

From reconstructing the first image of a black hole shadow to mapping invisible matter across the cosmic web — AI is reshaping what we can see.
How did machine learning help produce the image of M87*'s shadow, and how does AI reveal dark matter we cannot directly observe?

On April 10, 2019, the Event Horizon Telescope (EHT) collaboration released the first image ever captured of a black hole shadow — the supermassive black hole M87*, 55 million light-years away, surrounded by a glowing ring of accreting plasma. The image was not a conventional photograph. It was reconstructed by a machine-learning-assisted imaging algorithm called CLEAN and two newer methods — SMILI and eht-imaging — that filled in the enormous gaps between the eight radio telescopes forming a planet-spanning interferometer.

The Imaging Problem: Filling Gaps in the uv-Plane

Radio interferometry works by measuring the interference fringes between pairs of telescopes separated by thousands of kilometers. Each baseline (telescope pair) samples one Fourier component of the sky brightness distribution — one point in what astronomers call the uv-plane. The EHT in 2017 had only eight stations, providing sparse, irregular coverage. Reconstructing a 2D image from incomplete Fourier measurements is an ill-posed inverse problem: infinitely many images are mathematically consistent with the data.

Classical approaches like CLEAN, developed in the 1970s, iteratively subtract point sources. Newer regularized maximum-likelihood methods (the basis of eht-imaging and SMILI) add constraints — sparsity, smoothness, positivity — that encode prior knowledge about what astrophysical images typically look like. Machine learning enters at two points: training a network to learn the prior from a large corpus of simulated black hole images (general-relativistic magnetohydrodynamic simulations), and using that learned prior to regularize the reconstruction.

The EHT team ran four independent imaging pipelines, comparing results to ensure the bright ring and dark central depression were robust features rather than reconstruction artifacts. In 2022, the same collaboration released an image of Sagittarius A* — the black hole at the center of our own Milky Way — overcoming the additional challenge of rapid image variability on timescales of minutes.

Real Case — PRIMO Algorithm, 2023

In 2023, researchers from the Institute for Advanced Study published a new ML-based image reconstruction method called PRIMO (Principal-component Interferometric Modeling). Trained on 30,000 GRMHD simulated black hole images, PRIMO produced a sharper reconstruction of M87* than the original 2019 methods, revealing finer structure in the emission ring. The result demonstrated that as AI training data (simulations) improve, so too does our view of the universe's most extreme objects.

Weak Gravitational Lensing and Dark Matter Maps

Dark matter makes up roughly 27% of the universe's total energy content but emits no light. Its presence is inferred through gravity — specifically, by measuring how it distorts the shapes of background galaxies through weak gravitational lensing. The distortions are tiny (shears of 1–2% in galaxy ellipticity) and must be extracted from images containing billions of galaxies, each of which also has its own intrinsic shape.

The Dark Energy Survey (DES), the Kilo-Degree Survey (KiDS), and now Euclid all use AI-assisted shape measurement pipelines to extract lensing shear signals from hundreds of millions of galaxies. The challenge is calibration: the measured shape includes contributions from the atmosphere, telescope optics, and detector that must be precisely removed. Machine-learning methods including deep learning and Gaussian process regression are now used to model and subtract these point-spread function (PSF) contributions with sub-percent accuracy.

Cosmic Web and Large-Scale Structure

Beyond individual galaxy measurements, AI tools are used to reconstruct the three-dimensional distribution of dark matter — the cosmic web of filaments, voids, and nodes that structures the universe on scales of hundreds of millions of light-years. Neural networks trained on N-body simulations learn to predict 3D density fields from 2D projected maps, effectively performing a learned inversion of the lensing projection.

In 2022, a team using the DES Year 3 data and a CNN-based pipeline produced the most precise map of dark matter distribution across one-eighth of the sky, finding results consistent with — but slightly lower than — predictions from the standard cosmological model (Lambda-CDM). This tension, known as the S8 tension, is one of the most actively studied puzzles in observational cosmology, and AI-assisted lensing maps are central to its resolution.

