Ilyass Moummad

Ilyass Moummad, Nicolas Farrugia

Methods based on supervised learning using annotations in an end-to-end fashion have been the state-of-the-art for classification problems. However, they may be limited in their generalization capability, especially in the low data regime. In this study, we address this issue using supervised contrastive learning combined with available metadata to solve multiple pretext tasks that learn a good representation of data. We apply our approach on respiratory sound classification. This task is suited for this setting as demographic information such as sex and age are correlated with presence of lung diseases, and learning a system that implicitly encode this information may better detect anomalies. Supervised contrastive learning is a paradigm that learns similar representations to samples sharing the same class labels and dissimilar representations to samples with different class labels. The feature extractor learned using this paradigm extract useful features from the data, and we show that it outperforms cross-entropy in classifying respiratory anomalies in two different datasets. We also show that learning representations using only metadata, without class labels, obtains similar performance as using cross entropy with those labels only. In addition, when combining class labels with metadata using multiple supervised contrastive learning, an extension of supervised contrastive learning solving an additional task of grouping patients within the same sex and age group, more informative features are learned. This work suggests the potential of using multiple metadata sources in supervised contrastive settings, in particular in settings with class imbalance and few data. Our code is released at https://github.com/ilyassmoummad/scl_icbhi2017

Ilyass Moummad, Romain Serizel, Nicolas Farrugia

Bioacoustic sound event detection allows for better understanding of animal behavior and for better monitoring biodiversity using audio. Deep learning systems can help achieve this goal, however it is difficult to acquire sufficient annotated data to train these systems from scratch. To address this limitation, the Detection and Classification of Acoustic Scenes and Events (DCASE) community has recasted the problem within the framework of few-shot learning and organize an annual challenge for learning to detect animal sounds from only five annotated examples. In this work, we regularize supervised contrastive pre-training to learn features that can transfer well on new target tasks with animal sounds unseen during training, achieving a high F-score of 61.52%(0.48) when no feature adaptation is applied, and an F-score of 68.19%(0.75) when we further adapt the learned features for each new target task. This work aims to lower the entry bar to few-shot bioacoustic sound event detection by proposing a simple and yet effective framework for this task, by also providing open-source code.

Ilyass Moummad, Marius Miron, Lukas Rauch et al.

Audio-to-image retrieval offers an interpretable alternative to audio-only classification for bioacoustic species recognition, but learning aligned audio-image representations is challenging due to the scarcity of paired audio-image data. We propose a simple and data-efficient approach that enables audio-to-image retrieval without any audio-image supervision. Our proposed method uses text as a semantic intermediary: we distill the text embedding space of a pretrained image-text model (BioCLIP-2), which encodes rich visual and taxonomic structure, into a pretrained audio-text model (BioLingual) by fine-tuning its audio encoder with a contrastive objective. This distillation transfers visually grounded semantics into the audio representation, inducing emergent alignment between audio and image embeddings without using images during training. We evaluate the resulting model on multiple bioacoustic benchmarks. The distilled audio encoder preserves audio discriminative power while substantially improving audio-text alignment on focal recordings and soundscape datasets. Most importantly, on the SSW60 benchmark, the proposed approach achieves strong audio-to-image retrieval performance exceeding baselines based on zero-shot model combinations or learned mappings between text embeddings, despite not training on paired audio-image data. These results demonstrate that indirect semantic transfer through text is sufficient to induce meaningful audio-image alignment, providing a practical solution for visually grounded species recognition in data-scarce bioacoustic settings.

Ines Nolasco, Ilyass Moummad, Dan Stowell et al.

Acoustic identification of individual animals (AIID) is closely related to audio-based species classification but requires a finer level of detail to distinguish between individual animals within the same species. In this work, we frame AIID as a hierarchical multi-label classification task and propose the use of hierarchy-aware loss functions to learn robust representations of individual identities that maintain the hierarchical relationships among species and taxa. Our results demonstrate that hierarchical embeddings not only enhance identification accuracy at the individual level but also at higher taxonomic levels, effectively preserving the hierarchical structure in the learned representations. By comparing our approach with non-hierarchical models, we highlight the advantage of enforcing this structure in the embedding space. Additionally, we extend the evaluation to the classification of novel individual classes, demonstrating the potential of our method in open-set classification scenarios.

