Datasets:
File size: 3,146 Bytes
6c1c833 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 | {
"field_id": "F024",
"field": "bayesian learning",
"directions": [
{
"direction_id": "G0",
"name": "Scalable and reusable posterior inference",
"scope": "Make Bayesian posterior computation feasible and effective for exploration, approximate optimization, reuse across datasets, or distributed execution, encompassing samplers, general-purpose variational rules, amortized inference networks, and collaborative inference under communication constraints."
},
{
"direction_id": "G1",
"name": "Uncertainty-bearing models and informative priors",
"scope": "Construct Bayesian models that express useful uncertainty or inductive information, including stochastic neural predictors, flexible probabilistic process models for functions, and priors that encode functional, structural, auxiliary-data, or stability information and are used in subsequent inference."
},
{
"direction_id": "G2",
"name": "Inference of model structure and effective complexity",
"scope": "Infer from data which model components should exist or play a major role, encompassing posterior inference or search over discrete structures such as graphs, trees, and feature subsets, as well as coefficient shrinkage, sparse estimation, and neural-network pruning."
},
{
"direction_id": "G3",
"name": "Bayesian learning across related units and time",
"scope": "When information is distributed across different tasks, domains, systems, or observations arriving over time, use existing beliefs and update inference while accounting for differences between units or changes over time; this includes selective sharing, hierarchical pooling, online filtering, and continual adaptation."
},
{
"direction_id": "G4",
"name": "Predictive reliability and Bayesian assessment",
"scope": "Assess or improve whether Bayesian predictions and the conclusions drawn from their assessment are trustworthy, encompassing how confidence relates to error or coverage, uncertainty for unfamiliar inputs, and the effects of approximations, temperature, and model assumptions on generalization and assessment."
},
{
"direction_id": "G5",
"name": "Inference robust to model mismatch",
"scope": "When assumptions about the likelihood, observation noise, or simulator are unreliable, modify Bayesian or generalized-Bayesian updates so that inference about parameters and functions remains useful."
},
{
"direction_id": "G6",
"name": "Bayesian recovery of physical dynamics",
"scope": "Treat the dynamical or physically constrained system itself as the object of inference, recovering uncertain states, parameters, model forms, or differential-equation solutions and quantifying the associated uncertainty."
},
{
"direction_id": "G7",
"name": "Bayesian model comparison and predictive combination",
"scope": "Assess uncertainty across proposed candidate models or components, marginalize over them, or combine their predictions, rather than conditioning only on a single specified model."
}
]
}
|