Key Terms
uv-PlaneThe Fourier space sampled by radio interferometer baselines; sparse coverage creates the ill-posed image reconstruction problem.
Regularized Maximum LikelihoodAn imaging method that finds the image best fitting the data while penalizing solutions that violate smoothness, sparsity, or other priors.
Weak Gravitational LensingThe subtle distortion of background galaxy shapes by intervening mass — used to map dark matter without detecting it directly.
S8 TensionA discrepancy between the matter clustering amplitude measured by lensing surveys and predicted by the CMB-calibrated standard model.
What AI Made Possible

Both the EHT image and dark matter weak-lensing maps involve reconstructing invisible or indirect signals from noisy, incomplete data. Without AI-assisted regularization and learned priors, neither result would have been achievable at the precision that has driven new scientific questions. The tools themselves become instruments of discovery.

Lesson 3 Quiz

Imaging Black Holes and Dark Matter Mapping — 4 questions
The 2023 PRIMO algorithm produced a sharper reconstruction of M87*. On what type of data was it trained?
Correct. PRIMO was trained on 30,000 GRMHD simulated images, demonstrating that better simulation training data yields sharper real-world reconstructions.
Not quite. PRIMO was trained on 30,000 GRMHD (general-relativistic magnetohydrodynamic) simulated black hole accretion images.
Why is reconstructing an image from Event Horizon Telescope data described as an "ill-posed" problem?
Correct. With only eight stations and sparse Fourier coverage, the reconstruction problem is mathematically underdetermined — AI-learned priors constrain which solution to adopt.
Not quite. "Ill-posed" means the data (sparse Fourier samples) are compatible with infinitely many images. Regularization and learned priors are needed to select a physically meaningful solution.
What is weak gravitational lensing used for in observational cosmology?
Correct. Weak lensing measures the ~1–2% coherent distortion of galaxy ellipticities caused by intervening mass, enabling dark matter maps without direct detection.
Not quite. Weak gravitational lensing uses the subtle, coherent distortion of background galaxy shapes to map foreground dark matter distributions across the cosmic web.
The "S8 tension" refers to what cosmological discrepancy?
Correct. The S8 tension is the discrepancy between the matter clustering amplitude (σ8 times √(Ωm/0.3)) measured by weak-lensing surveys and inferred from the CMB.
Not quite. The S8 tension is specifically the discrepancy between the matter clustering strength measured by lensing surveys and the value predicted from CMB observations under Lambda-CDM.

Lab 3 — Black Hole Imaging & Dark Matter

Probe how AI reconstructs unseen structures from incomplete and indirect data

Your Task

You are a graduate student in observational cosmology. Discuss with the AI assistant how regularized maximum-likelihood methods work in EHT imaging, and how weak-lensing pipelines extract dark matter maps from galaxy shape catalogs. Focus on the role of priors and simulation-based training.

Suggested opening: "What does it mean to 'regularize' an ill-posed image reconstruction? How did the EHT pipelines decide which regularizers to apply, and how did ML-learned priors change that process?"
AI Lab Assistant
Imaging & Lensing
Welcome to Lab 3. I'm here to help you think through image reconstruction for radio interferometry and dark matter mapping through weak gravitational lensing. Both involve inferring structure from incomplete or indirect measurements — exactly where AI-learned priors add the most value. What aspect would you like to start with?
Lesson 4 · AI in Astronomy and Physics

AI in Particle Physics and Cosmological Simulation

At the LHC, AI sorts through 40 million collisions per second. In virtual universes, neural networks compress billion-particle simulations into seconds.
How is machine learning solving some of the hardest computational problems in fundamental physics?

The ATLAS and CMS detectors at CERN's Large Hadron Collider record proton-proton collisions at a rate of 40 million per second. Writing every event to disk is physically impossible — the raw data rate would exceed one exabyte per day. A multi-level trigger system must decide, in microseconds to milliseconds, which events to keep. At Level-1, programmable hardware uses simplified AI-like threshold logic. At the High-Level Trigger, neural networks trained on millions of simulated events evaluate collision topology in real time, rejecting 99.997% of all collisions while retaining essentially every event that might contain new physics.

