Alessandro Lameiras Koerich

Nicolas Richet, Soufiane Belharbi, Haseeb Aslam et al.

Systems for multimodal emotion recognition (ER) are commonly trained to extract features from different modalities (e.g., visual, audio, and textual) that are combined to predict individual basic emotions. However, compound emotions often occur in real-world scenarios, and the uncertainty of recognizing such complex emotions over diverse modalities is challenging for feature-based models. As an alternative, emerging large language models (LLMs) like BERT and LLaMA can rely on explicit non-verbal cues that may be translated from different non-textual modalities (e.g., audio and visual) into text. Textualization of modalities augments data with emotional cues to help the LLM encode the interconnections between all modalities in a shared text space. In such text-based models, prior knowledge of ER tasks is leveraged to textualize relevant non-verbal cues such as audio tone from vocal expressions, and action unit intensity from facial expressions. Since the pre-trained weights are publicly available for many LLMs, training on large-scale datasets is unnecessary, allowing to fine-tune for downstream tasks such as compound ER (CER). This paper compares the potential of text- and feature-based approaches for compound multimodal ER in videos. Experiments were conducted on the challenging C-EXPR-DB dataset in the wild for CER, and contrasted with results on the MELD dataset for basic ER. Our results indicate that multimodal textualization provides lower accuracy than feature-based models on C-EXPR-DB, where text transcripts are captured in the wild. However, higher accuracy can be achieved when the video data has rich transcripts. Our code is available.

Muhammad Osama Zeeshan, Masoumeh Sharafi, Benoît Savary et al.

Personalization in emotion recognition (ER) is essential for an accurate interpretation of subtle and subject-specific expressive patterns. Recent advances in vision-language models (VLMs) such as CLIP demonstrate strong potential for leveraging joint image-text representations in ER. However, CLIP-based methods either depend on CLIP's contrastive pretraining or on LLMs to generate descriptive text prompts, which are noisy, computationally expensive, and fail to capture fine-grained expressions, leading to degraded performance. In this work, we leverage Action Units (AUs) as structured textual prompts within CLIP to model fine-grained facial expressions. AUs encode the subtle muscle activations underlying expressions, providing localized and interpretable semantic cues for more robust ER. We introduce CLIP-AU, a lightweight AU-guided temporal learning method that integrates interpretable AU semantics into CLIP. It learns generic, subject-agnostic representations by aligning AU prompts with facial dynamics, enabling fine-grained ER without CLIP fine-tuning or LLM-generated text supervision. Although CLIP-AU models fine-grained AU semantics, it does not adapt to subject-specific variability in subtle expressions. To address this limitation, we propose CLIP-AUTT, a video-based test-time personalization method that dynamically adapts AU prompts to videos from unseen subjects. By combining entropy-guided temporal window selection with prompt tuning, CLIP-AUTT enables subject-specific adaptation while preserving temporal consistency. Our extensive experiments on three challenging video-based subtle ER datasets, BioVid, StressID, and BAH, indicate that CLIP-AU and CLIP-AUTT outperform state-of-the-art CLIP-based FER and TTA methods, achieving robust and personalized subtle ER. Our code is publicly available at: https://github.com/osamazeeshan/CLIP-AUTT.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

This paper investigates the impact of different standard environmental sound representations (spectrograms) on the recognition performance and adversarial attack robustness of a victim residual convolutional neural network, namely ResNet-18. Our main motivation for focusing on such a front-end classifier rather than other complex architectures is balancing recognition accuracy and the total number of training parameters. Herein, we measure the impact of different settings required for generating more informative Mel-frequency cepstral coefficient (MFCC), short-time Fourier transform (STFT), and discrete wavelet transform (DWT) representations on our front-end model. This measurement involves comparing the classification performance over the adversarial robustness. We demonstrate an inverse relationship between recognition accuracy and model robustness against six benchmarking attack algorithms on the balance of average budgets allocated by the adversary and the attack cost. Moreover, our experimental results have shown that while the ResNet-18 model trained on DWT spectrograms achieves a high recognition accuracy, attacking this model is relatively more costly for the adversary than other 2D representations. We also report some results on different convolutional neural network architectures such as ResNet-34, ResNet-56, AlexNet, and GoogLeNet, SB-CNN, and LSTM-based.

Jonathan de Matos, Luiz Eduardo Soares de Oliveira, Alceu de Souza Britto Junior et al.

