learn/steps in the work
Training
Training is the process of fitting a model to examples, adjusting its internal numbers until its outputs match the answers it is shown, so it can then handle data it has never seen.
What it is
A model of this kind starts as a set of adjustable numbers, called parameters, that turn an input into an output. At the start these numbers are arbitrary, so the output is nonsense. Training feeds the model examples where the right answer is already known, compares its output with that answer, and nudges the parameters a little to shrink the gap. Repeat this many times and the parameters settle into values that work on the examples, and usually on similar data the model has not seen. The known answers can come from people marking up data by hand, or from slower and more trusted calculations.
Training rarely starts from scratch. Fine-tuning takes a model already fitted to a large, general collection of data and continues training it on a smaller, specific set, so it keeps what it learned and adapts to the new task. Training also involves choices that are not learned from the data, such as how big the nudges are or how many parameters there should be. These are called hyperparameters, and they are usually set by trying several combinations and keeping whichever performs best on data held back from training.
Why scientists use it
Scientists use training when they can recognise a correct answer but cannot write down the rule that produces it. A researcher can point at a particle in an image, or confirm that a peak in a spectrum is real, without being able to state in equations what makes it so. Training converts that ability to judge examples into something a machine can repeat at scale.
So training stands in for a step that would otherwise be done by hand, or by a slower calculation. Where expert judgement is the bottleneck, a model trained on a modest set of hand-marked examples can apply the same judgement to far more data. Where the trusted method is an expensive physics calculation, a model trained on its results can approximate those answers much faster, letting researchers simulate systems far larger than the original method could reach.
In the record
Limits
Training only transfers what is in the examples. A model fitted to hand-marked data inherits the markers' mistakes and habits, and one fitted to calculated answers can be no more accurate than that calculation. Performance usually drops on data unlike the training set, and the model gives no warning when that happens; it returns a confident answer either way. A model can also fit its training examples closely and still generalise poorly, which is why some data must be held back and why a good score is not proof of understanding.
Written by AIxSci: the background from general knowledge, the examples from checked records.
Every article
aix-00279
Machine learning predicts where dopant atoms sit in hexaferrite crystals before refinement
aix-00278
Neural networks sort millions of stars to build a young-star candidate catalogue
aix-00276
Machine learning sorts a simulated polymer into ordered and disordered beads
aix-00275
AI model trained on protein folding predicts how mutations affect binding
aix-00274
Neural network trained on simulated data hunts solar wave trains in radio spectra
aix-00273
Deep learning model flags bacterial proteins likely to drive infection
aix-00272
Tree-based models predict sliding wear of carburised and plain engineering steels
aix-00271
Machine learning predicts galaxy 'relicness' from survey measurements, cutting spectroscopy demand
aix-00217
Reconstructing the early universe's density field with a smoothed map of cosmic structure
aix-00218
Sparse regression fits crystal force constants to model heat flow and vibrations
aix-00219
Neural network supplies the sideways flows needed to measure the quiet Sun's magnetic energy
aix-00220
Neural network predicts how titanium alloy grains round off during annealing
aix-00221
Neural network measures rotation periods for Kepler's main-sequence stars
aix-00222
Neural networks trained on molecule pairs predict light absorption in stacks of fifty
aix-00223
Neural networks weigh cluster X-ray maps to estimate galaxy cluster masses
aix-00227
Machine learning sifts a million candidate moving objects to find 258 new cool stars
aix-00235
Random forest sorts 130 million Magellanic Cloud sources into ten classes
aix-00238
Classifiers label individual TESS brightness readings as planet transits or not
aix-00216
Neural network models simulate how lead titanate loses its electrical polarisation on heating
aix-00215
Neural network ranks gravitational wave triggers in search for neutron star mergers
aix-00213
Neural networks predict how much ultrasound energy drives magnesia dissolution
