Topics
Topics
What the AI was for
The scientific job the AI did, described so that the same work in different fields looks the same.
85 articles
Property prediction
Predicting a measurable property of an existing object.
72 articles
Classification
Assigning existing objects to known categories, including deciding whether each belongs to a class of interest.
48 articles
Simulation surrogate
Standing in for a more expensive numerical simulation at lower cost.
35 articles
Structure determination
Recovering the structure of an object from indirect measurement or from sequence.
21 articles
Candidate generation
Proposing new objects for later evaluation.
21 articles
Experimental design
Choosing which experiment or measurement to perform next.
13 articles
Detection
Locating every instance of a known kind of object or event in an image, signal or data stream — not delineating its extent (segmentation) or sorting objects already found (classification).
8 articles
Denoising
Recovering signal from noisy or degraded measurement.
7 articles
Segmentation
Partitioning a signal or image into meaningful regions or objects.
4 articles
Anomaly detection
Identifying rare or unexpected instances against a background, without a predefined category for them. A classifier trained on known categories is classification.
1 articles
Literature synthesis
Extracting or combining claims across a body of published work.
Model families
The kind of model used.
76 articles
Multilayer perceptron
Fully connected feedforward network with no structural prior.
65 articles
Convolutional neural network
Convolutional architecture over gridded data.
48 articles
Transformer
Attention-based sequence architecture.
35 articles
Linear model
Linear or generalised-linear fit, including regularised variants.
34 articles
Random forest
Bagged ensemble of decision trees.
31 articles
Graph neural network
Message passing over an explicit graph structure.
26 articles
Gradient-boosted trees
Additive ensemble of decision trees fitted by boosting.
22 articles
Gaussian process
Non-parametric Bayesian model over functions.
22 articles
Clustering
Unsupervised grouping, including dimensionality reduction for it.
22 articles
Protein language model
Sequence model trained on protein sequences. Kept distinct from transformer because the corpus needs to count it separately; see CLAUDE.md §5 note on Model becoming a first-class entity later.
20 articles
Support vector machine
Maximum-margin classifier or regressor, optionally kernelised.
13 articles
Recurrent neural network
Sequential architecture with recurrent state.
12 articles
Autoencoder
Encoder/decoder trained to reconstruct its input.
10 articles
Probabilistic graphical model
A model of how many variables depend on one another, written as a graph of conditional dependencies — Markov random fields, Bayesian networks, maximum-entropy (Potts) models.
7 articles
Diffusion model
Iterative denoising generative model.
4 articles
Large language model
General-purpose text model applied as a component of the workflow.
3 articles
Normalising flow
Invertible transform between a simple and a complex distribution.
2 articles
Generative adversarial network
Generator trained against a discriminator.
1 articles
Symbolic regression
Search over closed-form expressions fitting the data.