Spiking neural networks (SNNs) underlie low-power, fault-tolerant information processing in the brain and could constitute a power-efficient alternative to conventional deep neural networks when implemented on suitable neuromorphic hardware accelerators. However, instantiating SNNs that solve complex computational tasks in-silico remains a significant challenge. Surrogate gradient (SG) techniques have emerged as a standard solution for training SNNs end-to-end. Still, their success depends on synaptic weight initialization, similar to conventional artificial neural networks (ANNs). Yet, unlike in the case of ANNs, it remains elusive what constitutes a good initial state for an SNN. Here, we develop a general initialization strategy for SNNs inspired by the fluctuation-driven regime commonly observed in the brain. Specifically, we derive practical solutions for data-dependent weight initialization that ensure fluctuation-driven firing in the widely used leaky integrate-and-fire (LIF) neurons. We empirically show that SNNs initialized following our strategy exhibit superior learning performance when trained with SGs. These findings generalize across several datasets and SNN architectures, including fully connected, deep convolutional, recurrent, and more biologically plausible SNNs obeying Dale's law. Thus fluctuation-driven initialization provides a practical, versatile, and easy-to-implement strategy for improving SNN training performance on diverse tasks in neuromorphic engineering and computational neuroscience.
Deep neural networks have significantly improved performance on a range of tasks with the increasing demand for computational resources, leaving deployment on low-resource devices (with limited memory and battery power) infeasible. Binary neural networks (BNNs) tackle the issue to an extent with extreme compression and speed-up gains compared to real-valued models. We propose a simple but effective method to accelerate inference through unifying BNNs with an early-exiting strategy. Our approach allows simple instances to exit early based on a decision threshold and utilizes output layers added to different intermediate layers to avoid executing the entire binary model. We extensively evaluate our method on three audio classification tasks and across four BNNs architectures. Our method demonstrates favorable quality-efficiency trade-offs while being controllable with an entropy-based threshold specified by the system user. It also results in better speed-ups (latency less than 6ms) with a single model based on existing BNN architectures without retraining for different efficiency levels. It also provides a straightforward way to estimate sample difficulty and better understanding of uncertainty around certain classes within the dataset.
The step function is one of the simplest and most natural activation functions for deep neural networks (DNNs). As it counts 1 for positive variables and 0 for others, its intrinsic characteristics (e.g., discontinuity and no viable information of subgradients) impede its development for several decades. Even if there is an impressive body of work on designing DNNs with continuous activation functions that can be deemed as surrogates of the step function, it is still in the possession of some advantageous properties, such as complete robustness to outliers and being capable of attaining the best learning-theoretic guarantee of predictive accuracy. Hence, in this paper, we aim to train DNNs with the step function used as an activation function (dubbed as 0/1 DNNs). We first reformulate 0/1 DNNs as an unconstrained optimization problem and then solve it by a block coordinate descend (BCD) method. Moreover, we acquire closed-form solutions for sub-problems of BCD as well as its convergence properties. Furthermore, we also integrate $\ell_{2,0}$-regularization into 0/1 DNN to accelerate the training process and compress the network scale. As a result, the proposed algorithm has a high performance on classifying MNIST and Fashion-MNIST datasets. As a result, the proposed algorithm has a desirable performance on classifying MNIST, FashionMNIST, Cifar10, and Cifar100 datasets.
Deep Learning (DL) algorithms have shown impressive performance in diverse domains. Among them, audio has attracted many researchers over the last couple of decades due to some interesting patterns--particularly in classification of audio data. For better performance of audio classification, feature selection and combination play a key role as they have the potential to make or break the performance of any DL model. To investigate this role, we conduct an extensive evaluation of the performance of several cutting-edge DL models (i.e., Convolutional Neural Network, EfficientNet, MobileNet, Supper Vector Machine and Multi-Perceptron) with various state-of-the-art audio features (i.e., Mel Spectrogram, Mel Frequency Cepstral Coefficients, and Zero Crossing Rate) either independently or as a combination (i.e., through ensembling) on three different datasets (i.e., Free Spoken Digits Dataset, Audio Urdu Digits Dataset, and Audio Gujarati Digits Dataset). Overall, results suggest feature selection depends on both the dataset and the model. However, feature combinations should be restricted to the only features that already achieve good performances when used individually (i.e., mostly Mel Spectrogram, Mel Frequency Cepstral Coefficients). Such feature combination/ensembling enabled us to outperform the previous state-of-the-art results irrespective of our choice of DL model.
