Learning vector quantization (LVQ) is a popular class of adaptive nearest prototype classifiers for multiclass classification, but learning algorithms from this family have so far been proposed on heuristic grounds. Here, we take a more principled approach and derive two variants of LVQ using a gaussian mixture ansatz. We propose an objective function based on a likelihood ratio and derive a learning rule using gradient descent. The new approach provides a way to extend the algorithms of the LVQ family to different distance measure and allows for the design of "soft" LVQ algorithms. Benchmark results show that the new methods lead to better classification performance than LVQ 2.1. An additional benefit of the new method is that model assumptions are made explicit, so that the method can be adapted more easily to different kinds of problems.

No abstract

Abnormalities across different domains of neuropsychological functioning may constitute a risk factor for heavy drinking during adolescence and for developing alcohol use disorders later in life. However, the exact nature of such multi-domain risk profiles is unclear, and it is further unclear whether these risk profiles differ between genders. We combined longitudinal and cross-sectional analyses on the large IMAGEN sample (N ≈ 1000) to predict heavy drinking at age 19 from gray matter volume as well as from psychosocial data at age 14 and 19-for males and females separately. Heavy drinking was associated with reduced gray matter volume in 19-year-olds' bilateral ACC, MPFC, thalamus, middle, medial and superior OFC as well as left amygdala and anterior insula and right inferior OFC. Notably, this lower gray matter volume associated with heavy drinking was stronger in females than in males. In both genders, we observed that impulsivity and facets of novelty seeking at the age of 14 and 19, as well as hopelessness at the age of 14, are risk factors for heavy drinking at the age of 19. Stressful life events with internal (but not external) locus of control were associated with heavy drinking only at age 19. Personality and stress assessment in adolescents may help to better target counseling and prevention programs. This might reduce heavy drinking in adolescents and hence reduce the risk of early brain atrophy, especially in females. In turn, this could additionally reduce the risk of developing alcohol use disorders later in adulthood.

In alcohol dependence, individual prediction of treatment outcome based on neuroimaging endophenotypes can help to tailor individual therapeutic offers to patients depending on their relapse risk. We built a prediction model for prospective relapse of alcohol-dependent patients that combines structural and functional brain images derived from an experiment in which 46 subjects were exposed to alcohol-related cues. The patient group had been subdivided post hoc regarding relapse behavior defined as a consumption of more than 60 g alcohol for male or more than 40 g alcohol for female patients on one occasion during the 3-month assessment period (16 abstainers and 30 relapsers). Naïve Bayes, support vector machines and learning vector quantization were used to infer prediction models for relapse based on the mean and maximum values of gray matter volume and brain responses on alcohol-related cues within a priori defined regions of interest. Model performance was estimated by leave-one-out cross-validation. Learning vector quantization yielded the model with the highest balanced accuracy (79.4 percent, p < 0.0001; 90 percent sensitivity, 68.8 percent specificity). The most informative individual predictors were functional brain activation features in the right and left ventral tegmental areas and the right ventral striatum, as well as gray matter volume features in left orbitofrontal cortex and right medial prefrontal cortex. In contrast, the best pure clinical model reached only chance-level accuracy (61.3 percent). Our results indicate that an individual prediction of future relapse from imaging measurement outperforms prediction from clinical measurements. The approach may help to target specific interventions at different risk groups.

No abstract

Real-time load information in public transport is of high importance for both passengers and service providers. Neural algorithms have shown a high performance on various object counting tasks and play a continually growing methodological role in developing automated passenger counting systems. However, the publication of public-space video footage is often contradicted by legal and ethical considerations to protect the passengers' privacy. This work proposes an end-to-end Long Short-Term Memory network with a problem-adapted cost function that learned to count boarding and alighting passengers on a publicly available, comprehensive dataset of approx. 13,000 manually annotated low-resolution 3D LiDAR video recordings (depth information only) from the doorways of a regional train. These depth recordings do not allow the identification of single individuals. For each door opening phase, the trained models predict the correct passenger count (ranging from 0 to 67) in approx. 96% of boarding and alighting, respectively. Repeated training with different training and validation sets confirms the independence of this result from a specific test set.

No abstract

No abstract