This paper presents a novel approach combining convolutional layers (CLs) and large-margin metric learning for training supervised models on small datasets for texture classification. The core of such an approach is a loss function that computes the distances between instances of interest and support vectors. The objective is to update the weights of CLs iteratively to learn a representation with a large margin between classes. Each iteration results in a large-margin discriminant model represented by support vectors based on such a representation. The advantage of the proposed approach w.r.t. convolutional neural networks (CNNs) is two-fold. First, it allows representation learning with a small amount of data due to the reduced number of parameters compared to an equivalent CNN. Second, it has a low training cost since the backpropagation considers only support vectors. The experimental results on texture and histopathologic image datasets have shown that the proposed approach achieves competitive accuracy with lower computational cost and faster convergence when compared to equivalent CNNs.

Manuela González-González, Soufiane Belharbi, Muhammad Osama Zeeshan et al.

Using behavioural science, health interventions focus on behaviour change by providing a framework to help patients acquire and maintain healthy habits that improve medical outcomes. In-person interventions are costly and difficult to scale, especially in resource-limited regions. Digital health interventions offer a cost-effective approach, potentially supporting independent living and self-management. Automating such interventions, especially through machine learning, has gained considerable attention recently. Ambivalence and hesitancy (A/H) play a primary role for individuals to delay, avoid, or abandon health interventions. A/H are subtle and conflicting emotions that place a person in a state between positive and negative evaluations of a behaviour, or between acceptance and refusal to engage in it. They manifest as affective inconsistency across modalities or within a modality, such as language, facial, vocal expressions, and body language. While experts can be trained to recognize A/H, integrating them into digital health interventions is costly and less effective. Automatic A/H recognition is therefore critical for the personalization and cost-effectiveness of digital health interventions. Here, we explore the application of deep learning models for A/H recognition in videos, a multi-modal task by nature. In particular, this paper covers three learning setups: supervised learning, unsupervised domain adaptation for personalization, and zero-shot inference via large language models (LLMs). Our experiments are conducted on the unique and recently published BAH video dataset for A/H recognition. Our results show limited performance, suggesting that more adapted multi-modal models are required for accurate A/H recognition. Better methods for modeling spatio-temporal and multimodal fusion are necessary to leverage conflicts within/across modalities.

Muhammad Haseeb Aslam, Marco Pedersoli, Alessandro Lameiras Koerich et al.

Human emotion is a complex phenomenon conveyed and perceived through facial expressions, vocal tones, body language, and physiological signals. Multimodal emotion recognition systems can perform well because they can learn complementary and redundant semantic information from diverse sensors. In real-world scenarios, only a subset of the modalities employed for training may be available at test time. Learning privileged information allows a model to exploit data from additional modalities that are only available during training. SOTA methods for PKD have been proposed to distill information from a teacher model (with privileged modalities) to a student model (without privileged modalities). However, such PKD methods utilize point-to-point matching and do not explicitly capture the relational information. Recently, methods have been proposed to distill the structural information. However, PKD methods based on structural similarity are primarily confined to learning from a single joint teacher representation, which limits their robustness, accuracy, and ability to learn from diverse multimodal sources. In this paper, a multi-teacher PKD (MT-PKDOT) method with self-distillation is introduced to align diverse teacher representations before distilling them to the student. MT-PKDOT employs a structural similarity KD mechanism based on a regularized optimal transport (OT) for distillation. The proposed MT-PKDOT method was validated on the Affwild2 and Biovid datasets. Results indicate that our proposed method can outperform SOTA PKD methods. It improves the visual-only baseline on Biovid data by 5.5%. On the Affwild2 dataset, the proposed method improves 3% and 5% over the visual-only baseline for valence and arousal respectively. Allowing the student to learn from multiple diverse sources is shown to increase the accuracy and implicitly avoids negative transfer to the student model.

Masoumeh Sharafi, Muhammad Osama Zeeshan, Soufiane Belharbi et al.