Feature representations derived from models pre-trained on large-scale datasets have shown their generalizability on a variety of audio analysis tasks. Despite this generalizability, however, task-specific features can outperform if sufficient training data is available, as specific task-relevant properties can be learned. Furthermore, the complex pre-trained models bring considerable computational burdens during inference. We propose to leverage both detailed task-specific features from spectrogram input and generic pre-trained features by introducing two regularization methods that integrate the information of both feature classes. The workload is kept low during inference as the pre-trained features are only necessary for training. In experiments with the pre-trained features VGGish, OpenL3, and a combination of both, we show that the proposed methods not only outperform baseline methods, but also can improve state-of-the-art models on several audio classification tasks. The results also suggest that using the mixture of features performs better than using individual features.
Supervised learning methods can solve the given problem in the presence of a large set of labeled data. However, the acquisition of a dataset covering all the target classes typically requires manual labeling which is expensive and time-consuming. Zero-shot learning models are capable of classifying the unseen concepts by utilizing their semantic information. The present study introduces image embeddings as side information on zero-shot audio classification by using a nonlinear acoustic-semantic projection. We extract the semantic image representations from the Open Images dataset and evaluate the performance of the models on an audio subset of AudioSet using semantic information in different domains; image, audio, and textual. We demonstrate that the image embeddings can be used as semantic information to perform zero-shot audio classification. The experimental results show that the image and textual embeddings display similar performance both individually and together. We additionally calculate the semantic acoustic embeddings from the test samples to provide an upper limit to the performance. The results show that the classification performance is highly sensitive to the semantic relation between test and training classes and textual and image embeddings can reach up to the semantic acoustic embeddings when the seen and unseen classes are semantically similar.
Spiking neural networks (SNNs), particularly the single-spike variant in which neurons spike at most once, are considerably more energy efficient than standard artificial neural networks (ANNs). However, single-spike SSNs are difficult to train due to their dynamic and non-differentiable nature, where current solutions are either slow or suffer from training instabilities. These networks have also been critiqued for their limited computational applicability such as being unsuitable for time-series datasets. We propose a new model for training single-spike SNNs which mitigates the aforementioned training issues and obtains competitive results across various image and neuromorphic datasets, with up to a $13.98\times$ training speedup and up to an $81\%$ reduction in spikes compared to the multi-spike SNN. Notably, our model performs on par with multi-spike SNNs in challenging tasks involving neuromorphic time-series datasets, demonstrating a broader computational role for single-spike SNNs than previously believed.
Deep neural networks have become popular in many supervised learning tasks, but they may suffer from overfitting when the training dataset is limited. To mitigate this, many researchers use data augmentation, which is a widely used and effective method for increasing the variety of datasets. However, the randomness introduced by data augmentation causes inevitable inconsistency between training and inference, which leads to poor improvement. In this paper, we propose a consistency regularization framework based on data augmentation, called CR-Aug, which forces the output distributions of different sub models generated by data augmentation to be consistent with each other. Specifically, CR-Aug evaluates the discrepancy between the output distributions of two augmented versions of each sample, and it utilizes a stop-gradient operation to minimize the consistency loss. We implement CR-Aug to image and audio classification tasks and conduct extensive experiments to verify its effectiveness in improving the generalization ability of classifiers. Our CR-Aug framework is ready-to-use, it can be easily adapted to many state-of-the-art network architectures. Our empirical results show that CR-Aug outperforms baseline methods by a significant margin.
