Search Results for author: Franz Pernkopf

Found 44 papers, 15 papers with code

On the Latent Variable Interpretation in Sum-Product Networks

no code implementations22 Jan 2016 Robert Peharz, Robert Gens, Franz Pernkopf, Pedro Domingos

We discuss conditional independencies in augmented SPNs, formally establish the probabilistic interpretation of the sum-weights and give an interpretation of augmented SPNs as Bayesian networks.

Fixed Points of Belief Propagation -- An Analysis via Polynomial Homotopy Continuation

no code implementations20 May 2016 Christian Knoll, Franz Pernkopf, Dhagash Mehta, Tianran Chen

Moreover, we show that this fixed point gives a good approximation, and the NPHC method is able to obtain this fixed point.

Safe Semi-Supervised Learning of Sum-Product Networks

1 code implementation10 Oct 2017 Martin Trapp, Tamas Madl, Robert Peharz, Franz Pernkopf, Robert Trappl

In several domains obtaining class annotations is expensive while at the same time unlabelled data are abundant.

Discrete-Valued Neural Networks Using Variational Inference

no code implementations ICLR 2018 Wolfgang Roth, Franz Pernkopf

The increasing demand for neural networks (NNs) being employed on embedded devices has led to plenty of research investigating methods for training low precision NNs.

Quantization Variational Inference

Sum-Product Networks for Sequence Labeling

no code implementations6 Jul 2018 Martin Ratajczak, Sebastian Tschiatschek, Franz Pernkopf

We consider higher-order linear-chain conditional random fields (HO-LC-CRFs) for sequence modelling, and use sum-product networks (SPNs) for representing higher-order input- and output-dependent factors.

Optical Character Recognition Optical Character Recognition (OCR)

Learning Deep Mixtures of Gaussian Process Experts Using Sum-Product Networks

1 code implementation12 Sep 2018 Martin Trapp, Robert Peharz, Carl E. Rasmussen, Franz Pernkopf

In this paper, we introduce a natural and expressive way to tackle these problems, by incorporating GPs in sum-product networks (SPNs), a recently proposed tractable probabilistic model allowing exact and efficient inference.

Gaussian Processes regression +1

Self-Guided Belief Propagation -- A Homotopy Continuation Method

no code implementations4 Dec 2018 Christian Knoll, Adrian Weller, Franz Pernkopf

Belief propagation (BP) is a popular method for performing probabilistic inference on graphical models.

N-Ary Quantization for CNN Model Compression and Inference Acceleration

no code implementations ICLR 2019 Günther Schindler, Wolfgang Roth, Franz Pernkopf, Holger Fröning

In this work we propose a method for weight and activation quantization that is scalable in terms of quantization levels (n-ary representations) and easy to compute while maintaining the performance close to full-precision CNNs.

Clustering Model Compression +1

Optimisation of Overparametrized Sum-Product Networks

1 code implementation20 May 2019 Martin Trapp, Robert Peharz, Franz Pernkopf

It seems to be a pearl of conventional wisdom that parameter learning in deep sum-product networks is surprisingly fast compared to shallow mixture models.

Bayesian Learning of Sum-Product Networks

1 code implementation NeurIPS 2019 Martin Trapp, Robert Peharz, Hong Ge, Franz Pernkopf, Zoubin Ghahramani

While parameter learning in SPNs is well developed, structure learning leaves something to be desired: Even though there is a plethora of SPN structure learners, most of them are somewhat ad-hoc and based on intuition rather than a clear learning principle.

Parameterized Structured Pruning for Deep Neural Networks

no code implementations12 Jun 2019 Guenther Schindler, Wolfgang Roth, Franz Pernkopf, Holger Froening

As a result, PSP maintains prediction performance, creates a substantial amount of sparsity that is structured and, thus, easy and efficient to map to a variety of massively parallel processors, which are mandatory for utmost compute power and energy efficiency.

Quantization

Learning a Behavior Model of Hybrid Systems Through Combining Model-Based Testing and Machine Learning (Full Version)

no code implementations10 Jul 2019 Bernhard K. Aichernig, Roderick Bloem, Masoud Ebrahimi, Martin Horn, Franz Pernkopf, Wolfgang Roth, Astrid Rupp, Martin Tappler, Markus Tranninger

Therefore, there is considerable interest in learning such hybrid behavior by means of machine learning which requires sufficient and representative training data covering the behavior of the physical system adequately.