Facial expression recognition (FER) in videos requires model personalization to capture the considerable variations across subjects. Vision-language models (VLMs) offer strong transfer to downstream tasks through image-text alignment, but their performance can still degrade under inter-subject distribution shifts. Personalizing models using test-time adaptation (TTA) methods can mitigate this challenge. However, most state-of-the-art TTA methods rely on unsupervised parameter optimization, introducing computational overhead that is impractical in many real-world applications. This paper introduces TTA through Cache Personalization (TTA-CaP), a cache-based TTA method that enables cost-effective (gradient-free) personalization of VLMs for video FER. Prior cache-based TTA methods rely solely on dynamic memories that store test samples, which can accumulate errors and drift due to noisy pseudo-labels. TTA-CaP leverages three coordinated caches: a personalized source cache that stores source-domain prototypes, a positive target cache that accumulates reliable subject-specific samples, and a negative target cache that stores low-confidence cases as negative samples to reduce the impact of noisy pseudo-labels. Cache updates and replacement are controlled by a tri-gate mechanism based on temporal stability, confidence, and consistency with the personalized cache. Finally, TTA-CaP refines predictions through fusion of embeddings, yielding refined representations that support temporally stable video-level predictions. Our experiments on three challenging video FER datasets, BioVid, StressID, and BAH, indicate that TTA-CaP can outperform state-of-the-art TTA methods under subject-specific and environmental shifts, while maintaining low computational and memory overhead for real-world deployment.

Paul Waligora, Haseeb Aslam, Osama Zeeshan et al.

Multimodal emotion recognition (MMER) systems typically outperform unimodal systems by leveraging the inter- and intra-modal relationships between, e.g., visual, textual, physiological, and auditory modalities. This paper proposes an MMER method that relies on a joint multimodal transformer (JMT) for fusion with key-based cross-attention. This framework can exploit the complementary nature of diverse modalities to improve predictive accuracy. Separate backbones capture intra-modal spatiotemporal dependencies within each modality over video sequences. Subsequently, our JMT fusion architecture integrates the individual modality embeddings, allowing the model to effectively capture inter- and intra-modal relationships. Extensive experiments on two challenging expression recognition tasks -- (1) dimensional emotion recognition on the Affwild2 dataset (with face and voice) and (2) pain estimation on the Biovid dataset (with face and biosensors) -- indicate that our JMT fusion can provide a cost-effective solution for MMER. Empirical results show that MMER systems with our proposed fusion allow us to outperform relevant baseline and state-of-the-art methods.

Soufiane Belharbi, Marco Pedersoli, Alessandro Lameiras Koerich et al.

Although state-of-the-art classifiers for facial expression recognition (FER) can achieve a high level of accuracy, they lack interpretability, an important feature for end-users. Experts typically associate spatial action units (\aus) from a codebook to facial regions for the visual interpretation of expressions. In this paper, the same expert steps are followed. A new learning strategy is proposed to explicitly incorporate \au cues into classifier training, allowing to train deep interpretable models. During training, this \au codebook is used, along with the input image expression label, and facial landmarks, to construct a \au heatmap that indicates the most discriminative image regions of interest w.r.t the facial expression. This valuable spatial cue is leveraged to train a deep interpretable classifier for FER. This is achieved by constraining the spatial layer features of a classifier to be correlated with \au heatmaps. Using a composite loss, the classifier is trained to correctly classify an image while yielding interpretable visual layer-wise attention correlated with \au maps, simulating the expert decision process. Our strategy only relies on image class expression for supervision, without additional manual annotations. Our new strategy is generic, and can be applied to any deep CNN- or transformer-based classifier without requiring any architectural change or significant additional training time. Our extensive evaluation on two public benchmarks \rafdb, and \affectnet datasets shows that our proposed strategy can improve layer-wise interpretability without degrading classification performance. In addition, we explore a common type of interpretable classifiers that rely on class activation mapping (CAM) methods, and show that our approach can also improve CAM interpretability.

Cristiano Mesquita Garcia, Ramon Simoes Abilio, Alessandro Lameiras Koerich et al.

The society produces textual data online in several ways, e.g., via reviews and social media posts. Therefore, numerous researchers have been working on discovering patterns in textual data that can indicate peoples' opinions, interests, etc. Most tasks regarding natural language processing are addressed using traditional machine learning methods and static datasets. This setting can lead to several problems, e.g., outdated datasets and models, which degrade in performance over time. This is particularly true regarding concept drift, in which the data distribution changes over time. Furthermore, text streaming scenarios also exhibit further challenges, such as the high speed at which data arrives over time. Models for stream scenarios must adhere to the aforementioned constraints while learning from the stream, thus storing texts for limited periods and consuming low memory. This study presents a systematic literature review regarding concept drift adaptation in text stream scenarios. Considering well-defined criteria, we selected 48 papers published between 2018 and August 2024 to unravel aspects such as text drift categories, detection types, model update mechanisms, stream mining tasks addressed, and text representation methods and their update mechanisms. Furthermore, we discussed drift visualization and simulation and listed real-world datasets used in the selected papers. Finally, we brought forward a discussion on existing works in the area, also highlighting open challenges and future research directions for the community.