Ilyass Moummad, Marius Miron, David Robinson et al.

Large-scale biodiversity monitoring platforms increasingly rely on multimodal wildlife observations. While recent foundation models enable rich semantic representations across vision, audio, and language, retrieving relevant observations from massive archives remains challenging due to the computational cost of high-dimensional similarity search. In this work, we introduce compact hypercube embeddings for fast text-based wildlife observation retrieval, a framework that enables efficient text-based search over large-scale wildlife image and audio databases using compact binary representations. Building on the cross-view code alignment hashing framework, we extend lightweight hashing beyond a single-modality setup to align natural language descriptions with visual or acoustic observations in a shared Hamming space. Our approach leverages pretrained wildlife foundation models, including BioCLIP and BioLingual, and adapts them efficiently for hashing using parameter-efficient fine-tuning. We evaluate our method on large-scale benchmarks, including iNaturalist2024 for text-to-image retrieval and iNatSounds2024 for text-to-audio retrieval, as well as multiple soundscape datasets to assess robustness under domain shift. Results show that retrieval using discrete hypercube embeddings achieves competitive, and in several cases superior, performance compared to continuous embeddings, while drastically reducing memory and search cost. Moreover, we observe that the hashing objective consistently improves the underlying encoder representations, leading to stronger retrieval and zero-shot generalization. These results demonstrate that binary, language-based retrieval enables scalable and efficient search over large wildlife archives for biodiversity monitoring systems.

Kawtar Zaher, Ilyass Moummad, Olivier Buisson et al.

Most self-supervised learning (SSL) methods learn continuous visual representations by aligning different views of the same input, offering limited control over how information is structured across representation dimensions. In this work, we frame visual self-supervised learning as a discrete communication process between a teacher and a student network, where semantic information is transmitted through a fixed-capacity binary channel. Rather than aligning continuous features, the student predicts multi-label binary messages produced by the teacher. Discrete agreement is enforced through an element-wise binary cross-entropy objective, while a coding-rate regularization term encourages effective utilization of the constrained channel, promoting structured representations. We further show that periodically reinitializing the projection head strengthens this effect by encouraging embeddings that remain predictive across multiple discrete encodings. Extensive experiments demonstrate consistent improvements over continuous agreement baselines on image classification, retrieval, and dense visual prediction tasks, as well as under domain shift through self-supervised adaptation. Beyond backbone representations, we analyze the learned binary codes and show that they form a compact and informative discrete language, capturing semantic factors reusable across classes.

Ilyass Moummad, Kawtar Zaher, Hervé Goëau et al.

Efficient large-scale retrieval requires representations that are both compact and discriminative. Foundation models provide powerful visual and multimodal embeddings, but nearest neighbor search in these high-dimensional spaces is computationally expensive. Hashing offers an efficient alternative by enabling fast Hamming distance search with binary codes, yet existing approaches often rely on complex pipelines, multi-term objectives, designs specialized for a single learning paradigm, and long training times. We introduce CroVCA (Cross-View Code Alignment), a simple and unified principle for learning binary codes that remain consistent across semantically aligned views. A single binary cross-entropy loss enforces alignment, while coding-rate maximization serves as an anti-collapse regularizer to promote balanced and diverse codes. To implement this, we design HashCoder, a lightweight MLP hashing network with a final batch normalization layer to enforce balanced codes. HashCoder can be used as a probing head on frozen embeddings or to adapt encoders efficiently via LoRA fine-tuning. Across benchmarks, CroVCA achieves state-of-the-art results in just 5 training epochs. At 16 bits, it particularly well-for instance, unsupervised hashing on COCO completes in under 2 minutes and supervised hashing on ImageNet100 in about 3 minutes on a single GPU. These results highlight CroVCA's efficiency, adaptability, and broad applicability.

Ilyass Moummad, Kawtar Zaher, Hervé Goëau et al.

