Comparing recurrent convolutional neural networks for large scale bird species classification

We present a deep learning approach towards the large-scale prediction and analysis of bird acoustics from 100 different bird species. We use spectrograms constructed on bird audio recordings from the Cornell Bird Challenge (CBC)2020 dataset, which includes recordings of multiple and potentially overlapping bird vocalizations with background noise. Our experiments show that a hybrid modeling approach that involves a Convolutional Neural Network (CNN) for learning the representation for a slice of the spectrogram, and a Recurrent Neural Network (RNN) for the temporal component to combine across time-points leads to the most accurate model on this dataset. We show results on a spectrum of models ranging from stand-alone CNNs to hybrid models of various types obtained by combining CNNs with other CNNs or RNNs of the following types: Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), and Legendre Memory Units (LMU). The best performing model achieves an average accuracy of 67% over the 100 different bird species, with the highest accuracy of 90% for the bird species, Red crossbill. We further analyze the learned representations visually and find them to be intuitive, where we find that related bird species are clustered close together. We present a novel way to empirically interpret the representations learned by the LMU-based hybrid model which shows how memory channel patterns change over time with the changes seen in the spectrograms.

Artificial Neural Network Assisted Error Correction for MLC NAND Flash Memory

The multilevel per cell technology and continued scaling down process technology significantly improves the storage density of NAND flash memory but also brings about a challenge in that data reliability degrades due to the serious noise. To ensure the data reliability, many noise mitigation technologies have been proposed. However, they only mitigate one of the noises of the NAND flash memory channel. In this paper, we consider all the main noises and present a novel neural network-assisted error correction (ANNAEC) scheme to increase the reliability of multi-level cell (MLC) NAND flash memory. To avoid using retention time as an input parameter of the neural network, we propose a relative log-likelihood ratio (LLR) to estimate the actual LLR. Then, we transform the bit detection into a clustering problem and propose to employ a neural network to learn the error characteristics of the NAND flash memory channel. Therefore, the trained neural network has optimized performances of bit error detection. Simulation results show that our proposed scheme can significantly improve the performance of the bit error detection and increase the endurance of NAND flash memory.

Holy Selfie as the Channel of Media Memory in the Digital World

Selfie as a special genre of digital photography performs a variety of functions, giving users the possibility to refer themselves to places, persons and events, thus personifying one's self-presentation and expressing the author's attitude to the world and their own experiences. Selfie is actively used in the representation of religious life, first of all, documenting the connection of the authors to sacred places, objects, persons and events, preserving the memory of significant moments in the life of an individual and making it available to the public. The memorial function of photography in the holy selfie format merges with its communicative function, changing the motivation of religious practice, redirecting it from the acquisition of religious experience to its sharing, empathy and participation, i.e. socializing religious experience. By analysing likes and reposts of selfie content, one can create strategies for the union of virtual religious communities around the offline experience of their members. In this article we will try to identify the differences in the ways of organizing the semantic space of holy selfie, practiced by the followers of Catholicism and Orthodoxy. Holy selfie will be studied as a new media memory channel to which users resort in order to correlate the practices of constructing personal and group memory for the reproduction of religious context by banal religion. Our work is based on the content analysis of selfie photos posted on Instagram.

Recurrent Convolutional Neural Networks for Large Scale Bird Species Classification

We present a deep learning approach towards the large-scale prediction and analysis of bird acoustics from 100 different bird species. We use spectrograms constructed on bird audio recordings from the Cornell Bird Challenge (CBC) dataset, which includes recordings with background noise, of multiple and potentially overlapping bird vocalizations per audio. Our experiments show that a hybrid modeling approach that involves a Convolutional Neural Network (CNN) for learning the representation for a slice of the spectrogram and a Recurrent Neural Network (RNN) for the temporal component to combine across time-points leads to the most accurate model on this dataset. We show results on a spectrum of models ranging from stand-alone CNNs to hybrid models of various types obtained by combining CNNs with CNNs or RNNs of the following types: Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU) and Legendre Memory Units (LMU). The best performing model achieves an average accuracy of 67% over the 100 different bird species, with the highest accuracy of 90% for the Red crossbill. We further analyze the learned representations visually and find them to be intuitive, where we find that related bird species are clustered close together. We present a novel way to empirically interpret the representations learned by the LMU-based hybrid model which shows how memory channel patterns over time change with spectrograms.

A robust W-state encoding for linear quantum optics

Error-detection and correction are necessary prerequisites for any scalable quantum computing architecture. Given the inevitability of unwanted physical noise in quantum systems and the propensity for errors to spread as computations proceed, computational outcomes can become substantially corrupted. This observation applies regardless of the choice of physical implementation. In the context of photonic quantum information processing, there has recently been much interest in passive linear optics quantum computing, which includes boson-sampling, as this model eliminates the highly-challenging requirements for feed-forward via fast, active control. That is, these systems are passive by definition. In usual scenarios, error detection and correction techniques are inherently active, making them incompatible with this model, arousing suspicion that physical error processes may be an insurmountable obstacle. Here we explore a photonic error-detection technique, based on W-state encoding of photonic qubits, which is entirely passive, based on post-selection, and compatible with these near-term photonic architectures of interest. We show that this W-state redundant encoding techniques enables the suppression of dephasing noise on photonic qubits via simple fan-out style operations, implemented by optical Fourier transform networks, which can be readily realised today. The protocol effectively maps dephasing noise into heralding failures, with zero failure probability in the ideal no-noise limit. We present our scheme in the context of a single photonic qubit passing through a noisy communication or quantum memory channel, which has not been generalised to the more general context of full quantum computation.

Implementation problems of high-speed DDR4 channels in a new generation Russian multi-core microprocessor

Solving the issue of compatibility for the new domestic developments with continuously implemented and used in global microelectronics industry cutting-edge standards requires substantial work on analysis and optimization of the implementation environment. The results of the new Elbrus 8SV microprocessor DDR4 random access memory channel study are provided in this article. The much lower than estimated channel data transfer speed has become the main issue. In order to overcome it the channel functioning study method has been developed and implemented. It is based on forming the analogs of eye diagrams, which allow estimating the area of operability and using the optimal settings. Studies held using this method allowed establishing the cause for unsatisfactory performance of the channel and objectively assessing design decisions made during development. After taking these results into account and applying changes to the chip and the circuit board of the microprocessor case, an improved version of the microprocessor was released. It became possible to achieve the calculated data transfer speed via the memory channel.

Многоуровневая запись в тонких пленках Ge-=SUB=-2-=/SUB=-Sb-=SUB=-2-=/SUB=-Te-=SUB=-5-=/SUB=-

I−V characteristics obtained on Ge2Sb2Te5 thin films in the current mode were studied. The effect of multilevel recording was established upon sequentially applying to the sample current pulses with increasing maximum value. It was shown that this effect can be associated with expansion of the memory channel. An estimate of the channel size is obtained. It is concluded that Ge2Sb2Te5 films can be used as a memristor.