Muhammad Osama Zeeshan, Muhammad Haseeb Aslam, Soufiane Belharbi et al.

Adapting a deep learning model to a specific target individual is a challenging facial expression recognition (FER) task that may be achieved using unsupervised domain adaptation (UDA) methods. Although several UDA methods have been proposed to adapt deep FER models across source and target data sets, multiple subject-specific source domains are needed to accurately represent the intra- and inter-person variability in subject-based adaption. This paper considers the setting where domains correspond to individuals, not entire datasets. Unlike UDA, multi-source domain adaptation (MSDA) methods can leverage multiple source datasets to improve the accuracy and robustness of the target model. However, previous methods for MSDA adapt image classification models across datasets and do not scale well to a more significant number of source domains. This paper introduces a new MSDA method for subject-based domain adaptation in FER. It efficiently leverages information from multiple source subjects (labeled source domain data) to adapt a deep FER model to a single target individual (unlabeled target domain data). During adaptation, our subject-based MSDA first computes a between-source discrepancy loss to mitigate the domain shift among data from several source subjects. Then, a new strategy is employed to generate augmented confident pseudo-labels for the target subject, allowing a reduction in the domain shift between source and target subjects. Experiments performed on the challenging BioVid heat and pain dataset with 87 subjects and the UNBC-McMaster shoulder pain dataset with 25 subjects show that our subject-based MSDA can outperform state-of-the-art methods yet scale well to multiple subject-based source domains.

Masoumeh Sharafi, Emma Ollivier, Muhammad Osama Zeeshan et al.

Facial Expression Recognition (FER) from videos is a crucial task in various application areas, such as human-computer interaction and health diagnosis and monitoring (e.g., assessing pain and depression). Beyond the challenges of recognizing subtle emotional or health states, the effectiveness of deep FER models is often hindered by the considerable inter-subject variability in expressions. Source-free (unsupervised) domain adaptation (SFDA) methods may be employed to adapt a pre-trained source model using only unlabeled target domain data, thereby avoiding data privacy, storage, and transmission issues. Typically, SFDA methods adapt to a target domain dataset corresponding to an entire population and assume it includes data from all recognition classes. However, collecting such comprehensive target data can be difficult or even impossible for FER in healthcare applications. In many real-world scenarios, it may be feasible to collect a short neutral control video (which displays only neutral expressions) from target subjects before deployment. These videos can be used to adapt a model to better handle the variability of expressions among subjects. This paper introduces the Disentangled SFDA (DSFDA) method to address the challenge posed by adapting models with missing target expression data. DSFDA leverages data from a neutral target control video for end-to-end generation and adaptation of target data with missing non-neutral data. Our method learns to disentangle features related to expressions and identity while generating the missing non-neutral expression data for the target subject, thereby enhancing model accuracy. Additionally, our self-supervision strategy improves model adaptation by reconstructing target images that maintain the same identity and source expression.

Manuela González-González, Soufiane Belharbi, Muhammad Osama Zeeshan et al.

Recognizing complex emotions linked to ambivalence and hesitancy (A/H) can play a critical role in the personalization and effectiveness of digital behaviour change interventions. These subtle and conflicting emotions are manifested by a discord between multiple modalities, such as facial and vocal expressions, and body language. Although experts can be trained to identify A/H, integrating them into digital interventions is costly and less effective. Automatic learning systems provide a cost-effective alternative that can adapt to individual users, and operate seamlessly within real-time, and resource-limited environments. However, there are currently no datasets available for the design of ML models to recognize A/H. This paper introduces a first Behavioural Ambivalence/Hesitancy (BAH) dataset collected for subject-based multimodal recognition of A/H in videos. It contains videos from 224 participants captured across 9 provinces in Canada, with different age, and ethnicity. Through our web platform, we recruited participants to answer 7 questions, some of which were designed to elicit A/H while recording themselves via webcam with microphone. BAH amounts to 1,118 videos for a total duration of 8.26 hours with 1.5 hours of A/H. Our behavioural team annotated timestamp segments to indicate where A/H occurs, and provide frame- and video-level annotations with the A/H cues. Video transcripts and their timestamps are also included, along with cropped and aligned faces in each frame, and a variety of participants meta-data. We include results baselines for BAH at frame- and video-level recognition in multi-modal setups, in addition to zero-shot prediction, and for personalization using unsupervised domain adaptation. The limited performance of baseline models highlights the challenges of recognizing A/H in real-world videos. The data, code, and pretrained weights are available.