aix-00245
Neural network trained on survey data retuned to spot transients for a liquid mirror telescope
aix-00254
Neural networks read a fast radio burst's dispersion straight from its spectrum
aix-00210
Language models suggest ingredients and firing temperatures for inorganic materials recipes
aix-00209
Neural network turns Kaguya camera images into global maps of lunar surface chemistry
aix-00208
Neural network predicts atomic charges from bond lists, without atomic coordinates
aix-00206
Neural network designs five hard metallic glasses, all confirmed by experiment
aix-00205
Machine learning fitted to flexible molecules predicts oxidation potentials and hydration energies
aix-00204
Neural networks stand in for a slow model of interstellar gas clouds
aix-00159
Machine learning assigns atomic charges to size up polarisation in bismuth vanadate conductors
aix-00157
Microscopy and a trained classifier map which E. coli proteins clump together
aix-00156
Protein language model and two classifiers predict where ATP binds on proteins
aix-00155
Neural network predicts crystal density of candidate explosives from molecular structure
aix-00153
Neural networks pick amino acid sequences for short peptides in protein binding sites
aix-00152
Neural networks trained to find the cutting site on serpin proteins
aix-00151
Classifier sorts true from false AlphaFold predictions of human protein pairs
aix-00150
Deep learning sorts crystal structures for single-molecule magnet behaviour
aix-00147
Testing whether symmetry-aware neural networks can tell apart atomic arrangements in perovskites
aix-00160
Neural networks predict how zeolites take up carbon dioxide from structure alone
aix-00163
Peptides designed by simulation and machine learning sit at condensate surfaces
aix-00166
Neural network assigns protein domains to structural families from sequence alone
aix-00167
Neural network scans eleven years of night-sky images for mesospheric wave fronts
aix-00168
Deep learning reads cryo-EM maps alongside AlphaFold3 models to build protein structures
aix-00169
AlphaFold's confidence scores used to predict which disordered protein segments bind LC8
aix-00170
Machine learning predicts strength of PLA plastic filled with boron nitride flakes
aix-00171
Neural network fills gaps in protein maps from raw diffraction data
aix-00172
Machine learning tested against three definitions of protein shape-shifting
aix-00146
Machine learning ranks stable materials by predicted superconducting temperature
aix-00144
Machine-learned potentials trained to simulate copper ion conductor Cu7PS6 faster
aix-00142
Seven machine-learned water models test how density functional choice shapes water's behaviour
aix-00140
Model labels which protein residues are active sites and what job they do
aix-00139
Robot laser-heats thin films while software picks each next heating condition
aix-00136
Diffusion model generates synthetic images of the Sun sorted by flare strength
aix-00173
Machine learning predicts nanofibre thickness from electrospinning settings, tested against new scaffolds
aix-00176
Neural network potentials trained by repeatedly checking where models disagree on hydrogen sticking to copper
aix-00178
Machine learning picked catalyst recipes for a loop of 44 lab cycles
aix-00135
Clustering algorithm sorts cluster stars from background stars in thirteen open clusters
aix-00134
Meta-learning lets a cosmology emulator adapt to new galaxy depth maps quickly
aix-00133
Regression models predict how graphene loading changes aluminium's electron emission
aix-00132
Five codes measured galaxy shapes in simulated Euclid images, one using neural networks
aix-00131
Machine-learned potential predicts how chromium sulfide layers rearrange during exfoliation
aix-00130
Machine-learned potential makes long simulations of CO hydrogenation on rhodium affordable
aix-00179
Neural network predicts protein pair binding strength from simplified molecular models
aix-00180
Teaching protein language models to rank mutants from a few dozen lab measurements
aix-00181
Protein language model embeddings used to predict succinylation sites in proteins
aix-00183
Neural network predicts nanostructure optics to design a two-colour laser collimator
aix-00184
Machine learning scores short protein stretches that mark proteins for destruction
aix-00185
Protein language model trained to spot lactylated lysines in rice proteins
aix-00186
Language model trained on protein sequences predicts shape and stability of disordered proteins
aix-00187
Machine-learned force field used to model two crystal forms of formamide
aix-00188
Machine learning sorts antibody heavy chains by the germ they target
aix-00190
