Search

Quantization is a widely adopted technique for deep neural networks to reduce the memory and computational resources required. However, when quantized, most models would need a suitable calibration process to keep their performance intact, which requires data from the target domain, such as a fraction of the dataset used in model training and model validation (i.e. calibration dataset). In this study, we investigate the use of synthetic data as a substitute for the calibration with real data for the quantization method. We propose a data generation method based on Generative Adversarial Networks that are trained prior to the model quantization step. We compare the performance of models quantized using data generated by StyleGAN2-ADA and our pre-trained DiStyleGAN, with quantization using real data and an alternative data generation method based on fractal images. Overall, the results of our experiments demonstrate the potential of leveraging synthetic data for calibration during the quantization process. In our experiments, the percentage of accuracy degradation of the selected models was less than 0.6%, with our best performance achieved on MobileNetV2 (0.05%). The code is available at: https://github.com/ThanosM97/gsoc2022-openvino

The Lie-group approach was applied to determine symmetries of the third-order non-linear equation formulated for description of shear elastic disturbances in soft solids. Invariant solutions to this equation are derived and it turned out that they could represent outgoing or incoming exponentially decaying or unbounded disturbances., Comment: 5 pages

Recent Neural Architecture Search (NAS) solutions have produced impressive results training super-networks and then deriving subnetworks, a.k.a. child models that outperform expert-crafted models from a pre-defined search space. Efficient and robust subnetworks can be selected for resource-constrained edge devices, allowing them to perform well in the wild. However, constructing super-networks for arbitrary architectures is still a challenge that often prevents the adoption of these approaches. To address this challenge, we present BootstrapNAS, a software framework for automatic generation of super-networks for NAS. BootstrapNAS takes a pre-trained model from a popular architecture, e.g., ResNet- 50, or from a valid custom design, and automatically creates a super-network out of it, then uses state-of-the-art NAS techniques to train the super-network, resulting in subnetworks that significantly outperform the given pre-trained model. We demonstrate the solution by generating super-networks from arbitrary model repositories and make available the resulting super-networks for reproducibility of the results., Comment: Accepted at AAAI2022 - Practical Deep Learning in the Wild

This paper presents a description of STC Ltd. systems submitted to the NIST 2021 Speaker Recognition Evaluation for both fixed and open training conditions. These systems consists of a number of diverse subsystems based on using deep neural networks as feature extractors. During the NIST 2021 SRE challenge we focused on the training of the state-of-the-art deep speaker embeddings extractors like ResNets and ECAPA networks by using additive angular margin based loss functions. Additionally, inspired by the recent success of the wav2vec 2.0 features in automatic speech recognition we explored the effectiveness of this approach for the speaker verification filed. According to our observation the fine-tuning of the pretrained large wav2vec 2.0 model provides our best performing systems for open track condition. Our experiments with wav2vec 2.0 based extractors for the fixed condition showed that unsupervised autoregressive pretraining with Contrastive Predictive Coding loss opens the door to training powerful transformer-based extractors from raw speech signals. For video modality we developed our best solution with RetinaFace face detector and deep ResNet face embeddings extractor trained on large face image datasets. The final results for primary systems were obtained by different configurations of subsystems fusion on the score level followed by score calibration.

We present a post-training weight pruning method for deep neural networks that achieves accuracy levels tolerable for the production setting and that is sufficiently fast to be run on commodity hardware such as desktop CPUs or edge devices. We propose a data-free extension of the approach for computer vision models based on automatically-generated synthetic fractal images. We obtain state-of-the-art results for data-free neural network pruning, with ~1.5% top@1 accuracy drop for a ResNet50 on ImageNet at 50% sparsity rate. When using real data, we are able to get a ResNet50 model on ImageNet with 65% sparsity rate in 8-bit precision in a post-training setting with a ~1% top@1 accuracy drop. We release the code as a part of the OpenVINO(TM) Post-Training Optimization tool.

