- Concept Drift description on Wikipedia
- A. Tsymbal, The problem of concept drift: definitions and related work local
- G. Forman, Tackling Concept Drift by Temporal Inductive Transfer local
- M. Masud, Mining Concept-Drifting Data Stream to Detect Peer-to-Peer Botnet Traffic local
- EDDM-IWKDDS-2006, M. Garcia, Early Drift Detection Method in The International Workshop on Knowledge Discovery from Data Stream, code and datasets
- Weka, Single Classifier Drift
- J. Gama, A Survey on Concept Drift Adaptation local
- R. Polikar,Guest Editorial Learning in Nonstationary and Evolving Environments local content
- I. Zliobaite et al., Active Learning with Drifting Streaming Data local
- M. Wozniak et al., Active Learning Classification With Drifted Streaming Data local
- S. Janardan et al., Concept drift in Streaming Data Classification: Algorithms, Platforms and Issues, overview
- S. Zubin et al., Concept Drift Detection and Adaptation with Hierarchical Hypothesis Testing
- J. Gama's workgroup, explore around for papers
- Classic IDSVM
- solution for two opposite patterns
- insert/remove a pattern from the solution
- compare g() for polynomial kernel
- insert 500 USPS patterns (kernel)
- remove 480 USPS patterns (from now on, kernel only)
- adapt code to be device-independent (CUDA is faster)
- reduce C from 5.0 to 1.0 and observe error vectors, repeating learn/unlearn 500 patterns
- Weighted IDSVM
- make C individual for each pattern, adapt Migration and extend its unit test
- also extend Migration unit test for lambda
- vary C for a specific pattern on basic 2-vector solution, probing for SVs, EVs and RVs
- define rectangular shift kernel and probe for a window of 20-40 patterns
- define a linear shift kernel (timing info is needed here, check bibliography)
- define an exponential shift kernel (same as above)
- comparison with classic SVM
- time comparison for Forest Covertype
- add bibliography
- add latex template with todonotes package
- define main structure
- add theoretical considerations on incr/decr SVM
- refine for weighted SVM section
- final draft