A transformer model segments cryo-electron tomograms from three viewing directions at once
aix-00191
Neural network stands in for slow spectral modelling to read 64 stars' chemistry
aix-00193
Researchers train a machine-learned force model to simulate carbon nanotubes' strength and heat limits
aix-00194
Diffusion model generates glass and silicon structures to order from target properties
aix-00129
Neural networks sort supernovae and estimate their distances from brightness alone
aix-00128
Phage proteins drawn as pictures, then sorted by image-recognition networks
aix-00126
Pretrained potentials give a structure search a head start in finding atomic arrangements
aix-00125
Classifying stars by type from a single wide-band telescope image
aix-00123
Neural networks swap places with black hole image simulations, both directions
aix-00122
Random Forest sorts 9,446 white dwarfs by type from Gaia spectra
aix-00121
Machine learning sorts mitochondrial proteins into three compartments from sequence alone
aix-00119
Neural network potential used to simulate how calcium oxide melts under pressure
aix-00118
Neural networks replace a slow statistical fit to find solar oscillations in Hα images
aix-00117
Deep learning model predicts polymer heat-softening across 48,208 designed candidates
aix-00116
Neural network reads protein surfaces to predict DNA and RNA binding
aix-00115
A shared test set for predicting which enzyme carries out a reaction
aix-00114
A compact way to describe molecules speeds up machine learning of their properties
aix-00113
Random atoms settle into molecules and crystals on a learned energy surface
aix-00112
Protein language model embeddings tested against classical descriptors for spotting resistance genes
aix-00111
Zero-padding lets one neural network encode crystals with differing element counts
aix-00110
Deep learning trained on robotic peptide mapping predicts where antibody proteins degrade
aix-00108
Oxidation potentials for 15,238 molecules assembled from quantum calculations and lab measurements
aix-00107
Machine learning sorts plant protein sequences into disease-resistance proteins or not
aix-00106
Volunteer labels train a model to find asteroid trails in Hubble archive images
aix-00104
Neural networks sort 400,000 JWST galaxies to trace when spirals and spheroids appeared
aix-00103
Neural network predicts yield of chitosan nanoparticles grown with olive leaf extract
aix-00101
Simulations map three sodium–bismuth compounds, then models predict their heat-to-electricity performance
aix-00100
Astronomers test pre-trained image models for sifting telescope alerts
aix-00097
Neural networks speed up forecasts of what gravitational wave detections LISA will catch
aix-00095
Neural networks estimate dark matter halo masses from a galaxy's neighbours
aix-00093
Neural network predicts gas pressure in galaxy clusters from dark matter alone
aix-00091
Neural network force field simulates how a ferroelectric crystal's molecules begin to spin
aix-00090
Machine learning reads planet-forming disk masses from archival ALMA observations
aix-00089
Neural networks trained on atomic surroundings score how mutations change proteins
aix-00088
Neural network retrained to predict how crystals absorb light, from few examples
aix-00087
Neural networks recover a simulated barred galaxy's gravity from one snapshot
aix-00086
Neural networks trained on simulated radio bursts sort real bursts by shape
aix-00084
Neural networks read simulated galaxy cluster catalogues to estimate cosmological parameters
aix-00083
A compact neural network learns to generate shape-shifting protein structures
aix-00082
Neural networks turn simulated cluster mass maps into X-ray and SZ images
aix-00081
Language models trained on protein sequences used to predict sugar-attachment sites
aix-00080
Machine learning sifts three-colour infrared survey data for cold brown dwarfs
aix-00079
Changing the training penalty improves how a network spots beta-sheets in blurry protein maps
aix-00078
Simulated water in graphene gaps shows acid's charge hugging the surface
aix-00077
Neural networks sift 812,118 quasar spectra for hidden gravitational lenses
aix-00076
Neural network strips mains-power hum from LIGO data in about a second
aix-00075
Neural network sifts Gaia data for 160,146 stars born outside the Milky Way
aix-00073
Neural networks sort eleven years of Mars plasma data into three regions
aix-00072
Attention patterns inside a protein language model used to cut sequences into reusable units
aix-00071
Simulations map how surface atoms rearrange in alloys when molecules attach
aix-00070
Neural network scores candidate composite microstructures in a strength–toughness design loop
aix-00069
Neural networks rebuild cosmic expansion history to calibrate gamma-ray burst energies
