Model explorer
One card per model: inputs, outputs, objective, assumptions, hyperparameters, failure modes, and how it compares with its neighbors. Filter by what you need.
29 of 29 models
AutoencoderSqueeze data through a narrow bottleneck and learn to rebuild it, so the bottleneck has to capture what matters.unsupervisedself superviseddimensionality reductiongenerativeConvolutional neural networkSlide small, shared pattern detectors across the image; stack layers so edges become textures, textures become parts, parts become objects.supervisedclassificationvisionDBSCANA point in a crowded neighborhood is a core point; clusters grow by chaining core points that are close to each other; isolated points are noise.unsupervisedclusteringDecision treeA learned flowchart of yes/no questions, each chosen to make the groups below it as pure as possible.supervisedclassificationregressioninterpretablelow latencysmall dataDiffusion modelLearn to undo noise one small step at a time; then start from pure noise and denoise your way to a realistic sample.self supervisedgenerativevisionEmbedding modelMap items to points in space so that similar meanings land close together; compare by cosine similarity.self supervisednlplow latencyGaussian mixture modelThe data comes from K overlapping bell-shaped clouds, each with its own center, spread, and orientation; every point gets a probability of belonging to each cloud.unsupervisedclusteringgenerativesmall dataGradient boostingStart with a simple guess, then repeatedly add a small tree that fixes part of the remaining error, each contribution scaled down by a learning rate.supervisedclassificationregressionGRUAn LSTM simplified: an update gate decides how much of the old state to keep, and a reset gate decides how much of it to use when proposing a new state. No separate cell state.supervisedclassificationregressionnlplow latencyHierarchical clusteringStart with every point alone and repeatedly merge the two closest groups, recording each merge, until everything is one group.unsupervisedclusteringinterpretablesmall dataK-meansDrop K pins; each point joins its nearest pin; each pin moves to the middle of its group; repeat until nothing moves.unsupervisedclusteringinterpretablelow latencyK-nearest neighborsTo label something new, find the K most similar examples you've already seen and let them vote. The training data is the model.supervisedclassificationregressioninterpretablesmall dataLarge language modelA very large decoder-only transformer trained to predict the next token, then tuned to follow instructions. It produces probable text, which is not the same as true text.self supervisedgenerativenlpLasso regressionLinear regression that charges a fee proportional to each weight's absolute size, so unhelpful weights are pushed all the way to zero and dropped.supervisedregressioninterpretablelow latencysmall dataLinear regressionFit the flat line (or hyperplane) that best explains the target as a weighted sum of features. Each weight is "how much changes per unit of that feature, holding the others fixed."supervisedregressioninterpretablelow latencysmall dataLogistic regressionCompute a linear score, then squeeze it into a probability between 0 and 1 with the sigmoid. Each weight shifts the log-odds.supervisedclassificationinterpretablelow latencysmall datanlpLSTMAn RNN with a conveyor belt (the cell state) and gates that decide what to forget, what to write, and what to reveal, so information and gradients can travel many steps.supervisedclassificationregressionnlpMultilayer perceptronA stack of layers that each mix their inputs with weights, add a bias, and bend the result with a nonlinearity, building more useful representations layer by layer.supervisedclassificationregressionNaive BayesStart from how common each class is, multiply by how typical each observed feature is of that class, and pick the class with the bigger product. Features are assumed independent given the class.supervisedclassificationnlplow latencysmall datainterpretablePCARotate the coordinate system so the first axis points where the data varies most, the next where it varies most at a right angle, and keep only the first few axes.unsuperviseddimensionality reductioninterpretablelow latencyRandom forestAsk many independent, slightly different experts (deep trees trained on bootstrap samples with random feature subsets) and average their answers so individual quirks cancel out.supervisedclassificationregressionsmall dataRecurrent neural networkRead one step at a time and keep a running summary (the hidden state), updated with the same weights at every step.supervisedclassificationregressionnlpRerankerThe retriever quickly finds a few dozen plausible candidates; the reranker reads each one alongside the question, carefully, and puts the best on top.supervisednlpRidge regressionLinear regression with a fee on large weights: it shrinks every coefficient smoothly toward zero, spreading weight across correlated features instead of letting them fight.supervisedregressioninterpretablelow latencysmall dataSupport vector machineChoose the separating boundary with the widest empty street around it; only the points on the street's edges (support vectors) matter. Kernels let the street curve.supervisedclassificationregressionnlpsmall dataTransformer decoderA transformer where each token may only look at tokens before it (a causal mask), trained to predict what comes next.self supervisedgenerativenlpTransformer encoderEvery token looks at every other token in both directions, repeatedly, so each token's vector comes to reflect its full context.self supervisedsupervisedclassificationnlpTwo-tower modelPlace users and items in the same space so that a user sits close to the items they will engage with. Because items never interact with users before the final dot product, every item vector can be computed ahead of time and searched with an approximate nearest neighbor index.supervisedlow latencyXGBoostGradient boosting engineered for accuracy and speed: each new tree is chosen using both the gradient and the curvature of the loss, with explicit regularization on tree complexity.supervisedclassificationregression