Trigger Systems and Real-Time Inference

The challenge of real-time triggering is one of the most demanding AI deployment environments in science. Inference must complete in under 4 microseconds at the hardware-level trigger (L1) and under 400 milliseconds at the High-Level Trigger (HLT). CERN engineers use field-programmable gate arrays (FPGAs) running quantized, compressed neural networks — a subfield called hls4ml (high-level synthesis for ML) — to achieve this speed.

In 2020, ATLAS deployed the first FPGA-based neural network trigger for real collision data, targeting boosted jets from massive particles decaying at high transverse momentum. The network outperformed the hand-crafted cut-based trigger it replaced while running faster. CMS followed with its own ML-based L1 trigger upgrades ahead of Run 3, including a boosted jet tagger and a displaced-vertex finder for long-lived particle signatures.

Real Case — Anomaly Detection at the LHC

A long-standing challenge in particle physics is searching for new physics without knowing exactly what to look for. In 2021, the CMS collaboration published results from an autoencoder-based anomaly detection trigger, the first of its kind deployed in a hadron collider experiment. The autoencoder was trained only on Standard Model events; deviations from its learned reconstruction — events it found "surprising" — were flagged as anomalous candidates. This unsupervised approach enables model-independent searches for new particles without precommitting to a specific theoretical prediction.

Particle Identification and Jet Tagging

Inside each collision, jets of hadrons spray outward from quark and gluon interactions. Identifying the type of quark or gluon that initiated each jet — called jet tagging — is critical for Higgs boson studies and searches for beyond-Standard-Model physics. Traditional taggers used a handful of hand-crafted variables. Modern deep-learning taggers like ParticleNet (CMS) and GN1/GN2 (ATLAS) treat each jet as a point cloud of particle tracks and calorimeter deposits, achieving state-of-the-art b-quark, charm-quark, and boosted-object identification with graph neural networks.

ATLAS's GN2 tagger, deployed in Run 3, improves b-jet identification efficiency by roughly 10% at the same false-positive rate compared to its predecessor — directly translating to more precise Higgs coupling measurements and stronger exclusion limits on supersymmetric particles.

Cosmological Simulations and Emulators

State-of-the-art N-body simulations of cosmic structure formation — such as IllustrisTNG or the FLAMINGO simulation — run on supercomputers for months and track billions of particles interacting gravitationally and hydrodynamically. Running many such simulations to explore different cosmological parameter combinations is computationally prohibitive.

Neural network emulators solve this by learning the mapping from input parameters (matter density, dark energy equation of state, neutrino mass) to output observables (power spectra, halo mass functions) from a small number of full simulations, then interpolating across parameter space in milliseconds. The Aemulus and EuclidEmulator2 projects use this approach, enabling Markov Chain Monte Carlo parameter inference that would otherwise require thousands of full simulations.

A related technique — field-level inference using simulation-based inference (SBI) — skips summary statistics entirely and trains neural networks to compare full simulation outputs to observed galaxy surveys, compressing the entire inference chain into differentiable neural architecture. The SIMBIG analysis, applied to BOSS galaxy survey data in 2022, used SBI to extract cosmological constraints from the full galaxy field rather than compressed power spectra, recovering information previously discarded.

Key Terms
hls4mlHigh-level synthesis for machine learning — a framework for compiling trained neural networks onto FPGAs for microsecond-scale inference in particle physics triggers.
Jet TaggingIdentifying the particle type (b-quark, c-quark, gluon, boosted Higgs) that initiated a hadronic jet in a collider detector.
Neural Network EmulatorA fast surrogate model trained on full simulations that interpolates outputs across parameter space, replacing expensive computation with millisecond inference.
Simulation-Based Inference (SBI)A likelihood-free Bayesian inference framework that trains neural networks to compare data directly to simulations, avoiding the need for tractable likelihoods.
The Common Thread

Across all four lessons of this module — from survey alert brokers to exoplanet vetters, from black hole imagers to collider triggers — AI is solving the same fundamental problem: extracting scientific signal from data volumes and complexity that exceed the capacity of classical methods and human analysts. The physics is different in every case; the computational strategy is strikingly similar.