Most curriculum learning methods require an approach to sort the data samples by difficulty, which is often cumbersome to perform. In this work, we propose a novel curriculum learning approach termed Learning Rate Curriculum (LeRaC), which leverages the use of a different learning rate for each layer of a neural network to create a data-agnostic curriculum during the initial training epochs. More specifically, LeRaC assigns higher learning rates to neural layers closer to the input, gradually decreasing the learning rates as the layers are placed farther away from the input. The learning rates increase at various paces during the first training iterations, until they all reach the same value. From this point on, the neural model is trained as usual. This creates a model-level curriculum learning strategy that does not require sorting the examples by difficulty and is compatible with any neural network, generating higher performance levels regardless of the architecture. We conduct comprehensive experiments on 12 data sets from the computer vision (CIFAR-10, CIFAR-100, Tiny ImageNet, ImageNet-200, Food-101, UTKFace, PASCAL VOC), language (BoolQ, QNLI, RTE) and audio (ESC-50, CREMA-D) domains, considering various convolutional (ResNet-18, Wide-ResNet-50, DenseNet-121, YOLOv5), recurrent (LSTM) and transformer (CvT, BERT, SepTr) architectures. We compare our approach with the conventional training regime, as well as with Curriculum by Smoothing (CBS), a state-of-the-art data-agnostic curriculum learning approach. Unlike CBS, our performance improvements over the standard training regime are consistent across all data sets and models. Furthermore, we significantly surpass CBS in terms of training time (there is no additional cost over the standard training regime for LeRaC). Our code is freely available at: https://github.com/CroitoruAlin/LeRaC.
We present a deep learning based automatic cough classifier which can discriminate tuberculosis (TB) coughs from COVID-19 coughs and healthy coughs. Both TB and COVID-19 are respiratory diseases, contagious, have cough as a predominant symptom and claim thousands of lives each year. The cough audio recordings were collected at both indoor and outdoor settings and also uploaded using smartphones from subjects around the globe, thus containing various levels of noise. This cough data include 1.68 hours of TB coughs, 18.54 minutes of COVID-19 coughs and 1.69 hours of healthy coughs from 47 TB patients, 229 COVID-19 patients and 1498 healthy patients and were used to train and evaluate a CNN, LSTM and Resnet50. These three deep architectures were also pre-trained on 2.14 hours of sneeze, 2.91 hours of speech and 2.79 hours of noise for improved performance. The class-imbalance in our dataset was addressed by using SMOTE data balancing technique and using performance metrics such as F1-score and AUC. Our study shows that the highest F1-scores of 0.9259 and 0.8631 have been achieved from a pre-trained Resnet50 for two-class (TB vs COVID-19) and three-class (TB vs COVID-19 vs healthy) cough classification tasks, respectively. The application of deep transfer learning has improved the classifiers' performance and makes them more robust as they generalise better over the cross-validation folds. Their performances exceed the TB triage test requirements set by the world health organisation (WHO). The features producing the best performance contain higher order of MFCCs suggesting that the differences between TB and COVID-19 coughs are not perceivable by the human ear. This type of cough audio classification is non-contact, cost-effective and can easily be deployed on a smartphone, thus it can be an excellent tool for both TB and COVID-19 screening.
Recognizing human non-speech vocalizations is an important task and has broad applications such as automatic sound transcription and health condition monitoring. However, existing datasets have a relatively small number of vocal sound samples or noisy labels. As a consequence, state-of-the-art audio event classification models may not perform well in detecting human vocal sounds. To support research on building robust and accurate vocal sound recognition, we have created a VocalSound dataset consisting of over 21,000 crowdsourced recordings of laughter, sighs, coughs, throat clearing, sneezes, and sniffs from 3,365 unique subjects. Experiments show that the vocal sound recognition performance of a model can be significantly improved by 41.9% by adding VocalSound dataset to an existing dataset as training material. In addition, different from previous datasets, the VocalSound dataset contains meta information such as speaker age, gender, native language, country, and health condition.
