Eszter I Lohn - Coma patients

"Notation" for abbreviations ecc
Description and preparation of the data:
First look at the relationship between expl var and score: things like plots, aggregate_plots, ecc.
Estimation: variable selection and predictive models
Evaluation of model-fitting procedures using CV: done patient-wise (more entries per patient) w/ RMSE and MAD; alternative: Bootstrap
CV Results: put a table and a list of comments (e.g. Lasso beats all of them: check and see that intercept model always selected: not informative variables)
Caution with within- and across-patients analysis
CV nested to find \(\lambda\) for Lasso removing extra patient Maindonald, 2006
Note that models needing tuning of some parameters could be too strict: need to tune par using restriction of data and might loss relationships of overall sample
RF has problem if only small subset of features is related (due to random selection at each branch): consider restricting to variables w/ significant correlation (Pearson and Spearman)
Caution: cannot do for linear model: fit on overall sample, overfitting (cheat), since corelation is measure of linear dependence; RF non-linear, not so bad; otherwise bias w/ linear models bias w/ linear mod
Best: do corr test inside CV to filter out vars
Weak cross-validation: "optimistic": remove var selection uncertainty since done on whole dataset

Anita Kaufmann - Crime Linkage

Chapter Literature Review: what others already did
Discussion of assumptions and Justifications

Solt Kovacks - Changepoint detection high dim cov matrix

Derivation of the proposal: considered problem and notation (assumptions, definitions, ...)
Standardization and resulting refinements:;
Approach 1: ...
Approach 2: ...
Preparations for the simulations: Models: 1. Chain, 2. Neigh Networks, ..
Performance measures
Choice of tuning param: BIC AIC, CV
Simulations: boxplots, barplots, ...
Real life example: Stock returns
Summary and future work

Christop Kovacs - Semi-supervisef label propagation

Classification Performance Mesures: Zero-one loss; accuracy, Hamming loss, F1, macreo- and micro-averaged F1
Multiple testing problem: multiple algo on multiple datasets (should not be the case)
repeated holdout and k-CV, bootstrap

S. Kessler - Composite training for TS

Contributions: what did this works helps for?
Related works

Kari Kolbeinsson - Model Selection for Predictions of Football Matches

The dataset: explanation of various variables, amount of data, of vars, when they will be used
Table explaining the dataset (example)
Response variable: what is: Result, Goal Difference
Explanatory vars: some boxplots, for any grouping to explain their effect
classification models: discriminant analysis [what it is, LDA, plot,subset selection,application,QDA], KNN, Tree-based methods[Bagging, RF,Boosting], SVM
regression models: GLM, Tree-based
ch: results, 

Beat Jäggi - Stat Analysis neuro impulses

Biologische Hintergrund: give an idea of stuff not known to a lot of people (e.g. MS, breath pattern, time ...)
State the idea at the beginning of Dataset + Experiment section

Christina Heinze - Random projections in high-dim

Notation and List of Acronyms divided in two