aix-00068
Machine learning predicts marine steel corrosion from seawater conditions and alloy make-up
aix-00066
Machine learning sifts 25,000 known materials for quantum-technology host candidates
aix-00065
Model predicts how fast an enzyme works on a substrate, with uncertainty attached
aix-00063
Adding a term for the unfolded protein sharpens AI stability predictions
aix-00062
A single neural network stands in for quantum calculations across 45 elements
aix-00061
A search engine for radio galaxy shapes, built on a fine-tuned image-text model
aix-00060
Neural networks stand in for slow cosmology code in parameter fitting
aix-00059
Machine-learnt stand-in for a planet model cuts interior inference from 42 hours to 8 minutes
aix-00058
Deep-learning model designs petrol blends from octane and soot targets
aix-00056
Deep learning tool predicts how single mutations weaken protein partnerships
aix-00055
Neural networks sift 315,000 galaxy images for rare polar ring systems
aix-00054
Bayesian program merges X-ray diffraction data from three experiment types in one step
aix-00051
Ultrasound-treated clay bleaches cooking oil, with machine learning picking the settings
aix-00050
Machine learning tool flags protein helices likely to bind DNA or RNA
aix-00049
Eight machine learning methods sorted supernova gravitational wave signals by nuclear physics model
aix-00047
Neural networks stand in for slow galaxy clustering calculations in cosmology analyses
aix-00046
Two neural networks predict how heat shrinks the band gap of silver-based crystals
aix-00045
Neural network reads simulated quasar light curves to estimate black hole properties
aix-00044
Machine-learnt electron densities drive molecular dynamics of a gold–saltwater interface
aix-00043
Neural network maps cosmic voids, walls and filaments from sparse galaxy tracers
aix-00042
Two unsupervised methods compress fast radio burst spectra into a handful of numbers
aix-00041
Astronomers hunt dust-buried young star clusters in eleven nearby galaxies
aix-00040
Statistical model of 5,441 perovskite solar cells predicts efficiency and suggests recipes
aix-00039
Machine learning predicts whether a dying massive star explodes, from its density profile
aix-00037
Hubble and Webb images reveal a candidate massive touching binary star in WLM
aix-00036
Four million simulated X-ray patterns used to test 21 crystal-symmetry classifiers
aix-00035
GPT-4 reads band gap values from paper sentences to train better predictors
aix-00034
Kepler planets sorted by host star's Galactic orbit show differing eccentricities
aix-00033
Neural network built from physics equations extracts solder deformation coefficients
aix-00032
Neural network turns galaxy shape distortions into maps of invisible cluster mass
aix-00023
Neural network writes amino acid sequences to fit fixed protein backbones
aix-00028
Neural network trained on quantum calculations maps how gallium melts and freezes
aix-00027
Neural networks read infrared spectra to work out how CO sits on platinum
aix-00025
Machine learning picks a new metal mixture, which casting and X-rays confirm
aix-00024
Machine learning picked which engineered enzymes to build for fatty alcohol production
aix-00022
Software works out protein shapes from electron microscope images using statistics
aix-00021
Neural network reads a million star spectra to measure temperature and chemistry
aix-00020
Neural network sifts Kepler signals and validates 301 new exoplanets
aix-00019
Neural network sorts six hundred thousand noise blips in LIGO gravitational-wave data
aix-00018
Telescope targets chosen nightly by a classifier learning from its own spectra
aix-00017
Neural network picks bright exploding stars from sky survey alerts and books telescope time
aix-00016
Machine learning picked which shape-memory alloys to make next, round by round
aix-00015
Transformer model predicts how gas-storing crystals take up gases, replacing slow simulations
aix-00013
Neural network segments 650 battery particles to measure their detachment after cycling
aix-00012
Machine-learned force model simulates silicon-oxygen structures from glass surfaces to monoxide grains
aix-00009
Gaussian process models pick light-sensitive ion channels that reach mammalian cell membranes
aix-00008
Neural network locates single protein particles in cryo-electron microscope images
aix-00007
Neural network predicts protein atom positions from amino acid sequence alone
aix-00006
Protein shapes computed from patterns of change across families of related sequences
aix-00004
Robot lab and learning agent redesign an enzyme to survive higher temperatures
aix-00003
Researchers map the sixteen-part Commander complex using crystals, cryo-EM and AlphaFold2
aix-00002