Cristiano Mesquita Garcia, Alessandro Lameiras Koerich, Alceu de Souza Britto et al.

Systems and individuals produce data continuously. On the Internet, people share their knowledge, sentiments, and opinions, provide reviews about services and products, and so on. Automatically learning from these textual data can provide insights to organizations and institutions, thus preventing financial impacts, for example. To learn from textual data over time, the machine learning system must account for concept drift. Concept drift is a frequent phenomenon in real-world datasets and corresponds to changes in data distribution over time. For instance, a concept drift occurs when sentiments change or a word's meaning is adjusted over time. Although concept drift is frequent in real-world applications, benchmark datasets with labeled drifts are rare in the literature. To bridge this gap, this paper provides four textual drift generation methods to ease the production of datasets with labeled drifts. These methods were applied to Yelp and Airbnb datasets and tested using incremental classifiers respecting the stream mining paradigm to evaluate their ability to recover from the drifts. Results show that all methods have their performance degraded right after the drifts, and the incremental SVM is the fastest to run and recover the previous performance levels regarding accuracy and Macro F1-Score.

Muhammad Osama Zeeshan, Marco Pedersoli, Alessandro Lameiras Koerich et al.

Personalized facial expression recognition (FER) involves adapting a machine learning model using samples from labeled sources and unlabeled target domains. Given the challenges of recognizing subtle expressions with considerable interpersonal variability, state-of-the-art unsupervised domain adaptation (UDA) methods focus on the multi-source UDA (MSDA) setting, where each domain corresponds to a specific subject, and improve model accuracy and robustness. However, when adapting to a specific target, the diverse nature of multiple source domains translates to a large shift between source and target data. State-of-the-art MSDA methods for FER address this domain shift by considering all the sources to adapt to the target representations. Nevertheless, adapting to a target subject presents significant challenges due to large distributional differences between source and target domains, often resulting in negative transfer. In addition, integrating all sources simultaneously increases computational costs and causes misalignment with the target. To address these issues, we propose a progressive MSDA approach that gradually introduces information from subjects based on their similarity to the target subject. This will ensure that only the most relevant sources from the target are selected, which helps avoid the negative transfer caused by dissimilar sources. We first exploit the closest sources to reduce the distribution shift with the target and then move towards the furthest while only considering the most relevant sources based on the predetermined threshold. Furthermore, to mitigate catastrophic forgetting caused by the incremental introduction of source subjects, we implemented a density-based memory mechanism that preserves the most relevant historical source samples for adaptation. Our extensive experiments on Biovid, UNBC-McMaster, Aff-Wild2, BAH, and in a cross-dataset setting.

Armita Mohammadi, Alessandro Lameiras Koerich, Laureano Moro-Velazquez et al.

Second language proficiency (L2) in English is usually perceptually evaluated by English teachers or expert evaluators, with the inherent intra- and inter-rater variability. This paper explores deep learning techniques for comprehensive L2 proficiency assessment, addressing both the speech signal and its correspondent transcription. We analyze spoken proficiency classification prediction using diverse architectures, including 2D CNN, frequency-based CNN, ResNet, and a pretrained wav2vec 2.0 model. Additionally, we examine text-based proficiency assessment by fine-tuning a BERT language model within resource constraints. Finally, we tackle the complex task of spontaneous dialogue assessment, managing long-form audio and speaker interactions through separate applications of wav2vec 2.0 and BERT models. Results from experiments on EFCamDat and ANGLISH datasets and a private dataset highlight the potential of deep learning, especially the pretrained wav2vec 2.0 model, for robust automated L2 proficiency evaluation.

Israel A. Laurensi, Alceu de Souza Britto, Jean Paul Barddal et al.

