A5038199211- Image Retrieval and Classification Techniques
- Medical Image Segmentation Techniques
- Neural Networks and Applications
- Advanced Image and Video Retrieval Techniques
- Bayesian Methods and Mixture Models
- Advanced Clustering Algorithms Research
- Face and Expression Recognition
- Machine Learning and Algorithms
- Cell Image Analysis Techniques
- Remote-Sensing Image Classification
- Neural dynamics and brain function
- Gaussian Processes and Bayesian Inference
- Complexity and Algorithms in Graphs
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Data Management and Algorithms
- Advanced Proteomics Techniques and Applications
- AI in cancer detection
- Blind Source Separation Techniques
- Sparse and Compressive Sensing Techniques
- Statistical Methods and Inference
- Radiomics and Machine Learning in Medical Imaging
- Gene expression and cancer classification
- Image and Signal Denoising Methods
- Advanced Vision and Imaging
ETH Zurich
2015-2024
Institute for Biomedical Engineering
2024
University of Zurich
2009-2021
Board of the Swiss Federal Institutes of Technology
2006-2018
École Polytechnique Fédérale de Lausanne
2003-2014
Data61
2013
SIB Swiss Institute of Bioinformatics
2013
Institute for Systems Biology
2009
Life Science Zurich
2009
Center for Pediatric Endocrinology Zurich
2009
An object recognition system based on the dynamic link architecture, an extension to classical artificial neural networks (ANNs), is presented. The architecture exploits correlations in fine-scale temporal structure of cellular signals group neurons dynamically into higher-order entities. These entities represent a rich and can code for high-level objects. To demonstrate capabilities program was implemented that recognize human faces other objects from video images. Memorized are represented...
Evaluating the performance of a classification algorithm critically requires measure degree to which unseen examples have been identified with their correct class labels. In practice, generalizability is frequently estimated by averaging accuracies obtained on individual cross-validation folds. This procedure, however, problematic in two ways. First, it does not allow for derivation meaningful confidence intervals. Second, leads an optimistic estimate when biased classifier tested imbalanced...
Data clustering describes a set of frequently employed techniques in exploratory data analysis to extract "natural" group structure data. Such groupings need be validated separate the signal from spurious structure. In this context, finding an appropriate number clusters is particularly important model selection question. We introduce measure cluster stability assess validity model. This quantifies reproducibility solutions on second sample, and it can interpreted as classification risk with...
Partitioning a data set and extracting hidden structure from the arises in different application areas of pattern recognition, speech image processing. Pairwise clustering is combinatorial optimization method for grouping which extracts proximity data. We describe deterministic annealing approach to pairwise shares robustness properties maximum entropy inference. The resulting Gibbs probability distributions are estimated by mean-field approximation. A new structure-preserving algorithm...
To stimulate progress in automating the reconstruction of neural circuits, we organized first international challenge on 2D segmentation electron microscopic (EM) images brain. Participants submitted boundary maps predicted for a test set images, and were scored based their agreement with consensus human expert annotations. The winning team had no prior experience EM employed convolutional network. This "deep learning" approach has since become accepted as standard images. continued to...
Comprehensive characterization of a proteome is fundamental goal in proteomics. To achieve saturation coverage or specific subproteome via tandem mass spectrometric identification tryptic protein sample digests, proteomics data sets are growing dramatically size and heterogeneity. The trend toward very large integrated poses so far unsolved challenges to control the uncertainty identifications going beyond well established confidence measures for peptide-spectrum matches. We present MAYU,...
In this paper we present a deep neural network topology that incorporates simple to implement transformationinvariant pooling operator (TI-POOLING). This is able efficiently handle prior knowledge on nuisance variations in the data, such as rotation or scale changes. Most current methods usually make use of dataset augmentation address issue, but requires larger number model parameters and more training results significantly increased time chance under-or overfitting. The main reason for...