Synthesizer is a type of electronic musical instrument that is now widely used in modern music production and sound design. Each parameters configuration of a synthesizer produces a unique timbre and can be viewed as a unique instrument. The problem of estimating a set of parameters configuration that best restore a sound timbre is an important yet complicated problem, i.e.: the synthesizer parameters estimation problem. We proposed a multi-modal deep-learning-based pipeline Sound2Synth, together with a network structure Prime-Dilated Convolution (PDC) specially designed to solve this problem. Our method achieved not only SOTA but also the first real-world applicable results on Dexed synthesizer, a popular FM synthesizer.
The abundance and ease of utilizing sound, along with the fact that auditory clues reveal a plethora of information about what happens in a scene, make the audio-visual space an intuitive choice for representation learning. In this paper, we explore the efficacy of audio-visual self-supervised learning from uncurated long-form content i.e movies. Studying its differences with conventional short-form content, we identify a non-i.i.d distribution of data, driven by the nature of movies. Specifically, we find long-form content to naturally contain a diverse set of semantic concepts (semantic diversity), where a large portion of them, such as main characters and environments often reappear frequently throughout the movie (reoccurring semantic concepts). In addition, movies often contain content-exclusive artistic artifacts, such as color palettes or thematic music, which are strong signals for uniquely distinguishing a movie (non-semantic consistency). Capitalizing on these observations, we comprehensively study the effect of emphasizing within-movie negative sampling in a contrastive learning setup. Our view is different from those of prior works who consider within-video positive sampling, inspired by the notion of semantic persistency over time, and operate in a short-video regime. Our empirical findings suggest that, with certain modifications, training on uncurated long-form videos yields representations which transfer competitively with the state-of-the-art to a variety of action recognition and audio classification tasks.
Self-supervised learning (SSL) learns knowledge from a large amount of unlabeled data, and then transfers the knowledge to a specific problem with a limited number of labeled data. SSL has achieved promising results in various domains. This work addresses the problem of segment-level general audio SSL, and proposes a new transformer-based teacher-student SSL model, named ATST. A transformer encoder is developed on a recently emerged teacher-student baseline scheme, which largely improves the modeling capability of pre-training. In addition, a new strategy for positive pair creation is designed to fully leverage the capability of transformer. Extensive experiments have been conducted, and the proposed model achieves the new state-of-the-art results on almost all of the downstream tasks.
Although few-shot learning has attracted much attention from the fields of image and audio classification, few efforts have been made on few-shot speaker identification. In the task of few-shot learning, overfitting is a tough problem mainly due to the mismatch between training and testing conditions. In this paper, we propose a few-shot speaker identification method which can alleviate the overfitting problem. In the proposed method, the model of a depthwise separable convolutional network with channel attention is trained with a prototypical loss function. Experimental datasets are extracted from three public speech corpora: Aishell-2, VoxCeleb1 and TORGO. Experimental results show that the proposed method exceeds state-of-the-art methods for few-shot speaker identification in terms of accuracy and F-score.
