Dominik Brugger

Automatic Cluster Detection in Kohonen's SOM

Kohonens self-organizing map (SOM) is a popular neural network architecture for solving problems in the field of explorative data analysis, clustering, and data visualization. One of the major drawbacks of the SOM algorithm is the difficulty for nonexpert users to interpret the information contained in a trained SOM. In this paper, this problem is addressed by introducing an enhanced version of the Clusot algorithm. This algorithm consists of two main steps: 1) the computation of the Clusot surface utilizing the information contained in a trained SOM and 2) the automatic detection of clusters in this surface. In the Clusot surface, clusters present in the underlying SOM are indicated by the local maxima of the surface. For SOMs with 2-D topology, the Clusot surface can, therefore, be considered as a convenient visualization technique. Yet, the presented approach is not restricted to a certain type of 2-D SOM topology and it is also applicable for SOMs having an n-dimensional grid topology.

Vibrotactile Discrimination in the Rat Whisker System is Based on Neuronal Coding of Instantaneous Kinematic Cues

Which physical parameter of vibrissa deflections is extracted by the rodent tactile system for discrimination? Particularly, it remains unclear whether perception has access to instantaneous kinematic parameters (i.e., the details of the trajectory) or relies on temporally integration of the movement trajectory such as frequency (e.g., spectral information) and intensity (e.g., mean speed). Here, we use a novel detection of change paradigm in head-fixed rats, which presents pulsatile vibrissa stimuli in seamless sequence for discrimination. This procedure ensures that processes of decision making can directly tap into sensory signals (no memory functions involved). We find that discrimination performance based on instantaneous kinematic cues far exceeds the ones provided by frequency and intensity. Neuronal modeling based on barrel cortex single units shows that small populations of sensitive neurons provide a transient signal that optimally fits the characteristic of the subjects perception. The present study is the first to show that perceptual read-out is superior in situations allowing the subject to base perception on detailed trajectory cues, that is, instantaneous kinematic variables. A possible impact of this finding on tactile systems of other species is suggested by evidence for instantaneous coding also in primates.

Real-Time Adaptive Microstimulation Increases Reliability of Electrically Evoked Cortical Potentials

Cortical neuroprostheses that employ repeated electrical stimulation of cortical areas with fixed stimulus parameters, are faced with the problem of large trial-by-trial variability of evoked potentials. This variability is caused by the ongoing cortical signal processing, but it is an unwanted phenomenon if one aims at imprinting neural activity as precisely as possible. Here, we use local field potentials measured by one microelectrode, located at a distance of 200 microns from the stimulation site, to drive the electrically evoked potential toward a desired target potential by real-time adaptation of the stimulus intensity. The functional relationship between ongoing cortical activity, evoked potential, and stimulus intensity was estimated by standard machine learning techniques (support vector regression with problem-specific kernel function) from a set of stimulation trials with randomly varied stimulus intensities. The smallest deviation from the target potential was achieved for low stimulus intensities. Further, the observed precision effect proved time sensitive, since it was abolished by introducing a delay between data acquisition and stimulation. These results indicate that local field potentials contain sufficient information about ongoing local signal processing to stabilize electrically evoked potentials. We anticipate that adaptive low intensity microstimulation will play an important role in future cortical prosthetic devices that aim at restoring lost sensory functions.

Support for the slip hypothesis from whisker-related tactile perception of rats in a noisy environment.

Rodents use active whisker movements to explore their environment. The “slip hypothesis” of whisker-related tactile perception entails that short-lived kinematic events (abrupt whisker movements, called “slips”, due to bioelastic whisker properties that occur during active touch of textures) carry the decisive texture information. Supporting this hypothesis, previous studies have shown that slip amplitude and frequency occur in a texture-dependent way. Further, experiments employing passive pulsatile whisker deflections revealed that perceptual performance based on pulse kinematics (i.e., signatures that resemble slips) is far superior to the one based on time-integrated variables like frequency and intensity. So far, pulsatile stimuli were employed in a noise free environment. However, the realistic scenario involves background noise (e.g., evoked by rubbing across the texture). Therefore, if slips are used for tactile perception, the tactile neuronal system would need to differentiate slip-evoked spikes from those evoked by noise. To test the animals under these more realistic conditions, we presented passive whisker-deflections to head-fixed trained rats, consisting of “slip-like” events (waveforms mimicking slips occurring with touch of real textures) embedded into background noise. Varying the (i) shapes (ramp or pulse); (ii) kinematics (amplitude, velocity, etc.); and (iii) the probabilities of occurrence of slip-like events, we observed that rats could readily detect slip-like events of different shapes against noisy background. Psychophysical curves revealed that the difference of slip event and noise amplitude determined perception, while increased probability of occurrence (frequency) had barely any effect. These results strongly support the notion that encoding of kinematics dominantly determines whisker-related tactile perception while the computation of frequency or intensity plays a minor role.