Wheel defects on railway wagons have been identified as an important source of damage to the infrastructure and rolling stock. They also cause noise vibration emissions that are costly mitigate. We propose two machine learning methods automatically detect these wheel defects, based vertical force measured by a permanently installed sensor system network. Our learn different types predict during normal operation if has defect or not. The first method is novel features for classifying time...
This paper empirically compares nine image dissimilarity measures that are based on distributions of color and texture features summarizing over 1,000 CPU hours computational experiments. Ground truth is collected via a novel random sampling scheme for an partitioning method texture. Quantitative performance evaluations given classification, retrieval, segmentation tasks, wide variety measures. It demonstrated how the selection measure, large scale evaluation, substantially improves quality...
In this paper we propose and examine non-parametric statistical tests to define similarity homogeneity measures for textures. The are applied the coefficients of images filtered by a multi-scale Gabor filter bank. We demonstrate that these useful both, texture based image retrieval unsupervised segmentation, hence offer unified approach closely related tasks. present results on Brodatz-like micro-textures collection real-word images.
Perceptual grouping organizes image parts in clusters based on psychophysically plausible similarity measures. We propose a novel method this paper, which stresses connectedness of elements via mediating rather than favoring high mutual similarity. This principle yields superior clustering results when objects are distributed low-dimensional extended manifolds feature space, and not as local point clouds. In addition to extracting connected structures, singled out outliers they too far away...
Hidden Markov models (HMMs) are increasingly being used in computer vision for applications such as: gesture analysis, action recognition from video, and illumination modeling. Their use involves an off-line learning step that is as a basis on-line decision making (i.e. stationarity assumption on the model parameters). But, real-world often non-stationary nature. This leads to need dynamic mechanism learn update topology well its parameters. paper presents new framework HMM parameter...
A resampling scheme for clustering with similarity to bootstrap aggregation (bagging) is presented. Bagging used improve the quality of path-based clustering, a data method that can extract elongated structures from in noise robust way. The results an agglomerative optimization are influenced by small fluctuations input data. To increase reliability solutions, stochastic developed infer consensus clusters. related measure allows us estimate number clusters, based on stability optimized...
We present a novel optimization framework for unsupervised texture segmentation that relies on statistical tests as measure of homogeneity. Texture is formulated data clustering problem based sparse proximity data. Dissimilarities pairs textured regions are computed from multiscale Gabor filter image representation. discuss and compare class objective functions which systematically derived invariance principles. As general framework, we propose deterministic annealing mean-field...
The goal of this paper is to show how modify associative memory such that it can discriminate several stored patterns in a composite input and represent them simultaneously. Segmention takes place the temporal domain, components one pattern becoming temporally correlated with each other anticorrelated all patterns. Correlations are created naturally by usual connections. In our simulations, take form oscillatory bursts activity. Model oscillators consist pairs local cell populations...
Pain is known to comprise sensory, cognitive, and affective aspects. Despite numerous previous fMRI studies, however, it remains open which spatial distribution of activity sufficient encode whether a stimulus perceived as painful or not. In this study, we analyzed data from perceptual decision-making task in participants were exposed near-threshold laser pulses. Using multivariate analyses on different scales, investigated the predictive capacity for decoding had been painful. Our analysis...
De novo sequencing of peptides poses one the most challenging tasks in data analysis for proteome research. In this paper, a generative hidden Markov model (HMM) mass spectra de peptide which constitutes novel view on how to solve problem Bayesian framework is proposed. Further extensions structure graphical and factorial HMM substantially improve identification results are demonstrated. Inference with estimates posterior probabilities amino acids rather than scores single symbols sequence....
A key barrier to the realization of personalized medicine for cancer is identification biomarkers. Here we describe a two-stage strategy discovery serum biomarker signatures corresponding specific cancer-causing mutations and its application prostate (PCa) in context commonly occurring phosphatase tensin homolog ( PTEN ) tumor-suppressor gene inactivation. In first stage our approach, identified 775 N -linked glycoproteins from sera tissue wild-type Pten -null mice. Using label-free...