Additional, optional reading is also listed below. The references are resolved in the Course materials.
Key to notation:
Date |
Topic |
Lecture Notes | >
Homework/Reading Assignment | >
Optional Reading |
||||||||
Aug 25 |
Orientation, Introduction, Motivations |
Reading: Kohonen Ch3, due Aug 30 |
Haykin Ch9.1-9.6 |
|||||||||
Aug 27 | Review of the basic SOM | EMvis-2 | Reading: Kohonen Ch3, Kohonen, IEEE 1990
Refresh Competitive Learning from COMP / ELEC / STAT 502 Refresh SOM exercises from COMP / ELEC / STAT 502 |
Haykin Ch9.1-9.6 | ||||||||
Sep 1 | Review of approach to manifold learning with neural maps, EM
Topology Representing Networks |
EMvis-3 |
Refresh SOM visualizations from COMP / ELEC / STAT 502 Reading assignment: for class on Sep 3, 8; no reviews: Ultsch&Siemon, 1990; Cottrell&deBodt, 1995; Kraaijveld et al., 1995 Individual asignments, for Sep 8 presentations: Cottrell&deBodt, 1995; Ultsch et al. 2003, 2005 (U*, P, AU, AU* matrices) |
Haykin Ch9.1-9.6 | Sep 3 |
Topology Representing Networks, cont'd (EM)
CONN measure and visualization |
EMvis-4 |
reading assignment for Sep 10: Martinetz and Schulten, 1994 |
tr> | Sep 8 | Student presentations / discussion of SOM visualizations | DLvis, PODvis, JTvis |
reading assignment for Sep 15: CONN paper (Tasdemir and Merenyi, IEEE TNN 2009) |
tr> | Sep 10 |
Finish CONN lecture (EM);
discuss U, U*, P, AU, AU*, CONN (and other) visualizations we studied. |
JTvis |
reading assignment for Sep 15: All papers above |
tr> | Sep 15 | Intro to the NeuroScope environment |
reading assignment for Sep 22: U, U*, P, AU, AU*, CONN (and other) visualizations we studied, cont'd; TNN paper (Martinetz and Schulten, 1994), relation to visualizations, to expression of topology. |
tr> | Sep 17 | More NeuroScope, and test data |
(repeat) reading assignment for Sep 22: U, U*, P, AU, AU*, CONN (and other) visualizations we studied, cont'd; TNN paper (Martinetz and Schulten, 1994), relation to visualizations, to expression of topology. Review assignment: Martinetz and Schulten, 1994. due Sep 22 See review requirements, and template under "Grading and homework policy". New reading assignment: for class on Sep 24 Tennenbaum et al., 2000, no review; Sammon, 1969 Experiment assignment: Use the MATLAB function for Sammon's mapping, apply to one of the {uniform or 4 Gaussians} AND one of the {iris or 6-D synthetic spectral image}, compare results with KSOM. Do the same with the ISOMAP MATLAB implementation. Results to be discussed in class on Sep 24, 5-7 minutes per student. |
Sep 22 | Variants of SOMS: Conscience SOM, EM; |
EMvis |
reading assignment: DeSieno, 1988; NeuralWare, Neural Computation, SOM; Experiment assignment: Implement conscince SOM (or use NeuroScope module), learn the same data as you learned with KSOM, compare behavior and results. (Choose one or two of the uniform, 4 Gaussians, Iris, 6-D spectral cube data sets, or Ultsch data sets.) Results to be discussed in class on Sep 29, 5-7 minutes per student. |
|||||||||
Sep 24 | Variants of SOMs: scaled distances (Kangas et al. 1990), Presentation and discussion lead: DL Discussion of Sammons / ISOMAP / KSOM simulations. |
DLvis, Experiment slides | reading assignment:
Linde, Buzo, Gray, 1980 (for class on Oct 1) |
|||||||||
Sep 29 |
Discussion of Sammons / ISOMAP / KSOM simulations. Discussion of KSOM vs Conscience SOM simulations. SOM-hybrid classification EM |
EMvis | reading assignment:
Linde, Buzo, Gray, 1980 Exercise: K-means vs KSOM / CSOM sluctering. Presentation of results / insights Oct 6. |
|||||||||
Oct 1 |
Wrap up discussion of exercises, KSOM vs Conscience SOM Topology Preservation in SOMs, EM |
EMvis | reading assignment:
Villmann et al, 1997; Bauer and Pawelzik, 1992 |
Oct 6 |
LBG (K-means): comparison with SOM (discussion, lead POD) |
PODvis | reading assignment:
Villmann et al, 1997; Bauer and Pawelzik, 1992 Zhang and Merenyi, ESANN 2006; |
|||||
Oct 8 |
Topology Preservation in SOMs (TP, TF, WDTF and discussion, lead JT) |
JTvis | reading assignment:
Villmann et al, 1997; Bauer and Pawelzik, 1992 Zhang and Merenyi, ESANN 2006 |
