Thomas Hofmann

ORCID: 0000-0003-4057-7165
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Biochemical Analysis and Sensing Techniques
  • Advanced Chemical Sensor Technologies
  • Fermentation and Sensory Analysis
  • Olfactory and Sensory Function Studies
  • Phytochemicals and Antioxidant Activities
  • Academic Writing and Publishing
  • Neural Networks and Applications
  • Medical Practices and Rehabilitation
  • Advanced Glycation End Products research
  • Natural Language Processing Techniques
  • Topic Modeling
  • Coffee research and impacts
  • Tea Polyphenols and Effects
  • Health and Medical Studies
  • Physics and Engineering Research Articles
  • Image Retrieval and Classification Techniques
  • Medical and Health Sciences Research
  • Machine Learning and Algorithms
  • Engineering and Materials Science Studies
  • Generative Adversarial Networks and Image Synthesis
  • Stochastic Gradient Optimization Techniques
  • Face and Expression Recognition
  • Gaussian Processes and Bayesian Inference
  • Meat and Animal Product Quality
  • Sparse and Compressive Sensing Techniques

Technical University of Munich
2016-2025

Hochschule Osnabrück
2020-2025

Membrane Technology & Research (United States)
2022-2025

Leibniz-Institute for Food Systems Biology at the Technical University of Munich
2015-2024

Bavarian State Research Center for Agriculture
2016-2024

Technical University of Darmstadt
2005-2024

ETH Zurich
2014-2023

Universidad Nacional de Colombia
2023

University of Stuttgart
2023

University Hospital and Clinics
2022

Probabilistic Latent Semantic Indexing is a novel approach to automated document indexing which based on statistical latent class model for factor analysis of count data. Fitted from training corpus text documents by generalization the Expectation Maximization algorithm, utilized able deal with domain{specific synonymy as well polysemous words. In contrast standard (LSI) Singular Value Decomposition, probabilistic variant has solid foundation and defines proper generative data model....

10.1145/3130348.3130370 article EN ACM SIGIR Forum 2017-08-02

Article Free AccessProbabilistic latent semantic indexing Author: Thomas Hofmann International Computer Science Institute, Berkeley, CA & EECS Department, CS Division, UC Berkeley BerkeleyView Profile Authors Info Claims SIGIR '99: Proceedings of the 22nd annual international ACM conference on Research and development in information retrievalAugust 1999 Pages 50–57https://doi.org/10.1145/312624.312649Published:01 August 1999Publication History 3,242citation13,366DownloadsMetricsTotal...

10.1145/312624.312649 article EN 1999-08-01

10.1023/a:1007617005950 article EN Machine Learning 2001-01-01

We review machine learning methods employing positive definite kernels. These formulate and estimation problems in a reproducing kernel Hilbert space (RKHS) of functions defined on the data domain, expanded terms kernel. Working linear spaces function has benefit facilitating construction analysis algorithms while at same time allowing large classes functions. The latter include nonlinear as well nonvectorial data. cover wide range methods, ranging from binary classifiers to sophisticated...

10.1214/009053607000000677 article EN The Annals of Statistics 2008-05-26

Collaborative filtering aims at learning predictive models of user preferences, interests or behavior from community data, that is, a database available preferences. In this article, we describe new family model-based algorithms designed for task. These rely on statistical modelling technique introduces latent class variables in mixture model setting to discover communities and prototypical interest profiles. We investigate several variations deal with discrete continuous response as well...

10.1145/963770.963774 article EN ACM transactions on office information systems 2004-01-01

Learning general functional dependencies is one of the main goals in machine learning. Recent progress kernel-based methods has focused on designing flexible and powerful input representations. This paper addresses complementary issue problems involving complex outputs such as multiple dependent output variables structured spaces. We propose to generalize multiclass Support Vector Machine learning a formulation that involves features extracted jointly from inputs outputs. The resulting...

10.1145/1015330.1015341 article EN 2004-01-01

Hormones, neurotransmitters, and growth factors give rise to calcium entry via receptor-activated cation channels that are activated downstream of phospholipase C activity. Members the transient receptor potential channel (TRPC) family have been characterized as molecular substrates mediating influx. TRPC assumed be composed multiple proteins. However, cellular principles governing assembly proteins into homo- or heteromeric ion still remain elusive. By pursuing four independent experimental...

10.1073/pnas.102596199 article EN Proceedings of the National Academy of Sciences 2002-05-14

Most successful object recognition systems rely on binary classification, deciding only if an is present or not, but not providing information the actual location. To perform localization, one can take a sliding window approach, this strongly increases computational cost, because classifier function has to be evaluated over large set of candidate subwindows. In paper, we propose simple yet powerful branch-and-bound scheme that allows efficient maximization class functions all possible...

10.1109/cvpr.2008.4587586 article EN 2009 IEEE Conference on Computer Vision and Pattern Recognition 2008-06-01

Recently, bioresponse-guided fractionation of black tea infusions indicated that neither the high molecular weight thearubigens nor theaflavins, but a series 14 flavon-3-ol glycopyranosides besides some catechins, might be important contributors to taste. To further bridge gap between pure structural chemistry and human taste perception, in present investigation 51 putative compounds have been quantified infusion, their dose-over-threshold (Dot) factors calculated on basis dose/threshold...

10.1021/jf050294d article EN Journal of Agricultural and Food Chemistry 2005-06-01

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...

10.1109/34.566806 article EN IEEE Transactions on Pattern Analysis and Machine Intelligence 1997-01-01

Application of taste dilution analyses on freshly prepared black tea infusions revealed neither the high molecular weight thearubigen-like polyphenols nor catechins and theaflavins, but a series 14 flavon-3-ol glycosides as main contributors to astringent perceived upon consumption. Among these glycosides, apigenin-8-C-[α-l-rhamnopyranosyl-(1→2)-O-β-d-glucopyranoside] was identified for first time in infusions. Depending structure, were found induce velvety mouth-coating sensation at very...

10.1021/jf049802u article EN Journal of Agricultural and Food Chemistry 2004-05-05

Mammalian transient receptor potential channels (TRPCs) form a family of Ca2+-permeable cation currently consisting seven members, TRPC1–TRPC7. These have been proposed to be molecular correlates for capacitative Ca2+ entry channels. There are only few studies on the regulation and properties subfamily TRPC4 TRPC5, there contradictory reports concerning possible role intracellular store depletion in channel activation. We therefore investigated regulatory biophysical murine TRPC5 (mTRPC4/5)...

10.1074/jbc.275.23.17517 article EN cc-by Journal of Biological Chemistry 2000-06-01

Most existing machine translation systems operate at the level of words, relying on explicit segmentation to extract tokens. We introduce a neural (NMT) model that maps source character sequence target without any segmentation. employ character-level convolutional network with max-pooling encoder reduce length representation, allowing be trained speed comparable subword-level models while capturing local regularities. Our character-to-character outperforms recently proposed baseline WMT’15...

10.1162/tacl_a_00067 article EN cc-by Transactions of the Association for Computational Linguistics 2017-12-01
Coming Soon ...