Online SVR Training by Solving the Primal Optimization Problem

Online estimation of regression functions becomes important in presence of drifts and rapid changes in the training data. In this article we propose a new online training algorithm for SVR, called Priona, which is based on the idea of computing approximate solutions to the primal optimization problem. For the solution of the primal SVR problem we investigated the trade-off between computation time and prediction accuracy for the gradient, diagonally scaled gradient, and Newton descent direction. The choice of a particular buffering strategy did not influence the performance of the algorithm. By using a line search Priona does not require a priori selection of a learning rate which facilitates its practical application. On various benchmark data sets Priona is shown to perform better in terms of prediction accuracy in comparison to the Norma and Silk online SVR algorithms. Further, tests on two artificial data sets show that the online SVR algorithms are able to track temporal changes and drifts of the regression function, if the buffer size and learning rate are selected appropriately.

Estimation of diffusion coefficients from voltammetric signals by support vector and gaussian process regression.

BackgroundSupport vector regression (SVR) and Gaussian process regression (GPR) were used for the analysis of electroanalytical experimental data to estimate diffusion coefficients.ResultsFor simulated cyclic voltammograms based on the EC, Eqr, and EqrC mechanisms these regression algorithms in combination with nonlinear kernel/covariance functions yielded diffusion coefficients with higher accuracy as compared to the standard approach of calculating diffusion coefficients relying on the Nicholson-Shain equation. The level of accuracy achieved by SVR and GPR is virtually independent of the rate constants governing the respective reaction steps. Further, the reduction of high-dimensional voltammetric signals by manual selection of typical voltammetric peak features decreased the performance of both regression algorithms compared to a reduction by downsampling or principal component analysis. After training on simulated data sets, diffusion coefficients were estimated by the regression algorithms for experimental data comprising voltammetric signals for three organometallic complexes.ConclusionsEstimated diffusion coefficients closely matched the values determined by the parameter fitting method, but reduced the required computational time considerably for one of the reaction mechanisms. The automated processing of voltammograms according to the regression algorithms yields better results than the conventional analysis of peak-related data.

Are spatial frequency cues used for whisker-based active discrimination?

Rats are highly skilled in discriminating objects and textures by palpatory movements of their whiskers. If they used spatial frequency cues, they would be able to optimize performance in a stimulus dependent way—by moving their whisker faster or slower across the texture surface, thereby shifting the frequency content of the neuronal signal toward an optimal working range for perception. We tested this idea by measuring discrimination performance of head-fixed rats that were trained to actively sample from virtual grids. The virtual grid mimicked discrete and repetitive whisker deflections generated by real objects (e.g., grove patterns) with single electrical microstimulation pulses delivered directly to the barrel cortex, and provided the critical advantage that stimuli could be controlled at highest precision. Surprisingly, rats failed to use the spatial frequency cue for discrimination as a matter of course, and also failed to adapt whisking patterns in order to optimally exploit frequency differences. In striking contrast they could be easily trained to discriminate stimuli differing in electrical pulse amplitudes, a stimulus property that is not malleable by whisking. Intermingling these “easy-to-discriminate” discriminanda with others that solely offered frequency/positional cues, rats could be guided to base discrimination on frequency and/or position, albeit on a lower level of performance. Following this training, abolishment of electrical amplitude cues and reducing positional cues led to initial good performance which, however, was unstable and ran down to very low levels over the course of hundreds of trials. These results clearly demonstrate that frequency cues, while definitely perceived by rats, are of minor importance and they are not able to support consistent modulation of whisking patterns to optimize performance.

Online SVR training by solving the primal optimization problem

Online regression estimation becomes important in presence of drifts and rapid changes in the training data. In this article we propose a new online training algorithm for SVR, called PRIONA, which is based on the idea of computing approximate solutions to the primal optimization problem. We explore different unconstrained optimization methods for the solution of the primal SVR problem and investigate the impact of different buffering strategies. By using a line search PRIONA does not require a priori selection of a learning rate which facilitates its practical application. Further PRIONA is shown to perform better in terms of prediction accuracy on various benchmark data sets in comparison to the NORMA and SILK online SVR algorithms.