Passive acoustic monitoring (PAM) has shown great promise in helping ecologists understand the health of animal populations and ecosystems. However, extracting insights from millions of hours of audio recordings requires the development of specialized recognizers. This is typically a challenging task, necessitating large amounts of training data and machine learning expertise. In this work, we introduce a general, scalable and data-efficient system for developing recognizers for novel bioacoustic problems in under an hour. Our system consists of several key components that tackle problems in previous bioacoustic workflows: 1) highly generalizable acoustic embeddings pre-trained for birdsong classification minimize data hunger; 2) indexed audio search allows the efficient creation of classifier training datasets, and 3) precomputation of embeddings enables an efficient active learning loop, improving classifier quality iteratively with minimal wait time. Ecologists employed our system in three novel case studies: analyzing coral reef health through unidentified sounds; identifying juvenile Hawaiian bird calls to quantify breeding success and improve endangered species monitoring; and Christmas Island bird occupancy modeling. We augment the case studies with simulated experiments which explore the range of design decisions in a structured way and help establish best practices. Altogether these experiments showcase our system's scalability, efficiency, and generalizability, enabling scientists to quickly address new bioacoustic challenges.

Perch is a performant pre-trained model for bioacoustics. It was trained in supervised fashion, providing both off-the-shelf classification scores for thousands of vocalizing species as well as strong embeddings for transfer learning. In this new release, Perch 2.0, we expand from training exclusively on avian species to a large multi-taxa dataset. The model is trained with self-distillation using a prototype-learning classifier as well as a new source-prediction training criterion. Perch 2.0 obtains state-of-the-art performance on the BirdSet and BEANS benchmarks. It also outperforms specialized marine models on marine transfer learning tasks, despite having almost no marine training data. We present hypotheses as to why fine-grained species classification is a particularly robust pre-training task for bioacoustics.

Perch 2.0 is a supervised bioacoustics foundation model pretrained on 14,597 species, including birds, mammals, amphibians, and insects, and has state-of-the-art performance on multiple benchmarks. Given that Perch 2.0 includes almost no marine mammal audio or classes in the training data, we evaluate Perch 2.0 performance on marine mammal and underwater audio tasks through few-shot transfer learning. We perform linear probing with the embeddings generated from this foundation model and compare performance to other pretrained bioacoustics models. In particular, we compare Perch 2.0 with previous multispecies whale, Perch 1.0, SurfPerch, AVES-bio, BirdAVES, and Birdnet V2.3 models, which have open-source tools for transfer-learning and agile modeling. We show that the embeddings from the Perch 2.0 model have consistently high performance for few-shot transfer learning, generally outperforming alternative embedding models on the majority of tasks, and thus is recommended when developing new linear classifiers for marine mammal classification with few labeled examples.

Machine learning has the potential to revolutionize passive acoustic monitoring (PAM) for ecological assessments. However, high annotation and computing costs limit the field’s adoption. Generalizable pretrained networks can overcome these costs, but high-quality pretraining requires vast annotated libraries, limiting their current development to data-rich bird taxa. Here, we identify the optimum pretraining strategy for data-deficient domains, using tropical reefs as a representative case study. We assembled ReefSet, an annotated library of 57 000 reef sounds taken across 16 datasets, though still modest in scale compared to annotated bird libraries. We performed multiple pretraining experiments and found that pretraining on a library of bird audio 50 times the size of ReefSet provides notably superior generalizability on held-out reef datasets, with a mean area under the receiver operating characteristic curve (AUC-ROC) of 0.881 (±0.11), compared to pretraining on ReefSet itself or unrelated audio, with a mean AUC-ROC of 0.724 (±0.05) and 0.834 (±0.05), respectively. However, our key findings show that cross-domain mixing, where bird, reef and unrelated audio are combined during pretraining, provides superior transfer learning performance, with an AUC-ROC of 0.933 (±0.02). SurfPerch, our optimum pretrained network, provides a strong foundation for automated analysis of tropical reef and related PAM data with minimal annotation and computing costs. This article is part of the theme issue ‘Acoustic monitoring for tropical ecology and conservation’.

The ability for a machine learning model to cope with differences in training and deployment conditions--e.g. in the presence of distribution shift or the generalization to new classes altogether--is crucial for real-world use cases. However, most empirical work in this area has focused on the image domain with artificial benchmarks constructed to measure individual aspects of generalization. We present BIRB, a complex benchmark centered on the retrieval of bird vocalizations from passively-recorded datasets given focal recordings from a large citizen science corpus available for training. We propose a baseline system for this collection of tasks using representation learning and a nearest-centroid search. Our thorough empirical evaluation and analysis surfaces open research directions, suggesting that BIRB fills the need for a more realistic and complex benchmark to drive progress on robustness to distribution shifts and generalization of ML models.

