Deep learning (DL) has greatly advanced audio classification, yet the field is limited by the scarcity of large-scale benchmark datasets that have propelled progress in other domains. While AudioSet is a pivotal step to bridge this gap as a universal-domain dataset, its restricted accessibility and limited range of evaluation use cases challenge its role as the sole resource. Therefore, we introduce BirdSet, a large-scale benchmark data set for audio classification focusing on avian bioacoustics. BirdSet surpasses AudioSet with over 6,800 recording hours ($\uparrow17\%$) from nearly 10,000 classes ($\uparrow18\times$) for training and more than 400 hours ($\uparrow7\times$) across eight strongly labeled evaluation datasets. It serves as a versatile resource for use cases such as multi-label classification, covariate shift or self-supervised learning. We benchmark six well-known DL models in multi-label classification across three distinct training scenarios and outline further evaluation use cases in audio classification. We host our dataset on Hugging Face for easy accessibility and offer an extensive codebase to reproduce our results.

Abstract In the field of deep learning based computer vision, the development of deep object detection has led to unique paradigms (e.g., two-stage or set-based) and architectures (e.g., Faster-RCNN or DETR ) which enable outstanding performance on challenging benchmark datasets. Despite this, the trained object detectors typically do not reliably assess uncertainty regarding their own knowledge, and the quality of their probabilistic predictions is usually poor. As these are often used to make subsequent decisions, such inaccurate probabilistic predictions must be avoided. In this work, we investigate the uncertainty calibration properties of different pretrained object detection architectures in a multi-class setting. We propose a framework to ensure a fair, unbiased, and repeatable evaluation and conduct detailed analyses assessing the calibration under distributional changes (e.g., distributional shift and application to out-of-distribution data). Furthermore, by investigating the influence of different detector paradigms, post-processing steps, and suitable choices of metrics, we deliver novel insights into why poor detector calibration emerges. Based on these insights, we are able to improve the calibration of a detector by simply finetuning its last layer.

Training with noisy class labels impairs neural networks’ generalization performance. In this context, mixup is a popular regularization technique to improve training robustness by making memorizing false class labels more difficult. However, mixup neglects that multiple annotators, e.g., crowdworkers, typically provide class labels. Therefore, we propose an extension of mixup, which handles multiple class labels per instance while considering which class label originates from which annotator. Integrated into our multi-annotator classification framework annot-mix, it performs superiorly to eleven (mostly state-of-the-art) approaches in an evaluation study with eleven datasets comprising noisy class labels from either human or simulated annotators. Our code is publicly available through our GitHub repository at https://github.com/ies-research/multi-annotator-machine-learning/tree/annot-mix.

Human annotators typically provide annotated data for training machine learning models, such as neural networks. Yet, human annotations are subject to noise, impairing generalization performances. Methodological research on approaches counteracting noisy annotations requires corresponding datasets for a meaningful empirical evaluation. Consequently, we introduce a novel benchmark dataset, dopanim, consisting of about 15,750 animal images of 15 classes with ground truth labels. For approximately 10,500 of these images, 20 humans provided over 52,000 annotations with an accuracy of circa 67%. Its key attributes include (1) the challenging task of classifying doppelganger animals, (2) human-estimated likelihoods as annotations, and (3) annotator metadata. We benchmark well-known multi-annotator learning approaches using seven variants of this dataset and outline further evaluation use cases such as learning beyond hard class labels and active learning. Our dataset and a comprehensive codebase are publicly available to emulate the data collection process and to reproduce all empirical results.

Abstract Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active learning aims at reducing the labeling costs by an efficient and effective allocation of costly labeling resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes the gain in misclassification error, and (2) uses a Bayesian approach by introducing a conjugate prior distribution to determine the class posterior to deal with uncertainties. By reformulating existing selection strategies within our proposed model, we can explain which aspects are not covered in current state-of-the-art and why this leads to the superior performance of our approach. Extensive experiments on a large variety of datasets and different kernels validate our claims.

Classifiers require annotations of instances, i.e., class labels, for training. An annotation process is often costly due to its manual execution through human annotators. Active learning (AL) aims at reducing the annotation costs by selecting instances from which the classifier is expected to learn the most. Many AL strategies assume the availability of a single omniscient annotator. In this article, we overcome this limitation by considering multiple error-prone annotators. We propose the novel AL strategy multi-annotator probabilistic active learning (MaPAL). Due to the nature of learning with error-prone annotators, it must not only select instances but annotators, too. MaPAL builds on a decision-theoretic framework and selects instance-annotator pairs maximizing the classifier's expected performance. Experiments on a variety of data sets demonstrate MaPAL's superior performance compared to five related AL strategies.

Despite the success of deep neural networks (DNN) in many applications, their ability to model uncertainty is still significantly limited. For example, in safety-critical applications such as autonomous driving, it is crucial to obtain a prediction that reflects different types of uncertainty to address life-threatening situations appropriately. In such cases, it is essential to be aware of the risk (i.e., aleatoric uncertainty) and the reliability (i.e., epistemic uncertainty) that comes with a prediction. We present AE-DNN, a model allowing the separation of aleatoric and epistemic uncertainty while maintaining a proper generalization capability. AE-DNN is based on deterministic DNN, which can determine the respective uncertainty measures in a single forward pass. In analyses with synthetic and image data, we show that our method improves the modeling of epistemic uncertainty while providing an intuitively understandable separation of risk and reliability.

Class assignment (label) is not the only information that can be queried from annotators. Additional feedback, in form of confidences, quantifies the reliability of the provided label. Multiple annotators classifying the same sample with different levels of expertise posses different levels of confidences. In this article, we discuss different types of confidence inspired by real-world application and discuss how they relate to each other. The α-confidence is inspired by humans, the β-confidence works with probabilistic outputs from intelligent machines (e.g., robots), and γ-confidence models non-normalized confidences. We consider uncertain, benevolent annotators, thus the provided labels can be contradictory. To overcome this problem, we propose two fusion strategies that combine the confidences from multiple uncertain annotators to a single one. Numerical and graphical evaluation indicates superior performance of our strategy compared to related strategies, namely, confidence weighted majority vote, c-Certainty, and a maximum likelihood estimation.

As robots are employed for automating processes in industry, there is a strong demand on robots being able to solve new tasks without costly adaptions. In this article, we present a probabilistic active learning approach for adapting robots to tasks involving object sorting by means of classification according to a human understanding of the problem and its solution. In the beginning, the robot extracts appropriate features of images of the available objects. These features are used by the robot to actively learn to solve the sorting task by asking a human teacher. A novel visualization tool allows to supervise the learning process and to determine when training is complete. Then, the robot is able to successfully sort the objects autonomously. We show our method's superiority compared to other active learning strategies and present results of its application on a real robot to prove the above concept.