Mark Herbster

Tracking the Best Expert

AbstractWe generalize the recent relative loss bounds for on-line algorithms where the additional loss of the algorithm on the whole sequence of examples over the loss of the best expert is bounded. The generalization allows the sequence to be partitioned into segments, and the goal is to bound the additional loss of the algorithm over the sum of the losses of the best experts for each segment. This is to model situations in which the examples change and different experts are best for certain segments of the sequence of examples. In the single segment case, the additional loss is proportional to log n, where n is the number of experts and the constant of proportionality depends on the loss function. Our algorithms do not produce the best partition; however the loss bound shows that our predictions are close to those of the best partition. When the number of segments is k+1 and the sequence is of length &ell, we can bound the additional loss of our algorithm over the best partition by

Online learning over graphs

O\left( {klogn + k\log \left( {{\ell \mathord{\left/ {\vphantom {\ell k}} \right. \kern-\nulldelimiterspace} k}} \right)} \right)

Tracking the best regressor

. For the case when the loss per trial is bounded by one, we obtain an algorithm whose additional loss over the loss of the best partition is independent of the length of the sequence. The additional loss becomes

Epigenetic differences in monozygotic twins discordant for major depressive disorder

O\left( {klogn + k\log \left( {{\ell \mathord{\left/ {\vphantom {\ell k}} \right. \kern-\nulldelimiterspace} k}} \right)} \right)

Highly polygenic architecture of antidepressant treatment response: Comparative analysis of SSRI and NRI treatment in an animal model of depression

, where L is the loss of the best partitionwith k+1 segments. Our algorithms for tracking the predictions of the best expert aresimple adaptations of Vovks original algorithm for the single best expert case. As in the original algorithms, we keep one weight per expert, and spend O(1) time per weight in each trial.

Epigenetic Differences In Monozygotic Twins Discordant For Major Depressive Disorder

Summary Background We investigated the potential of proteomic fingerprinting with mass spectrometric serum profiling, coupled with pattern recognition methods, to identify biomarkers that could improve diagnosis of tuberculosis. Methods We obtained serum proteomic profiles from patients with active tuberculosis and controls by surface-enhanced laser desorption ionisation time of flight mass spectrometry. A supervised machine-learning approach based on the support vector machine (SVM) was used to obtain a classifier that distinguished between the groups in two independent test sets. We used k-fold cross validation and random sampling of the SVM classifier to assess the classifier further. Relevant mass peaks were selected by correlational analysis and assessed with SVM. We tested the diagnostic potential of candidate biomarkers, identified by peptide mass fingerprinting, by conventional immunoassays and SVM classifiers trained on these data. Findings Our SVM classifier discriminated the proteomic profile of patients with active tuberculosis from that of controls with overlapping clinical features. Diagnostic accuracy was 94% (sensitivity 93·5%, specificity 94·9%) for patients with tuberculosis and was unaffected by HIV status. A classifier trained on the 20 most informative peaks achieved diagnostic accuracy of 90%. From these peaks, two peptides (serum amyloid A protein and transthyretin) were identified and quantitated by immunoassay. Because these peptides reflect inflammatory states, we also quantitated neopterin and C reactive protein. Application of an SVM classifier using combinations of these values gave diagnostic accuracies of up to 84% for tuberculosis. Validation on a second, prospectively collected testing set gave similar accuracies using the whole proteomic signature and the 20 selected peaks. Using combinations of the four biomarkers, we achieved diagnostic accuracies of up to 78%. Interpretation The potential biomarkers for tuberculosis that we identified through proteomic fingerprinting and pattern recognition have a plausible biological connection with the disease and could be used to develop new diagnostic tests.