In the context of passive acoustic monitoring (PAM) better models are needed to reliably gain insights from large amounts of raw, unlabeled data. Bioacoustics foundation models, which are general-purpose, adaptable models that can be used for a wide range of downstream tasks, are an effective way to meet this need. Measuring the capabilities of such models is essential for their development, but the design of robust evaluation procedures is a complex process. In this review we discuss a variety of fields that are relevant for the evaluation of bioacoustics models, such as sound event detection, machine learning metrics, and transfer learning (including topics such as few-shot learning and domain generalization). We contextualize these topics using the particularities of bioacoustics data, which is characterized by large amounts of noise, strong class imbalance, and distribution shifts (differences in the data between training and deployment stages). Our hope is that these insights will help to inform the design of evaluation protocols that can more accurately predict the ability of bioacoustics models to be deployed reliably in a wide variety of settings.

Source-free domain adaptation (SFDA) is compelling because it allows adapting an off-the-shelf model to a new domain using only unlabelled data. In this work, we apply existing SFDA techniques to a challenging set of naturally-occurring distribution shifts in bioacoustics, which are very different from the ones commonly studied in computer vision. We find existing methods perform differently relative to each other than observed in vision benchmarks, and sometimes perform worse than no adaptation at all. We propose a new simple method which outperforms the existing methods on our new shifts while exhibiting strong performance on a range of vision datasets. Our findings suggest that existing SFDA methods are not as generalizable as previously thought and that considering diverse modalities can be a useful avenue for designing more robust models.

Source-free domain adaptation (SFDA) is a compelling problem as it allows to leverage any off-the-shelf model without requiring access to its original training set and adapts it using only unlabelled data. While several SFDA approaches have recently been proposed, their evaluation focuses on a narrow set of distribution shifts for vision tasks, and their generalizability outside of that scope has not yet been investigated. We put those recent approaches to the test by evaluating them on a new set of challenging---due to extreme covariate and label shift---and naturally-occurring distribution shifts in the audio domain. We study the task of adapting a bird species classifier trained on focalized recordings of bird songs to datasets of passive recordings for various geographical locations. Interestingly, we find that some recent SFDA methods underperform doing no adaptation at all. Drawing inspiration from those findings and insights, we propose a new method that improves on noisy student approaches by adjusting the teacher's pseudo-labels through Laplacian regularization. Our approach enjoys increased stability and significantly better performance on several of our proposed distribution shifts. We then look back at SFDA benchmarks in the vision domain and find that our approach is competitive with the state-of-the-art there as well.

Average-case analysis computes the complexity of an algorithm averaged over all possible inputs. Compared to worst-case analysis, it is more representative of the typical behavior of an algorithm, but remains largely unexplored in optimization. One difficulty is that the analysis can depend on the probability distribution of the inputs to the model. However, we show that this is not the case for a class of large-scale problems trained with first-order methods including random least squares and one-hidden layer neural networks with random weights. In fact, the halting time exhibits a universality property : it is independent of the probability distribution. With this barrier for average-case analysis removed, we provide the first explicit average-case convergence rates showing a tighter complexity not captured by traditional worst-case analysis. Finally, numerical simulations suggest this universality property holds for a more general class of algorithms and problems.

The architecture and the parameters of neural networks are often optimized independently, which requires costly retraining of the parameters whenever the architecture is modified. In this work we instead focus on growing the architecture without requiring costly retraining. We present a method that adds new neurons during training without impacting what is already learned, while improving the training dynamics. We achieve the latter by maximizing the gradients of the new weights and find the optimal initialization efficiently by means of the singular value decomposition (SVD). We call this technique Gradient Maximizing Growth (GradMax) and demonstrate its effectiveness in variety of vision tasks and architectures.