Facial expression recognition is a pivotal component in machine learning, facilitating various applications. However, convolutional neural networks (CNNs) are often plagued by catastrophic forgetting, impeding their adaptability. The proposed method, emotion-centered generative replay (ECgr), tackles this challenge by integrating synthetic images from generative adversarial networks. Moreover, ECgr incorporates a quality assurance algorithm to ensure the fidelity of generated images. This dual approach enables CNNs to retain past knowledge while learning new tasks, enhancing their performance in emotion recognition. The experimental results on four diverse facial expression datasets demonstrate that incorporating images generated by our pseudo-rehearsal method enhances training on the targeted dataset and the source dataset while making the CNN retain previously learned knowledge.

Cristiano Mesquita Garcia, Alessandro Lameiras Koerich, Alceu de Souza Britto et al.

The proliferation of textual data on the Internet presents a unique opportunity for institutions and companies to monitor public opinion about their services and products. Given the rapid generation of such data, the text stream mining setting, which handles sequentially arriving, potentially infinite text streams, is often more suitable than traditional batch learning. While pre-trained language models are commonly employed for their high-quality text vectorization capabilities in streaming contexts, they face challenges adapting to concept drift - the phenomenon where the data distribution changes over time, adversely affecting model performance. Addressing the issue of concept drift, this study explores the efficacy of seven text sampling methods designed to selectively fine-tune language models, thereby mitigating performance degradation. We precisely assess the impact of these methods on fine-tuning the SBERT model using four different loss functions. Our evaluation, focused on Macro F1-score and elapsed time, employs two text stream datasets and an incremental SVM classifier to benchmark performance. Our findings indicate that Softmax loss and Batch All Triplets loss are particularly effective for text stream classification, demonstrating that larger sample sizes generally correlate with improved macro F1-scores. Notably, our proposed WordPieceToken ratio sampling method significantly enhances performance with the identified loss functions, surpassing baseline results.

Muhammad Osama Zeeshan, Natacha Gillet, Alessandro Lameiras Koerich et al.

Personalized expression recognition (ER) involves adapting a machine learning model to subject-specific data for improved recognition of expressions with considerable interpersonal variability. Subject-specific ER can benefit significantly from multi-source domain adaptation (MSDA) methods, where each domain corresponds to a specific subject, to improve model accuracy and robustness. Despite promising results, state-of-the-art MSDA approaches often overlook multimodal information or blend sources into a single domain, limiting subject diversity and failing to explicitly capture unique subject-specific characteristics. To address these limitations, we introduce MuSACo, a multi-modal subject-specific selection and adaptation method for ER based on co-training. It leverages complementary information across multiple modalities and multiple source domains for subject-specific adaptation. This makes MuSACo particularly relevant for affective computing applications in digital health, such as patient-specific assessment for stress or pain, where subject-level nuances are crucial. MuSACo selects source subjects relevant to the target and generates pseudo-labels using the dominant modality for class-aware learning, in conjunction with a class-agnostic loss to learn from less confident target samples. Finally, source features from each modality are aligned, while only confident target features are combined. Our experimental results on challenging multimodal ER datasets: BioVid and StressID, show that MuSACo can outperform UDA (blending) and state-of-the-art MSDA methods.

Masoumeh Sharafi, Soufiane Belharbi, Houssem Ben Salem et al.

Facial expression recognition (FER) models are employed in many video-based affective computing applications, such as human-computer interaction and healthcare monitoring. However, deep FER models often struggle with subtle expressions and high inter-subject variability, limiting their performance in real-world applications. To improve their performance, source-free domain adaptation (SFDA) methods have been proposed to personalize a pretrained source model using only unlabeled target domain data, thereby avoiding data privacy, storage, and transmission constraints. This paper addresses a challenging scenario where source data is unavailable for adaptation, and only unlabeled target data consisting solely of neutral expressions is available. SFDA methods are not typically designed to adapt using target data from only a single class. Further, using models to generate facial images with non-neutral expressions can be unstable and computationally intensive. In this paper, personalized feature translation (PFT) is proposed for SFDA. Unlike current image translation methods for SFDA, our lightweight method operates in the latent space. We first pre-train the translator on the source domain data to transform the subject-specific style features from one source subject into another. Expression information is preserved by optimizing a combination of expression consistency and style-aware objectives. Then, the translator is adapted on neutral target data, without using source data or image synthesis. By translating in the latent space, PFT avoids the complexity and noise of face expression generation, producing discriminative embeddings optimized for classification. Using PFT eliminates the need for image synthesis, reduces computational overhead (using a lightweight translator), and only adapts part of the model, making the method efficient compared to image-based translation.