In this work we present a new framework for neural networks compression with fine-tuning, which we called Neural Network Compression Framework (NNCF). It leverages recent advances of various network compression methods and implements some of them, such as sparsity, quantization, and binarization. These methods allow getting more hardware-friendly models which can be efficiently run on general-purpose hardware computation units (CPU, GPU) or special Deep Learning accelerators. We show that the developed methods can be successfully applied to a wide range of models to accelerate the inference time while keeping the original accuracy. The framework can be used within the training samples, which are supplied with it, or as a standalone package that can be seamlessly integrated into the existing training code with minimal adaptations. Currently, a PyTorch version of NNCF is available as a part of OpenVINO Training Extensions at https://github.com/openvinotoolkit/nncf., Comment: 13 pages, 1 figure

Speaker recognition systems based on deep speaker embeddings have achieved significant performance in controlled conditions according to the results obtained for early NIST SRE (Speaker Recognition Evaluation) datasets. From the practical point of view, taking into account the increased interest in virtual assistants (such as Amazon Alexa, Google Home, AppleSiri, etc.), speaker verification on short utterances in uncontrolled noisy environment conditions is one of the most challenging and highly demanded tasks. This paper presents approaches aimed to achieve two goals: a) improve the quality of far-field speaker verification systems in the presence of environmental noise, reverberation and b) reduce the system qualitydegradation for short utterances. For these purposes, we considered deep neural network architectures based on TDNN (TimeDelay Neural Network) and ResNet (Residual Neural Network) blocks. We experimented with state-of-the-art embedding extractors and their training procedures. Obtained results confirm that ResNet architectures outperform the standard x-vector approach in terms of speaker verification quality for both long-duration and short-duration utterances. We also investigate the impact of speech activity detector, different scoring models, adaptation and score normalization techniques. The experimental results are presented for publicly available data and verification protocols for the VoxCeleb1, VoxCeleb2, and VOiCES datasets., Comment: Submitted to Odyssey 2020

Imaging is an important means by which information is gathered regarding the physical world. Spatial resolution and signal-to-noise ratio are underpinning concepts. There is a paucity of rigorous definitions for these quantities, which are general enough to be useful in a broad range of imaging problems, while being also sufficiently specific to enable precise quantitative evaluation of the relevant properties of imaging systems. This is particularly true for many modern forms of imaging that include digital processing of the acquired imaging data as an integral step leading to final images presented to an end-user. Here, both the well-known historical definitions of spatial resolution and some more recent approaches suitable for many forms of modern computational imaging are discussed. An intrinsic duality of spatial resolution and signal-to-noise exists in almost all types of imaging, with the related uncertainty relationship determining a trade-off between the two quantities. Examples are presented with applications to super-resolution imaging, inline holography and ghost imaging.

In this work we present a new efficient approach to Human Action Recognition called Video Transformer Network (VTN). It leverages the latest advances in Computer Vision and Natural Language Processing and applies them to video understanding. The proposed method allows us to create lightweight CNN models that achieve high accuracy and real-time speed using just an RGB mono camera and general purpose CPU. Furthermore, we explain how to improve accuracy by distilling from multiple models with different modalities into a single model. We conduct a comparison with state-of-the-art methods and show that our approach performs on par with most of them on famous Action Recognition datasets. We benchmark the inference time of the models using the modern inference framework and argue that our approach compares favorably with other methods in terms of speed/accuracy trade-off, running at 56 FPS on CPU. The models and the training code are available., Comment: 8 pages, 3 figures

It is proved that the width of a function and the width of the distribution of its values cannot be made arbitrarily small simultaneously. In the case of ergodic stochastic processes, an ensuing uncertainty relationship is demonstrated for the product of correlation length and variance. A closely related uncertainty principle is also established for the average degree of fourth-order coherence and the spatial width of modes of bosonic quantum fields. However, it is shown that, in the case of stochastic and quantum observables, certain non-classical states with sub-Poissonian statistics, such as for example photon number squeezed states in quantum optics, can overcome the "classical" noise-resolution uncertainty limit. The described uncertainty relationship, which is fundamentally different from the Heisenberg and related uncertainty principles, can define an upper limit for the information capacity of communication and imaging systems. It is expected to be useful in a variety of problems in classical and quantum physics., Comment: 12 pages, 1 figure, to be submitted to a journal