BIG-bench Machine Learning

Deep Structured Mixtures of Gaussian Processes

1 code implementation10 Oct 2019 Martin Trapp, Robert Peharz, Franz Pernkopf, Carl E. Rasmussen

Gaussian Processes (GPs) are powerful non-parametric Bayesian regression models that allow exact posterior inference, but exhibit high computational and memory costs.

Gaussian Processes

Graph Tracking in Dynamic Probabilistic Programs via Source Transformations

no code implementations pproximateinference AABI Symposium 2019 Philipp Gabler, Martin Trapp, Hong Ge, Franz Pernkopf

Many modern machine learning algorithms, such as automatic differentiation (AD) and versions of approximate Bayesian inference, can be understood as a particular case of message passing on some computation graph.

BIG-bench Machine Learning Probabilistic Programming

Wasserstein Routed Capsule Networks

no code implementations22 Jul 2020 Alexander Fuchs, Franz Pernkopf

Capsule networks offer interesting properties and provide an alternative to today's deep neural network architectures.

Resource-Efficient Speech Mask Estimation for Multi-Channel Speech Enhancement

no code implementations22 Jul 2020 Lukas Pfeifenberger, Matthias Zöhrer, Günther Schindler, Wolfgang Roth, Holger Fröning, Franz Pernkopf

While machine learning techniques are traditionally resource intensive, we are currently witnessing an increased interest in hardware and energy efficient approaches.

BIG-bench Machine Learning Speech Enhancement

Differentiable TAN Structure Learning for Bayesian Network Classifiers

1 code implementation21 Aug 2020 Wolfgang Roth, Franz Pernkopf

Learning the structure of Bayesian networks is a difficult combinatorial optimization problem.

Combinatorial Optimization

Nonlinear Residual Echo Suppression using a Recurrent Neural Network

1 code implementation Interspeech 2020 Lukas Pfeifenberger, Franz Pernkopf

The acoustic front-end of hands-free communication de-vices introduces a variety of distortions to the linear echo pathbetween the loudspeaker and the microphone.

Acoustic echo cancellation

Quantized Neural Networks for Radar Interference Mitigation

no code implementations25 Nov 2020 Johanna Rock, Wolfgang Roth, Paul Meissner, Franz Pernkopf

Radar sensors are crucial for environment perception of driver assistance systems as well as autonomous vehicles.

Autonomous Vehicles Denoising +1

Deep Interference Mitigation and Denoising of Real-World FMCW Radar Signals

no code implementations4 Dec 2020 Johanna Rock, Mate Toth, Paul Meissner, Franz Pernkopf

We combine real measurements with simulated interference in order to create input-output data suitable for training the model.

Denoising Transfer Learning

Resource-efficient DNNs for Keyword Spotting using Neural Architecture Search and Quantization

2 code implementations18 Dec 2020 David Peter, Wolfgang Roth, Franz Pernkopf

This paper introduces neural architecture search (NAS) for the automatic discovery of small models for keyword spotting (KWS) in limited resource environments.

Keyword Spotting Neural Architecture Search +1

Blind Speech Separation and Dereverberation using Neural Beamforming

1 code implementation24 Mar 2021 Lukas Pfeifenberger, Franz Pernkopf

In this paper, we present the Blind Speech Separation and Dereverberation (BSSD) network, which performs simultaneous speaker separation, dereverberation and speaker identification in a single neural network.

Speaker Identification Speaker Separation +1

End-to-end Keyword Spotting using Neural Architecture Search and Quantization

no code implementations14 Apr 2021 David Peter, Wolfgang Roth, Franz Pernkopf

This paper introduces neural architecture search (NAS) for the automatic discovery of end-to-end keyword spotting (KWS) models in limited resource environments.

Ranked #15 on Keyword Spotting on Google Speech Commands (Google Speech Commands V2 12 metric)

Keyword Spotting Neural Architecture Search +1

Complex-valued Convolutional Neural Networks for Enhanced Radar Signal Denoising and Interference Mitigation

no code implementations29 Apr 2021 Alexander Fuchs, Johanna Rock, Mate Toth, Paul Meissner, Franz Pernkopf

Our experiments show, that the use of CVCNNs increases data efficiency, speeds up network training and substantially improves the conservation of phase information during interference removal.