Automated plant recognition plays a crucial role in biodiversity monitoring and conservation, yet current approaches rely heavily on supervised learning, which is limited by the availability of expert-labeled data. Self-supervised learning (SSL) offers a scalable alternative, but existing methods and training protocols are largely designed for coarse-grained visual tasks and may not transfer well to fine-grained domains such as plant species recognition. In this work, we investigate SSL for plant image representation learning. We show that commonly used augmentations in SSL pipelines - such as Gaussian blur, grayscale conversion, and solarization - are detrimental in the context of plant images, as they remove subtle discriminative cues essential for fine-grained recognition. We instead identify alternative transformations, including affine and posterization, that are better suited to this domain. We further demonstrate that training SimDINOv2 on the iNaturalist 2021 Plantae subset yields significantly stronger representations than training on ImageNet-1K, highlighting the importance of domain-specific data for SSL. Our findings are consistent across both ViT-Base and ViT-Large architectures. Moreover, our models achieve competitive performance and sometimes outperform strong supervised baselines Pl@ntCLEF and BioCLIP on downstream plant recognition tasks in few-shot settings. Overall, our results highlight the critical importance of domain-adapted augmentation strategies and dataset selection in self-supervised learning, and provide practical guidelines for building scalable models for biodiversity monitoring.

Ilyass Moummad, Reda Bensaid, Kawtar Zaher et al.

Recent advances in large-scale visual representation learning have significantly improved performance in plant species and plant disease recognition tasks. However, state-of-the-art models, often based on high-capacity vision transformers or multimodal foundation models, remain computationally expensive and difficult to deploy in resource-constrained environments such as mobile or edge devices. This limitation hinders the scalability of automated biodiversity monitoring and precision agriculture systems, where efficiency is as critical as accuracy. In this work, we investigate knowledge distillation as an effective approach to transfer the representational capacity of large pretrained models into smaller, more efficient architectures. We focus on plant species and disease recognition, and conduct an extensive empirical study on two challenging benchmarks: Pl@ntNet300K-v2 and Deep-Plant-Disease. We evaluate four representative architectures, including two ConvNeXt models and two vision transformers, under multiple training regimes: from-scratch training and pretrained initialization, each with and without distillation. In total, we train and evaluate 70 models. Our results show that knowledge distillation consistently improves performance across tasks and architectures. Distilled models are able to match the performance of significantly larger models while maintaining substantially lower computational cost. These findings demonstrate the potential of knowledge distillation techniques to enable efficient and scalable deployment of plant recognition systems in real-world environmental applications.

Ilyass Moummad, Romain Serizel, Nicolas Farrugia

Self-supervised learning (SSL) in audio holds significant potential across various domains, particularly in situations where abundant, unlabeled data is readily available at no cost. This is pertinent in bioacoustics, where biologists routinely collect extensive sound datasets from the natural environment. In this study, we demonstrate that SSL is capable of acquiring meaningful representations of bird sounds from audio recordings without the need for annotations. Our experiments showcase that these learned representations exhibit the capacity to generalize to new bird species in few-shot learning (FSL) scenarios. Additionally, we show that selecting windows with high bird activation for self-supervised learning, using a pretrained audio neural network, significantly enhances the quality of the learned representations.

Lukas Rauch, René Heinrich, Ilyass Moummad et al.

Masked Autoencoders (MAEs) learn rich semantic representations in audio classification through an efficient self-supervised reconstruction task. However, general-purpose models fail to generalize well when applied directly to fine-grained audio domains. Specifically, bird-sound classification requires distinguishing subtle inter-species differences and managing high intra-species acoustic variability, revealing the performance limitations of general-domain Audio-MAEs. This work demonstrates that bridging this domain gap domain gap requires full-pipeline adaptation, not just domain-specific pretraining data. We systematically revisit and adapt the pretraining recipe, fine-tuning methods, and frozen feature utilization to bird sounds using BirdSet, a large-scale bioacoustic dataset comparable to AudioSet. Our resulting Bird-MAE achieves new state-of-the-art results in BirdSet's multi-label classification benchmark. Additionally, we introduce the parameter-efficient prototypical probing, enhancing the utility of frozen MAE representations and closely approaching fine-tuning performance in low-resource settings. Bird-MAE's prototypical probes outperform linear probing by up to 37 percentage points in mean average precision and narrow the gap to fine-tuning across BirdSet downstream tasks. Bird-MAE also demonstrates robust few-shot capabilities with prototypical probing in our newly established few-shot benchmark on BirdSet, highlighting the potential of tailored self-supervised learning pipelines for fine-grained audio domains.