Lesson 4 Quiz

AI in Particle Physics and Cosmological Simulation — 4 questions
What does the hls4ml framework accomplish in particle physics experiments?
Correct. hls4ml converts trained ML models into synthesizable hardware descriptions for FPGAs, enabling inference in under 4 microseconds at the hardware trigger level.
Not quite. hls4ml is a framework for deploying trained neural networks on FPGAs in real time — essential for meeting the microsecond latency requirements of particle physics hardware triggers.
What was novel about the CMS autoencoder-based anomaly detection trigger published in 2021?
Correct. Trained only on Standard Model events, the autoencoder flags deviations as anomalous — enabling searches for new physics without committing to a specific model.
Not quite. The autoencoder's novelty is that it enables model-independent new physics searches: trained on SM events, it flags anything unusual without needing a specific theoretical prediction to target.
What problem do neural network emulators solve in cosmological parameter inference?
Correct. Emulators like EuclidEmulator2 learn the parameter-to-observable mapping from a small set of full simulations, enabling MCMC chains that would otherwise require thousands of full runs.
Not quite. Neural network emulators act as fast surrogates — trained on a small number of full simulations, they interpolate outputs for any parameter combination in milliseconds, making MCMC inference tractable.
The ATLAS GN2 jet tagger uses what architecture to treat jets as sets of individual particles?
Correct. Graph neural networks like ParticleNet and GN2 naturally handle the variable-length, unordered set of particles in each jet — a key advantage over grid-based or sequence-based approaches.
Not quite. Both ParticleNet and GN2 use graph neural networks, treating each jet as a point cloud of particle constituents with edges connecting nearby particles — allowing permutation-invariant processing.

Lab 4 — Colliders and Cosmic Simulation

Explore AI trigger design at the LHC and neural network emulators for cosmology

Your Task

You are a physics student studying the computational frontiers of high-energy physics and cosmology. Discuss with the AI assistant how autoencoder-based anomaly detection works as a new-physics search strategy, and how simulation-based inference changes cosmological parameter estimation.

Suggested opening: "How does an autoencoder learn what a 'normal' LHC collision looks like, and why is its reconstruction error useful for spotting new physics? What are the risks of false positives in this approach?"
AI Lab Assistant
Particle Physics & Cosmology
Welcome to Lab 4. I'm ready to explore the cutting edge of AI in particle physics and cosmological simulation with you. We can cover LHC trigger design, autoencoder anomaly detection, graph neural network jet taggers, or simulation-based inference for cosmological parameters. What would you like to dig into?