Autonomous Driving Denoising

Lung Sound Classification Using Co-tuning and Stochastic Normalization

no code implementations4 Aug 2021 Truc Nguyen, Franz Pernkopf

In this paper, we use pre-trained ResNet models as backbone architectures for classification of adventitious lung sounds and respiratory diseases.

Ranked #6 on Audio Classification on ICBHI Respiratory Sound Database (using extra training data)

Audio Classification Data Augmentation +1

Acoustic Echo Cancellation with Cross-Domain Learning

1 code implementation Interspeech 2021 Lukas Pfeifenberger, Matthias Zoehrer, Franz Pernkopf

This paper proposes the Cross-Domain Echo-Controller(CDEC), submitted to the Interspeech 2021 AEC-Challenge. The algorithm consists of three building blocks: (i) a Time-Delay Compensation (TDC) module, (ii) a frequency-domainblock-based Acoustic Echo Canceler (AEC), and (iii) a Time-Domain Neural-Network (TD-NN) used as a post-processor. Our system achieves an overall MOS score of 3. 80, while onlyusing 2. 1 million parameters at a system latency of 32ms.

Acoustic echo cancellation

Distribution Mismatch Correction for Improved Robustness in Deep Neural Networks

no code implementations5 Oct 2021 Alexander Fuchs, Christian Knoll, Franz Pernkopf

The most common normalization method, batch normalization, reduces the distribution shift during training but is agnostic to changes in the input distribution during test time.

Active Bayesian Causal Inference

1 code implementation4 Jun 2022 Christian Toth, Lars Lorch, Christian Knoll, Andreas Krause, Franz Pernkopf, Robert Peharz, Julius von Kügelgen

In this work, we propose Active Bayesian Causal Inference (ABCI), a fully-Bayesian active learning framework for integrated causal discovery and reasoning, which jointly infers a posterior over causal models and queries of interest.

Active Learning Causal Discovery +2

Variational Message Passing-Based Respiratory Motion Estimation and Detection Using Radar Signals

no code implementations14 Oct 2022 Jakob Möderl, Erik Leitinger, Franz Pernkopf, Klaus Witrisal

We present a variational message passing (VMP) approach to detect the presence of a person based on their respiratory chest motion using multistatic ultra-wideband (UWB) radar.

Motion Estimation

Variational Inference of Structured Line Spectra Exploiting Group-Sparsity

no code implementations6 Mar 2023 Jakob Möderl, Franz Pernkopf, Klaus Witrisal, Erik Leitinger

The spectral lines in each group are associated with a group parameter common to all spectral lines within the group.

Variational Inference

Fast Variational Block-Sparse Bayesian Learning

1 code implementation1 Jun 2023 Jakob Möderl, Franz Pernkopf, Klaus Witrisal, Erik Leitinger

We present a fast update rule for variational block-sparse Bayesian learning (SBL) methods.

"UWBCarGraz" Dataset for Car Occupancy Detection using Ultra-Wideband Radar

no code implementations17 Nov 2023 Jakob Möderl, Stefan Posch, Franz Pernkopf, Klaus Witrisal

Our presented ResNet architecture is able to outperform the VMP algorithm in terms of the area under the receiver operating curve (AUC) at low signal-to-noise ratios (SNRs) for all three activity levels of the target.

End-to-End Training of Neural Networks for Automotive Radar Interference Mitigation

no code implementations15 Dec 2023 Christian Oswald, Mate Toth, Paul Meissner, Franz Pernkopf

In this paper we propose a new method for training neural networks (NNs) for frequency modulated continuous wave (FMCW) radar mutual interference mitigation.

Object object-detection +1

Angle-Equivariant Convolutional Neural Networks for Interference Mitigation in Automotive Radar

no code implementations18 Dec 2023 Christian Oswald, Mate Toth, Paul Meissner, Franz Pernkopf

In automotive applications, frequency modulated continuous wave (FMCW) radar is an established technology to determine the distance, velocity and angle of objects in the vicinity of the vehicle.

Rao-Blackwellising Bayesian Causal Inference

no code implementations22 Feb 2024 Christian Toth, Christian Knoll, Franz Pernkopf, Robert Peharz

Specifically, we decompose the problem of inferring the causal structure into (i) inferring a topological order over variables and (ii) inferring the parent sets for each variable.

Causal Inference Gaussian Processes +1

Cannot find the paper you are looking for? You can Submit a new open access paper.