Relation extraction (RE) is an important in-001 formation extraction task which provides es-002 sential information to many NLP applications 003 such as knowledge base population and ques-004 tion answering. In this paper, we present 005 a novel generative model for relation extrac-006 tion (which we call GenRE), where RE is 007 modeled as a sequence-to-sequence generation 008 task. We explore various encoding schemes 009 for the source and target sequences, and de-010 sign effective schemes that enable GenRE to 011 achieve state-of-the-art performance on three 012 benchmark RE datasets. In addition, we in-013 troduce negative sampling and decoding scal-014 ing techniques which provide a flexible tool 015 to tune the precision and recall performance 016 of our GenRE model. Our approach can be 017 extended to extract all relation triples from a 018 sentence in one pass. Although the one-pass 019 approach incurs certain performance loss, it is 020 much more computationally efficient. 021

The speed at which one can minimize an expected loss using stochastic methods depends on two properties: the curvature of the loss and the variance of the gradients. While most previous works focus on one or the other of these properties, we explore how their interaction affects optimization speed. Further, as the ultimate goal is good generalization performance, we clarify how both curvature and noise are relevant to properly estimate the generalization gap. Realizing that the limitations of some existing works stems from a confusion between these matrices, we also clarify the distinction between the Fisher matrix, the Hessian, and the covariance matrix of the gradients.

Graph neural networks have become increasingly popular in recent years due to their ability to naturally encode relational input data and their ability to scale to large graphs by operating on a sparse representation of graph adjacency matrices. As we look to scale up these models using custom hardware, a natural assumption would be that we need hardware tailored to sparse operations and/or dynamic control flow. In this work, we question this assumption by scaling up sparse graph neural networks using a platform targeted at dense computation on fixed-size data. Drawing inspiration from optimization of numerical algorithms on sparse matrices, we develop techniques that enable training the sparse graph neural network model from Allamanis et al. [2018] in 13 minutes using a 512-core TPUv2 Pod, whereas the original training takes almost a day.

The need to efficiently calculate first-and higher-order derivatives of increasingly complex models expressed in Python has stressed or exceeded the capabilities of available tools. In this work, we explore techniques from the field of automatic differentiation (AD) that can give researchers expressive power, performance and strong usability. These include source-code transformation (SCT), flexible gradient surgery, efficient in-place array operations, and higher-order derivatives. We implement and demonstrate these ideas in the Tangent software library for Python, the first AD framework for a dynamic language that uses SCT.

We review the current state of automatic differentiation (AD) for array programming in machine learning (ML), including the different approaches such as operator overloading (OO) and source transformation (ST) used for AD, graph-based intermediate representations for programs, and source languages. Based on these insights, we introduce a new graph-based intermediate representation (IR) which specifically aims to efficiently support fully-general AD for array programming. Unlike existing dataflow programming representations in ML frameworks, our IR naturally supports function calls, higher-order functions and recursion, making ML models easier to implement. The ability to represent closures allows us to perform AD using ST without a tape, making the resulting derivative (adjoint) program amenable to ahead-of-time optimization using tools from functional language compilers, and enabling higher-order derivatives. Lastly, we introduce a proof of concept compiler toolchain called Myia which uses a subset of Python as a front end.

Automatic differentiation (AD) is an essential primitive for machine learning programming systems. Tangent is a new library that performs AD using source code transformation (SCT) in Python. It takes numeric functions written in a syntactic subset of Python and NumPy as input, and generates new Python functions which calculate a derivative. This approach to automatic differentiation is different from existing packages popular in machine learning, such as TensorFlow and Autograd. Advantages are that Tangent generates gradient code in Python which is readable by the user, easy to understand and debug, and has no runtime overhead. Tangent also introduces abstractions for easily injecting logic into the generated gradient code, further improving usability.

We propose a generalization of neural network sequence models. Instead of predicting one symbol at a time, our multi-scale model makes predictions over multiple, potentially overlapping multi-symbol tokens. A variation of the byte-pair encoding (BPE) compression algorithm is used to learn the dictionary of tokens that the model is trained with. When applied to language modelling, our model has the flexibility of character-level models while maintaining many of the performance benefits of word-level models. Our experiments show that this model performs better than a regular LSTM on language modeling tasks, especially for smaller models.

Theano is a Python library that allows to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently. Since its introduction, it has been one of the most used CPU and GPU mathematical compilers - especially in the machine learning community - and has shown steady performance improvements. Theano is being actively and continuously developed since 2008, multiple frameworks have been built on top of it and it has been used to produce many state-of-the-art machine learning models. The present article is structured as follows. Section I provides an overview of the Theano software and its community. Section II presents the principal features of Theano and how to use them, and compares them with other similar projects. Section III focuses on recently-introduced functionalities and improvements. Section IV compares the performance of Theano against Torch7 and TensorFlow on several machine learning models. Section V discusses current limitations of Theano and potential ways of improving it.