Machine learning spots cosmic explosions in Fermi telescope data

Researchers trained a ConvLSTM neural network on ten years of simulated Fermi-LAT gamma-ray observations to learn the expected sky baseline, then taught it to spot anomalies by reconstructing what should be there and measuring departures. The network flags localized excesses consistent with transients like flares and gamma-ray bursts, while filtering out noise. This provides an automated early-warning system for the next generation of time-domain astronomy.