## TL;DR
Astronomy has become a data-driven science drowning in data — the Vera Rubin Observatory (LSST) will generate 20 TB of images per night, JWST produces terabytes weekly, and LIGO streams continuous gravitational wave data. AI is the only viable way to process, classify, and discover in this data deluge. From finding 100+ hidden exoplanets in old NASA data to classifying billions of galaxies, AI is becoming astronomy's most productive tool.

## Core Explanation
AI astronomy applications: (1) Galaxy morphology — CNNs trained on Galaxy Zoo citizen-science labels classify millions of galaxies into Hubble types (elliptical, spiral, irregular) and detect rare morphologies (mergers, ring galaxies). Transfer learning from ImageNet-pretrained models works surprisingly well on astronomical images despite domain differences; (2) Exoplanet detection — transit method: planet crossing in front of star causes tiny brightness dip. CNNs and LSTMs process Kepler/TESS light curves to distinguish planet transits from instrumental noise, stellar variability, and eclipsing binaries; (3) Gravitational waves — matched filtering (traditional method) is computationally expensive; CNNs and normalizing flows (DINGO) accelerate waveform parameter estimation by 100-1000x; (4) Strong lensing — AI detects gravitationally lensed galaxies (Einstein rings, arcs) in wide-field surveys; (5) Transient astronomy — real-time classification of nightly alerts as supernovae, kilonovae, asteroids, or artifacts.

## Detailed Analysis
JWST era: NIRCam and MIRI instruments produce ultra-deep infrared images revealing galaxies at z>10 (first billion years of universe). AI-powered photometric redshift estimation (predicting galaxy distances from multi-band photometry) replaces traditional SED fitting, achieving 100x speedup at comparable accuracy. LSST/Vera Rubin Observatory (first light 2025): 3.2 gigapixel camera, 10 million transient alerts per night. The ANTARES broker system and ALeRCE use ML classifiers to triage alerts in real-time, flagging the 0.1% most scientifically interesting events for follow-up spectroscopy. Nature 2025 exoplanet CNN: training on simulated transits with injected noise patterns achieves 98% recall and 96% precision on Kepler data. AI pipleline at Warwick (2026): the system reanalyzed all 200,000+ Kepler/TESS light curves using ensemble CNNs, uncovering 100+ planets missed by the original Kepler pipeline due to subtle signal patterns. Gravitational wave cosmology: AI accelerates inference for compact binary coalescence (black hole/neutron star mergers) from hours to seconds, enabling real-time electromagnetic follow-up (LIGO-Virgo-KAGRA alerts). MNRAS surveys: machine learning for strong lensing detection, cosmic web classification, and dark matter substructure inference from galaxy-galaxy strong lensing.

## Further Reading
- Galaxy Zoo: Citizen Science Galaxy Classification
- ZTF & LSST Alert Brokers (ANTARES, ALeRCE, Fink)
- AstroML: Machine Learning for Astronomy (Python Library)