While efficient architectures and a plethora of augmentations for end-to-end image classification tasks have been suggested and heavily investigated, state-of-the-art techniques for audio classifications still rely on numerous representations of the audio signal together with large architectures, fine-tuned from large datasets. By utilizing the inherited lightweight nature of audio and novel audio augmentations, we were able to present an efficient end-to-end network with strong generalization ability. Experiments on a variety of sound classification sets demonstrate the effectiveness and robustness of our approach, by achieving state-of-the-art results in various settings. Public code is available at: \href{https://github.com/Alibaba-MIIL/AudioClassfication}{this http url}
Spiking Neural Network (SNN) is considered more biologically realistic and power-efficient as it imitates the fundamental mechanism of the human brain. Recently, backpropagation (BP) based SNN learning algorithms that utilize deep learning frameworks have achieved good performance. However, bio-interpretability is partially neglected in those BP-based algorithms. Toward bio-plausible BP-based SNNs, we consider three properties in modeling spike activities: Multiplicity, Adaptability, and Plasticity (MAP). In terms of multiplicity, we propose a Multiple-Spike Pattern (MSP) with multiple spike transmission to strengthen model robustness in discrete time-iteration. To realize adaptability, we adopt Spike Frequency Adaption (SFA) under MSP to decrease spike activities for improved efficiency. For plasticity, we propose a trainable convolutional synapse that models spike response current to enhance the diversity of spiking neurons for temporal feature extraction. The proposed SNN model achieves competitive performances on neuromorphic datasets: N-MNIST and SHD. Furthermore, experimental results demonstrate that the proposed three aspects are significant to iterative robustness, spike efficiency, and temporal feature extraction capability of spike activities. In summary, this work proposes a feasible scheme for bio-inspired spike activities with MAP, offering a new neuromorphic perspective to embed biological characteristics into spiking neural networks.
Currently available benchmarks for few-shot learning (machine learning with few training examples) are limited in the domains they cover, primarily focusing on image classification. This work aims to alleviate this reliance on image-based benchmarks by offering the first comprehensive, public and fully reproducible audio based alternative, covering a variety of sound domains and experimental settings. We compare the few-shot classification performance of a variety of techniques on seven audio datasets (spanning environmental sounds to human-speech). Extending this, we carry out in-depth analyses of joint training (where all datasets are used during training) and cross-dataset adaptation protocols, establishing the possibility of a generalised audio few-shot classification algorithm. Our experimentation shows gradient-based meta-learning methods such as MAML and Meta-Curvature consistently outperform both metric and baseline methods. We also demonstrate that the joint training routine helps overall generalisation for the environmental sound databases included, as well as being a somewhat-effective method of tackling the cross-dataset/domain setting.
In this paper, we propose a simple yet powerful improvement over the recent Self-Supervised Audio Spectrogram Transformer (SSAST) model for speech and audio classification. Specifically, we leverage the insight that the SSAST uses a very high masking ratio (75%) during pretraining, meaning that the vast majority of self-attention compute is performed on mask tokens. We address this by integrating the encoder-decoder architecture from Masked Autoencoders are Scalable Vision Learners (MAE) into the SSAST, where a deep encoder operates on only unmasked input, and a shallow decoder operates on encoder outputs and mask tokens. We find that MAE-like pretraining can provide a 3x speedup and 2x memory usage reduction over the vanilla SSAST using current audio pretraining strategies with ordinary model and input sizes. When fine-tuning on downstream tasks, which only uses the encoder, we find that our approach outperforms the SSAST on a variety of downstream tasks. We further conduct comprehensive evaluations into different strategies of pretraining and explore differences in MAE-style pretraining between the visual and audio domains.
Deep audio classification, traditionally cast as training a deep neural network on top of mel-filterbanks in a supervised fashion, has recently benefited from two independent lines of work. The first one explores "learnable frontends", i.e., neural modules that produce a learnable time-frequency representation, to overcome limitations of fixed features. The second one uses self-supervised learning to leverage unprecedented scales of pre-training data. In this work, we study the feasibility of combining both approaches, i.e., pre-training learnable frontend jointly with the main architecture for downstream classification. First, we show that pretraining two previously proposed frontends (SincNet and LEAF) on Audioset drastically improves linear-probe performance over fixed mel-filterbanks, suggesting that learnable time-frequency representations can benefit self-supervised pre-training even more than supervised training. Surprisingly, randomly initialized learnable filterbanks outperform mel-scaled initialization in the self-supervised setting, a counter-intuitive result that questions the appropriateness of strong priors when designing learnable filters. Through exploratory analysis of the learned frontend components, we uncover crucial differences in properties of these frontends when used in a supervised and self-supervised setting, especially the affinity of self-supervised filters to diverge significantly from the mel-scale to model a broader range of frequencies.