Luciana Trinkaus Menon, Luiz Carlos Ribeiro Neduziak, Jean Paul Barddal et al.

The study of human emotions, traditionally a cornerstone in fields like psychology and neuroscience, has been profoundly impacted by the advent of artificial intelligence (AI). Multiple channels, such as speech (voice) and facial expressions (image), are crucial in understanding human emotions. However, AI's journey in multimodal emotion recognition (MER) is marked by substantial technical challenges. One significant hurdle is how AI models manage the absence of a particular modality - a frequent occurrence in real-world situations. This study's central focus is assessing the performance and resilience of two strategies when confronted with the lack of one modality: a novel multimodal dynamic modality and view selection and a cross-attention mechanism. Results on the RECOLA dataset show that dynamic selection-based methods are a promising approach for MER. In the missing modalities scenarios, all dynamic selection-based methods outperformed the baseline. The study concludes by emphasizing the intricate interplay between audio and video modalities in emotion prediction, showcasing the adaptability of dynamic selection methods in handling missing modalities.

Steve Tsham Mpinda Ataky, Alessandro Lameiras Koerich

Breast cancer is a health problem that affects mainly the female population. An early detection increases the chances of effective treatment, improving the prognosis of the disease. In this regard, computational tools have been proposed to assist the specialist in interpreting the breast digital image exam, providing features for detecting and diagnosing tumors and cancerous cells. Nonetheless, detecting tumors with a high sensitivity rate and reducing the false positives rate is still challenging. Texture descriptors have been quite popular in medical image analysis, particularly in histopathologic images (HI), due to the variability of both the texture found in such images and the tissue appearance due to irregularity in the staining process. Such variability may exist depending on differences in staining protocol such as fixation, inconsistency in the staining condition, and reagents, either between laboratories or in the same laboratory. Textural feature extraction for quantifying HI information in a discriminant way is challenging given the distribution of intrinsic properties of such images forms a non-deterministic complex system. This paper proposes a method for characterizing texture across HIs with a considerable success rate. By employing ecological diversity measures and discrete wavelet transform, it is possible to quantify the intrinsic properties of such images with promising accuracy on two HI datasets compared with state-of-the-art methods.

Safaa Allamy, Alessandro Lameiras Koerich

This paper proposes a 1D residual convolutional neural network (CNN) architecture for music genre classification and compares it with other recent 1D CNN architectures. The 1D CNNs learn a representation and a discriminant directly from the raw audio signal. Several convolutional layers capture the time-frequency characteristics of the audio signal and learn various filters relevant to the music genre recognition task. The proposed approach splits the audio signal into overlapped segments using a sliding window to comply with the fixed-length input constraint of the 1D CNNs. As a result, music genre classification can be carried out on a single audio segment or on the aggregation of the predictions on several audio segments, which improves the final accuracy. The performance of the proposed 1D residual CNN is assessed on a public dataset of 1,000 audio clips. The experimental results have shown that it achieves 80.93% of mean accuracy in classifying music genres and outperforms other 1D CNN architectures.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

This paper proposes a new defense approach for counteracting state-of-the-art white and black-box adversarial attack algorithms. Our approach fits into the implicit reactive defense algorithm category since it does not directly manipulate the potentially malicious input signals. Instead, it reconstructs a similar signal with a synthesized spectrogram using a cyclic generative adversarial network. This cyclic framework helps to yield a stable generative model. Finally, we feed the reconstructed signal into the speech-to-text model for transcription. The conducted experiments on targeted and non-targeted adversarial attacks developed for attacking DeepSpeech, Kaldi, and Lingvo models demonstrate the proposed defense's effectiveness in adverse scenarios.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

This paper introduces a novel adversarial algorithm for attacking the state-of-the-art speech-to-text systems, namely DeepSpeech, Kaldi, and Lingvo. Our approach is based on developing an extension for the conventional distortion condition of the adversarial optimization formulation using the Cramèr integral probability metric. Minimizing over this metric, which measures the discrepancies between original and adversarial samples' distributions, contributes to crafting signals very close to the subspace of legitimate speech recordings. This helps to yield more robust adversarial signals against playback over-the-air without employing neither costly expectation over transformation operations nor static room impulse response simulations. Our approach outperforms other targeted and non-targeted algorithms in terms of word error rate and sentence-level-accuracy with competitive performance on the crafted adversarial signals' quality. Compared to seven other strong white and black-box adversarial attacks, our proposed approach is considerably more resilient against multiple consecutive playbacks over-the-air, corroborating its higher robustness in noisy environments.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