Module 6 Test

AI in Astronomy and Physics — 15 questions · 80% to pass
1. What alert-brokering system classifies roughly one million ZTF events per night in near-real time?
Correct. ALeRCE (Automatic Learning for the Rapid Classification of Events) is one of the community brokers classifying ZTF's ~1 million nightly alerts.
ALeRCE is the correct answer — one of the community alert brokers handling ZTF's million-event nightly stream.
2. The Rubin Observatory's LSST will catalog approximately how many galaxies over its 10-year survey?
Correct. LSST is expected to catalog roughly 20 billion galaxies and 17 billion stars over its 10-year run.
The expected figure is approximately 20 billion galaxies — a scale that makes AI classification indispensable.
3. In the context of LIGO gravitational-wave detection, what are "glitches"?
Correct. Glitches are non-astrophysical noise transients that mimic signal features; ML classifiers trained on thousands of labeled glitches identify and veto them in near-real time.
Glitches are instrumental noise events — seismic, scattered light, electronic — that must be identified and vetted by ML classifiers before gravitational-wave candidates are confirmed.
4. What was the transit depth challenge for detecting Earth-sized planets with Kepler?
Correct. A transit depth of ~84 ppm requires exceptional photometric precision — well below the noise floor that challenged purely manual analysis.
The key challenge was an ~84 ppm signal depth — far smaller than stellar variability — requiring AI to extract genuine transits from noisy light curves.
5. Shallue and Vanderburg's 2017 neural network used which two complementary views of a transit signal for classification?
Correct. The dual-input architecture used a global view (full phase-folded dip) and a local view (zoomed transit window), achieving 96% accuracy on the test set.
The network used a "global" view of the full phase-folded light curve and a "local" zoomed view of the transit region — a dual-input architecture that became widely adopted.
6. As of 2024, approximately how many exoplanets have been confirmed in total?
Correct. As of 2024, over 5,600 exoplanets have been confirmed, with AI-assisted vetting central to the accelerating discovery rate.
As of 2024, more than 5,600 exoplanets have been confirmed — a figure that reflects the scaling power of AI-assisted survey pipelines.
7. Which 2019 milestone resulted in the Kepler-90 system being recognized for the most known planets around any star?
Correct. Kepler-90i was discovered by Shallue and Vanderburg's neural network in archived Kepler data, bringing the system's total to eight confirmed planets.
The neural network discovery of Kepler-90i in 2017 brought the system to eight confirmed planets — tying our solar system at the time.
8. The Event Horizon Telescope produced its first black hole image in what year, and of which object?
Correct. The EHT's first released black hole image in April 2019 showed M87*, the supermassive black hole 55 million light-years away in galaxy M87.
The first EHT black hole image was released on April 10, 2019, showing M87* — a 6.5-billion-solar-mass black hole at the center of galaxy M87.
9. What makes the EHT's image reconstruction problem "ill-posed"?
Correct. With only eight stations providing sparse, irregular uv-plane coverage, infinitely many images fit the data — requiring AI-learned priors to select physically meaningful solutions.
The problem is ill-posed because sparse uv-plane sampling leaves many images consistent with the data. Regularization and ML-learned priors from GRMHD simulations constrain the solution.
10. The S8 tension in cosmology was sharpened by AI-assisted analysis of which type of observational data?
Correct. Weak lensing surveys like DES Year 3 used ML-assisted shape measurement pipelines to produce dark matter maps that revealed the S8 tension with CMB predictions.
The S8 tension is driven by weak gravitational lensing surveys — DES, KiDS — which use AI-calibrated shape measurements to map dark matter and find clustering weaker than CMB predictions.
11. At what approximate rate do LHC proton-proton collisions occur in ATLAS and CMS detectors?
Correct. The LHC beam crossing rate is 40 MHz — 40 million bunch crossings per second — requiring real-time AI trigger systems to select which events to retain.
The LHC operates at 40 MHz — 40 million collisions per second. Only AI-assisted hardware triggers can select the ~0.003% of events worth keeping in under 4 microseconds.
12. What architecture does ATLAS's GN2 jet tagger use to process the variable-length set of particles in a jet?
Correct. Graph neural networks handle unordered, variable-length particle sets naturally — each particle is a node, with edges connecting nearby particles in angular or momentum space.
GN2 uses a graph neural network — representing each jet as a point cloud of particle tracks and calorimeter hits, with message passing between neighboring particles.
13. What is the key advantage of a neural network emulator over running full N-body cosmological simulations in MCMC inference?
Correct. Trained on a small set of full simulations, emulators return predictions in milliseconds — making MCMC chains with thousands of steps computationally feasible.
The emulator's power is speed: once trained, it returns predictions in milliseconds rather than the months a full simulation would require, enabling thorough MCMC parameter exploration.
14. The CMS 2021 autoencoder anomaly detection trigger was trained exclusively on which type of data?
Correct. Trained only on SM events, the autoencoder learns what "normal" collisions look like — events with anomalously high reconstruction error are flagged as potential new physics.
The autoencoder was trained only on Standard Model events. It learns normal topology; anything it reconstructs poorly is flagged as anomalous — a model-independent search strategy.
15. Which of the following best describes simulation-based inference (SBI) in cosmological data analysis?
Correct. SBI bypasses the need for a tractable likelihood by training neural networks to compress and compare full data fields to simulations, enabling parameter inference without summary statistics.
SBI (simulation-based inference) trains neural networks to directly compare observed data to simulations, enabling likelihood-free Bayesian parameter estimation on full data fields — as in the SIMBIG BOSS analysis.