Background/Objectives: The aim of this paper was to compare voice and speech characteristics between post-COVID-19 control subjects. hypothesis that acoustic parameters may differentiate subjects infected by COVID-19 from Additionally, we expected observe the persistence symptoms in women. Methods: In total, 134 participated study, were selected for convenience divided into two groups: 70 64 subjects, with an average time 8.7 months after infection. recordings made using SPIRA software...

The hippocampus is known to play a crucial role in the formation of long-term memory. For this, fast replays previously experienced activities during sleep or after reward experiences are believed be crucial. But how such generated still completely unclear. In this paper we propose possible mechanism for this: present model that can store trajectories on behavioral timescale single run, and subsequently bidirectionally replay trajectories, thereby omitting any specifics previous behavior...

Evolutionary Strategies (ES) are known to be an effective black-box optimization technique for deep neural networks when the true gradients cannot computed, such as in Reinforcement Learning. We continue a recent line of research that uses surrogate improve gradient estimation ES. propose novel method optimally incorporate information. Our approach, unlike previous work, needs no information about quality and is always guaranteed find descent direction better than gradient. This allows...

Hebbian changes of excitatory synapses are driven by and enhance correlations between pre- postsynaptic neuronal activations, forming a positive feedback loop that can lead to instability in simulated neural networks. Because learning may occur on time scales seconds minutes, it is conjectured some form fast stabilization firing necessary avoid runaway excitation, but both the theoretical underpinning biological implementation for such homeostatic mechanism be fully investigated. Supported...

We study unbiased (1 + 1) evolutionary algorithms on linear functions with an unknown number n of bits non-zero weight. Static achieve optimal runtime O(n(ln n)2+ε), however, it remained unclear whether more dynamic parameter policies could yield better guarantees. consider two setups: one where the mutation rate follows a fixed schedule, and may be adapted depending history run. For first setup, we give schedule that achieves (1±o(1))βn ln n, β ≈ 3.552, which is asymptotic improvement over...

One of the central goals Recurrent Neural Networks (RNNs) is to learn long-term dependencies in sequential data. Nevertheless, most popular training method, Truncated Backpropagation through Time (TBPTT), categorically forbids learning beyond truncation horizon. In contrast, online algorithm Real Learning (RTRL) provides untruncated gradients, with disadvantage impractically large computational costs. Recently published approaches reduce these costs by providing noisy approximations RTRL. We...

In computational neuroscience, synaptic plasticity rules are often formulated in terms of firing rates. The predominant description vivo neuronal activity, however, is the instantaneous rate (or spiking probability). this article we resolve discrepancy by showing that fluctuations membrane potential carry enough information to permit a precise estimate balanced networks. As consequence, find based not restricted activity stable for hundreds milliseconds seconds, but can be carried over...

We investigate the asymptotic version of Erdős–Ko–Rado theorem for random k -uniform hypergraph $\mathcal{H}$ ( n, p ). For 2⩽ n ) ⩽ /2, let $N=\binom{n}k$ and $D=\binom{n-k}k$ . show that with probability tending to 1 as → ∞, largest intersecting subhypergraph has size $$(1+o(1))p\ffrac kn N$$ any $$p\gg \ffrac nk\ln^2\biggl(\ffrac nk\biggr)D^{-1}.$$ This lower bound on is asymptotically best possible = Θ( this range , we are able stability well. A different behaviour occurs when o In case,...

During the coronavirus disease 2019 (COVID-19) pandemic, various research disciplines collaborated to address impacts of severe acute respiratory syndrome coronavirus-2 infections. This paper presents an interpretability analysis a convolutional neural network-based model designed for COVID-19 detection using audio data. We explore input features that play crucial role in model’s decision-making process, including spectrograms, fundamental frequency (F0), F0 standard deviation,...

We contrast high effectiveness of state the art deep learning architectures designed for general audio classification tasks, refined respiratory insufficiency (RI) detection and blood oxygen saturation (SpO2) estimation through automated analysis. Recently, multiple have been proposed to detect RI in COVID patients analysis, achieving accuracy above 95% F1-score 0.93. is a condition associated with low SpO2 levels, commonly defined as threshold <92%. While serves crucial determinant RI,...

This work explores speech as a biomarker and investigates the detection of respiratory insufficiency (RI) by analyzing samples. Previous \cite{spira2021} constructed dataset COVID-19 patient utterances analyzed it means convolutional neural network achieving an accuracy $87.04\%$, validating hypothesis that one can detect RI through speech. Here, we study how Transformer architectures improve performance on detection. approach enables construction acoustic model. By choosing correct...

The connection density of nearby neurons in the cortex has been observed to be around 0.1, whereas longer-range connections are present with much sparser (Kalisman, Silberberg, & Markram, 2005 ). We propose a memory association model that qualitatively explains these empirical observations. we consider is multiassociative, sparse, Willshaw-like consisting binary threshold and synapses. It uses recurrent synapses for iterative retrieval stored memories. quantify usefulness by simulating small...

The goal of speech emotion recognition (SER) is to identify the emotional aspects speech. SER challenge for Brazilian Portuguese was proposed with short snippets which are classified as neutral, non-neutral female and male according paralinguistic elements (laughing, crying, etc). This dataset contains about $50$ minutes As leans on small side, we investigate whether a combination transfer learning data augmentation techniques can produce positive results. Thus, by combining technique called...

We investigate the asymptotic version of Erdős-Ko-Rado theorem for random $k$-uniform hypergraph $\mathcal{H}^k(n,p)$. For $2 \leq k(n) n/2$, let $N=\binom{n}k$ and $D=\binom{n-k}k$. show that with probability tending to 1 as $n\to\infty$, largest intersecting subhypergraph $\mathcal{H}^k(n,p)$ has size $(1+o(1))p\frac kn N$, any $p\gg \frac nk\ln^2\!\left(\frac nk\right)D^{-1}$. This lower bound on $p$ is asymptotically best possible $k=Θ(n)$. this range $k$ $p$, we are able stability well....

An acoustic model, trained on a significant amount of unlabeled data, consists self-supervised learned speech representation useful for solving downstream tasks, perhaps after fine-tuning the model in respective task. In this work, we build an Brazilian Portuguese Speech through Transformer neural network. This was pretrained more than $800$ hours Speech, using combination pretraining techniques. Using labeled dataset collected detection respiratory insufficiency speakers, fine-tune network...

Abstract The hippocampus is known to play a crucial role in the formation of long-term memory. For this, fast replays previously experienced activities during sleep or after reward experiences are believed be crucial. But how such generated still completely unclear. In this paper we propose possible mechanism for this: present model that can store trajectories on behavioral timescale single run, and subsequently bidirectionally replay trajectories, thereby omitting any specifics previous...

Abstract The sharp wave ripple complex in rodent hippocampus is associated with a network burst CA3 (NB) that triggers synchronous event the CA1 population (SE). number of pyramidal cells participating SE has been observed to follow lognormal distribution. However, origin this skewed and heavy‐tailed distribution synchrony remains unknown. Because size SEs likely originate from NBs underlying neural circuitry, we model CA3‐CA1 circuit study mechanisms their functional implications. We show...