|||||||||
Oct 13 |
Fall recess, NO CLASS |
|||||||||||
Oct 15 |
EM out of town. Learn cluster extraction, TF, supervised SOM-hybrid classification, and evaluation of classification accuracy in NeuroScope. (remap, TF, ann-SOMconsc, class_acc_eval; clstat, vecplot) |
POD tutorial | reading assignment: for upcoming classes on SOM magnification: Bauer, Der, Hermann, 1996; Ritter & Schulten, 1986
Experiments: cluster extraction; SOM-hybrid classification. Data? | |||||||||
Oct 20 | Magnification in SOMs. Introduction, EM. | EMvis | reading assignment for SOM magnification (Oct 20 - Nov 6): Bauer, Der, Hermann, 1996; Ritter and Schulten, 1986; Ritter, 1991; Dersch and Tavan, 1995; Merenyi et al., 2007 | |||||||||
Oct 22 | Magnification in SOMs. The stationary state of the KSOM, Ritter and Schulten, 1986, DL. | DLvis | reading assignment for SOM magnification (Oct 20 - Nov 6): Bauer, Der, Hermann, 1996; Ritter and Schulten, 1986; Ritter, 1991; Dersch and Tavan, 1995; Merenyi et al., 2007 | |||||||||
Oct 25 | Magnification in SOMs. Controlling the magnification, "BDH paper". POD; Magnification for "forbidden data", Merenyi et al, 2007, EM. | PODvis | reading assignment for SOM magnification (Oct 20 - Nov 6): Bauer, Der, Hermann, 1996; Ritter and Schulten, 1986; Ritter, 1991; Dersch and Tavan, 1995; Merenyi et al., 2007 | |||||||||
Oct 27 |
Discussion of reviews; magnification issues. Leftover case studies with SOM-hybrid claassification and regression, EM. |
EMvis | reading assignment for SOM magnification (Oct 20 - Nov 6): Bauer, Der, Hermann, 1996; Ritter and Schulten, 1986; Ritter, 1991; Dersch and Tavan, 1995; Merenyi et al., 2007 | |||||||||
Nov 3 |
Magnification: Ritter, 1991; Dersch and Tavan, 1995. Presentation and discussion lead: JT |
JTvis | reading assignment for GRLVQ (Nov 10 - Nov 12): Hammer and Villmann, 2002; optional: Mendenhall and Merenyi, 2008
Review assignment: Hammer and Villmann, 2002, due Nov 10. | |||||||||
Nov 5 |
Review of past material: Self-assessment discussion, EM lead. GRLVQ introduction, EM |
EMvis | reading assignment for GRLVQ (Nov 10 - Nov 12): Hammer and Villmann, 2002; optional: Mendenhall and Merenyi, 2008 | |||||||||
Nov 10 |
GRLVQ, Hammer and Villmann, 2002. Presentation and discussion lead: DL Guest: Alex Wang, GRLVQ experiments with iris. (To be confirmed for Nov 10 or 12.) |
DLvis | reading assignment for GRLVQ (Nov 10 - Nov 12): Hammer and Villmann, 2002; optional: Mendenhall and Merenyi, 2008 Experiment assignment: Run GRLVQ experiments on the iris data, and on a fully labeled 6-d synthetic cube. Present your observations on Nov 17, 10 minutes. | |||||||||
Nov 12 |
High-dimensional GRLVQ case study, open questions. EM. GSOM, Bauer and Villmann, TNN 1997, Presentation and discussion lead: POD (Note that reading the GSOM paper ahead of time is not a requirement for this class.) |
EMvis, PODvis | reading assignment for Cluster Validity Indices (Nov 17 - Nov 19): Tentative list, to be settled: Bezdek and Pal, 1998; DBI, GDI (use of in Vesanto and Alhoniemi, 2000; PBM, CDbw, and others, CONN_index, Tasdemir and Merenyi, 2011 Review assignment: Bauer and Villmann, TNN 1997, due Nov 24. | |||||||||
Nov 17 |
Presentation of GRLVQ experiments. Leftover issues. Project discussion. |
|||||||||||
Nov 19 |
Cluster Validity Indices, Presentation and discussion lead: JT Experiment guide: EM, POD. |
JTvis, EMvis, PODvis | Experiment assignment: Run CVI experiments on data with clusters of varied properties, such as the Clown, Lsun, Wingnuts. Compare the judgement of the indexes. (In NeuroScope, the module CVI has several indexes implemented. The 6-d synthetic cubes are also good data for this.) Present experiments on December 1. | |||||||||
Nov 24 |
NO CLASS (Allowance for projects) |
|||||||||||
Nov 26 |
NO CLASS (Thanksgiving) |
Dec 3 |
CVI experiment presentations Leftover issues. |
|||||||||
Dec 3 |
Class project presentations |
Approx. 15 min + 5 minutes for discussion, for each talk Turn in a soft copy of your presentation and your project report by Dec 4. |
Return to ELEC 602 / COMP 602 home page