Ilyass Moummad, Kawtar Zaher, Lukas Rauch et al.

Information retrieval with compact binary embeddings, also referred to as hashing, is crucial for scalable fast search applications, yet state-of-the-art hashing methods require expensive, scenario-specific training. In this work, we introduce Hashing-Baseline, a strong training-free hashing method leveraging powerful pretrained encoders that produce rich pretrained embeddings. We revisit classical, training-free hashing techniques: principal component analysis, random orthogonal projection, and threshold binarization, to produce a strong baseline for hashing. Our approach combines these techniques with frozen embeddings from state-of-the-art vision and audio encoders to yield competitive retrieval performance without any additional learning or fine-tuning. To demonstrate the generality and effectiveness of this approach, we evaluate it on standard image retrieval benchmarks as well as a newly introduced benchmark for audio hashing.

Lukas Rauch, René Heinrich, Houtan Ghaffari et al.

Although probing frozen models has become a standard evaluation paradigm, self-supervised learning in audio defaults to fine-tuning. A key reason is that global pooling creates an information bottleneck causing linear probes to misrepresent the embedding quality: The $\texttt{cls}$-token discards crucial token information about dispersed, localized events in multi-label audio. This weakness is rooted in the mismatch between the pretraining objective (operating globally) and the downstream task (localized events). Across a comprehensive benchmark of 13 datasets and 6 spectrogram-based encoders, we first investigate the global pooling bottleneck. We then introduce binarized prototypical probes: a lightweight and simple pooling method that learns prototypes to perform class-wise information aggregation. Despite its simplicity, our method notably outperforms linear and attentive probing. Our work establishes probing as a competitive and efficient paradigm for evaluating audio SSL models, challenging the reliance on costly fine-tuning.

Ilyass Moummad, Nicolas Farrugia, Romain Serizel et al.

Multi-label imbalanced classification poses a significant challenge in machine learning, particularly evident in bioacoustics where animal sounds often co-occur, and certain sounds are much less frequent than others. This paper focuses on the specific case of classifying anuran species sounds using the dataset AnuraSet, that contains both class imbalance and multi-label examples. To address these challenges, we introduce Mixture of Mixups (Mix2), a framework that leverages mixing regularization methods Mixup, Manifold Mixup, and MultiMix. Experimental results show that these methods, individually, may lead to suboptimal results; however, when applied randomly, with one selected at each training iteration, they prove effective in addressing the mentioned challenges, particularly for rare classes with few occurrences. Further analysis reveals that Mix2 is also proficient in classifying sounds across various levels of class co-occurrences.

Ilyass Moummad, Romain Serizel, Nicolas Farrugia

Deep learning has been widely used recently for sound event detection and classification. Its success is linked to the availability of sufficiently large datasets, possibly with corresponding annotations when supervised learning is considered. In bioacoustic applications, most tasks come with few labelled training data, because annotating long recordings is time consuming and costly. Therefore supervised learning is not the best suited approach to solve bioacoustic tasks. The bioacoustic community recasted the problem of sound event detection within the framework of few-shot learning, i.e. training a system with only few labeled examples. The few-shot bioacoustic sound event detection task in the DCASE challenge focuses on detecting events in long audio recordings given only five annotated examples for each class of interest. In this paper, we show that learning a rich feature extractor from scratch can be achieved by leveraging data augmentation using a supervised contrastive learning framework. We highlight the ability of this framework to transfer well for five-shot event detection on previously unseen classes in the training data. We obtain an F-score of 63.46\% on the validation set and 42.7\% on the test set, ranking second in the DCASE challenge. We provide an ablation study for the critical choices of data augmentation techniques as well as for the learning strategy applied on the training set.