Completed
ProtoDash: Fast Interpretable Prototype Selection
Class Central Classrooms beta
YouTube videos curated by Class Central.
Classroom Contents
ProtoDash - Fast Interpretable Prototype Selection by Karthik Gurumoorthy
Automatically move to the next video in the Classroom when playback concludes
- 1 ProtoDash: Fast Interpretable Prototype Selection
- 2 Objective: Extract compact synapses of large data sets
- 3 Prototypes and Criticisms
- 4 Some applications:
- 5 Causal reasoning
- 6 Metric to quantify best for prototype selection
- 7 Reformulation as maximizing a set function
- 8 Properties of set function: Submodularity
- 9 Weak submodularity
- 10 ProtoDash algorithm: Greedily build the set based on gradients
- 11 ProtoGreedy algorithm: Greedily build the set based on maximum function increment
- 12 ProtoDash Vs ProtoGreedy
- 13 Classification accuracy Vs sparsity
- 14 Classification accuracy Vs skew
- 15 Prototype selection quality
- 16 Computation time
- 17 Visualizing selected prototypes
- 18 Visualizing selected prototypes from the same data set
- 19 Theoretical guarantees
- 20 Set function value Vs sparsity
- 21 Criticism selection
- 22 Greedy algorithm for choosing criticisms
- 23 Visualizing selected criticisms
- 24 Application 2: Finding relationships between data sets
- 25 Approach to identify relationships
- 26 Visualizing relationships
- 27 Human expert based evaluation
- 28 Application 3: Improving prediction accuracy by training data selection
- 29 Training data selection
- 30 Prediction accuracy
- 31 Interpreting selected prototype
- 32 Interpreting selected prototypes
- 33 Thank You
- 34 Q&A