Handling Probabilistic Data Ander de Keijzer Scalable Uncertainty - PowerPoint PPT Presentation

Handling Probabilistic Data Ander de Keijzer Scalable Uncertainty Management

Medical domain § Physicians are inference systems § Physicians working together, means exchanging data But…. § Physicians really don’t like uncertainty!! Scalable Uncertainty Management

Brief overview of uncertain data § Data has an associated probability Patient (name, diagnosis*) Amy, (Epilepsy || Parkinson) Scalable Uncertainty Management

Brief overview of uncertain data § Data has an associated probability § Granularity of uncertainty depends on level with which the probability is associated Patient (name, diagnosis*) Patient (name, diagnosis*) Amy, (Epilepsy || Parkinson) Amy, (Epilepsy:.8 || Parkinson:.2) Scalable Uncertainty Management

Brief overview of uncertain data § Data has an associated probability § Granularity of uncertainty depends on level with which the probability is associated EEGFeatures(patient, feature) AccelData(patient, feature) Amy, FocalBurst Amy, ShortStride ? ? Patient (name, diagnosis*) Patient (name, diagnosis*) Amy, (Epilepsy || Parkinson) Amy, (Epilepsy:.8 || Parkinson:.2) Scalable Uncertainty Management

Brief overview of uncertain data § Data has an associated probability § Granularity of uncertainty depends on level with which the probability is associated EEGFeatures(patient, feature) AccelData(patient, feature) Amy, FocalBurst Amy, ShortStride ? ? 10 20 Patient (name, diagnosis*) Amy, (Epilepsy:.8 || Parkinson:.2) 30 λ (30,1)= 10; λ (30,2)=20 Scalable Uncertainty Management

Or… § Probabilities given as conditional dependence probabilities, which incorporate lineage. Epilepsy Parkinson FocalBurst ShortStride Falling Scalable Uncertainty Management

Or… . Falling α 1- α EEG Accel FocalBurst ShortStride Scalable Uncertainty Management

Discussion Scalable Uncertainty Management

Discussion § Where do the initial probabilities come from? Scalable Uncertainty Management

Discussion § Where do the initial probabilities come from? § Are they even probabilities?? Or is it confidence scores? Scalable Uncertainty Management

Discussion § Where do the initial probabilities come from? § Are they even probabilities?? Or is it confidence scores? § Lineage…or provenance….or inference rules?? § How to represent the data to end users? § Database researchers usually ignore this one! § Is uncertainty (in databases) only useful with large amounts of data? § Can we scale? § What is large scale? Scalable Uncertainty Management