spike-and-slab
A classic prior structure in Bayesian variable selection, formed by mixing a spike (a degenerate distribution with zero coefficient) and a slab (a wide diffuse distribution with nonzero coefficient). Each variable carries a prior inclusion weight controlling its probability of being selected; group spike-and-slab can be extended to variable groups or interaction terms.
Timeline (last 90 days)
SelectPFN is pretrained in a one-off manner on synthetic tasks through a meta-prior built around a group spike-and-slab sparse input neural network.
Spike-and-slab is a classic prior structure in Bayesian variable selection, in which the spike corresponds to a degenerate distribution with a coefficient of zero and the slab is a wide diffuse distribution when the coefficient is nonzero; group spike-and-slab extends this idea to variable groups or interaction terms.
All Facts (2)
SelectPFN is pretrained in a one-off manner on synthetic tasks through a meta-prior built around a group spike-and-slab sparse input neural network.
50%UnverifiedSpike-and-slab is a classic prior structure in Bayesian variable selection, in which the spike corresponds to a degenerate distribution with a coefficient of zero and the slab is a wide diffuse distribution when the coefficient is nonzero; group spike-and-slab extends this idea to variable groups or interaction terms.
50%