In this work, we outline the set of problems, which any Object Detection CNN faces when its development comes to the deployment stage and propose methods to deal with such difficulties. We show that these practices allow one to get Object Detection network, which can recognize two classes: vehicles and pedestrians and achieves more than 60 frames per second inference speed on Core$^{TM}$ i5-6500 CPU. The proposed model is built on top of the popular Single Shot MultiBox Object Detection framework but with substantial improvements, which were inspired by the discovered problems. The network has just 1.96 GMAC complexity and less than 7 MB model size. It is publicly available as a part of Intel$\circledR$ OpenVINO$^{TM}$ Toolkit.

Simulations of radiation damage in single molecule imaging using a X-ray free electron laser use atomic rates calculated in the lowest order. We investigate the difference in ion yield predictions using Hartree-Fock and Hartree-Fock-Slater pproximations for light and heavy elements of biological significance. The results show that for the biologically abundant elements of the second and third rows of the periodic table both approximations agree to about 6%. For the heavier elements beyond the fourth row the discrepancy rices to 11% for the range of the pulse parameters covered in this work. Presented analysis can be used for an error estimation in a wide range of ab initio simulations of the X-ray pulse interaction with biological molecules. We also discuss other atomic structure effects and show that their account has considerably smaller effect on the ion yields of respective elements compared to the choice of the approximation.

A reciprocal relationship between the autocovariance of the light intensity in the source plane and in the far-field detector plane is presented in a form analogous to the classical van Cittert - Zernike theorem, but involving intensity correlation functions. A "classical" version of the reciprocity relationship is considered first, based on the assumption of circular Gaussian statistics of the complex amplitudes in the source plane. The result is consistent with the theory of Hanbury Brown - Twiss interferometry, but it is shown to be also applicable to estimation of the source size or the spatial resolution of the detector from the noise power spectrum of flat-field images. An alternative version of the van Cittert - Zernike theorem for intensity correlations is then derived for a quantized electromagnetic beam in a coherent state, which leads to Poisson statistics for the intrinsic intensity of the beam.

This paper presents the Speech Technology Center (STC) replay attack detection systems proposed for Automatic Speaker Verification Spoofing and Countermeasures Challenge 2017. In this study we focused on comparison of different spoofing detection approaches. These were GMM based methods, high level features extraction with simple classifier and deep learning frameworks. Experiments performed on the development and evaluation parts of the challenge dataset demonstrated stable efficiency of deep learning approaches in case of changing acoustic conditions. At the same time SVM classifier with high level features provided a substantial input in the efficiency of the resulting STC systems according to the fusion systems results., Comment: 11 pages, 3 figures, accepted for Specom 2017

An uncertainty inequality is presented that establishes a lower limit for the product of the variance of the time-averaged intensity of a mode of a quantized electromagnetic field and the degree of its spatial localization. The lower limit is determined by the vacuum fluctuations within the volume corresponding to the width of the mode. This result also leads to a generalized form of the Heisenberg uncertainty principle for boson fields in which the lower limit for the product of uncertainties in the spatial and momentum localization of a mode is equal to the product of Planck's constant and a dimensionless functional which reflects the joint signal-to-noise ratio of the position and momentum of vacuum fluctuations in the region of the phase space occupied by the mode. Experimental X-ray synchrotron measurements provide an initial verification of the proposed theory in the case of Poisson statistics.

We study a wide range of neutral atoms and ions suitable for ultra-precise atomic optical clocks with naturally suppressed black body radiation shift of clock transition frequency. Calculations show that scalar polarizabilities of clock states cancel each other for at least one order of magnitude for considered systems. Results for calculations of frequencies, quadrupole moments of the states, clock transition amplitudes and natural widths of upper clock states are presented.