This paper introduces a defense approach against end-to-end adversarial attacks developed for cutting-edge speech-to-text systems. The proposed defense algorithm has four major steps. First, we represent speech signals with 2D spectrograms using the short-time Fourier transform. Second, we iteratively find a safe vector using a spectrogram subspace projection operation. This operation minimizes the chordal distance adjustment between spectrograms with an additional regularization term. Third, we synthesize a spectrogram with such a safe vector using a novel GAN architecture trained with Sobolev integral probability metric. To improve the model's performance in terms of stability and the total number of learned modes, we impose an additional constraint on the generator network. Finally, we reconstruct the signal from the synthesized spectrogram and the Griffin-Lim phase approximation technique. We evaluate the proposed defense approach against six strong white and black-box adversarial attacks benchmarked on DeepSpeech, Kaldi, and Lingvo models. Our experimental results show that our algorithm outperforms other state-of-the-art defense algorithms both in terms of accuracy and signal quality.

Steve Tsham Mpinda Ataky, Alessandro Lameiras Koerich

Texture can be defined as the change of image intensity that forms repetitive patterns, resulting from physical properties of the object's roughness or differences in a reflection on the surface. Considering that texture forms a complex system of patterns in a non-deterministic way, biodiversity concepts can help texture characterization in images. This paper proposes a novel approach capable of quantifying such a complex system of diverse patterns through species diversity and richness and taxonomic distinctiveness. The proposed approach considers each image channel as a species ecosystem and computes species diversity and richness measures as well as taxonomic measures to describe the texture. The proposed approach takes advantage of ecological patterns' invariance characteristics to build a permutation, rotation, and translation invariant descriptor. Experimental results on three datasets of natural texture images and two datasets of histopathological images have shown that the proposed texture descriptor has advantages over several texture descriptors and deep methods.

Jonathan de Matos, Steve Tsham Mpinda Ataky, Alceu de Souza Britto et al.

Histopathological images (HIs) are the gold standard for evaluating some types of tumors for cancer diagnosis. The analysis of such images is not only time and resource consuming, but also very challenging even for experienced pathologists, resulting in inter- and intra-observer disagreements. One of the ways of accelerating such an analysis is to use computer-aided diagnosis (CAD) systems. In this paper, we present a review on machine learning methods for histopathological image analysis, including shallow and deep learning methods. We also cover the most common tasks in HI analysis, such as segmentation and feature extraction. In addition, we present a list of publicly available and private datasets that have been used in HI research.

Thomas Teixeira, Eric Granger, Alessandro Lameiras Koerich

Facial expressions are one of the most powerful ways for depicting specific patterns in human behavior and describing human emotional state. Despite the impressive advances of affective computing over the last decade, automatic video-based systems for facial expression recognition still cannot handle properly variations in facial expression among individuals as well as cross-cultural and demographic aspects. Nevertheless, recognizing facial expressions is a difficult task even for humans. In this paper, we investigate the suitability of state-of-the-art deep learning architectures based on convolutional neural networks (CNNs) for continuous emotion recognition using long video sequences captured in-the-wild. This study focuses on deep learning models that allow encoding spatiotemporal relations in videos considering a complex and multi-dimensional emotion space, where values of valence and arousal must be predicted. We have developed and evaluated convolutional recurrent neural networks combining 2D-CNNs and long short term-memory units, and inflated 3D-CNN models, which are built by inflating the weights of a pre-trained 2D-CNN model during fine-tuning, using application-specific videos. Experimental results on the challenging SEWA-DB dataset have shown that these architectures can effectively be fine-tuned to encode the spatiotemporal information from successive raw pixel images and achieve state-of-the-art results on such a dataset.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

In this paper we propose a novel defense approach against end-to-end adversarial attacks developed to fool advanced speech-to-text systems such as DeepSpeech and Lingvo. Unlike conventional defense approaches, the proposed approach does not directly employ low-level transformations such as autoencoding a given input signal aiming at removing potential adversarial perturbation. Instead of that, we find an optimal input vector for a class conditional generative adversarial network through minimizing the relative chordal distance adjustment between a given test input and the generator network. Then, we reconstruct the 1D signal from the synthesized spectrogram and the original phase information derived from the given input signal. Hence, this reconstruction does not add any extra noise to the signal and according to our experimental results, our defense-GAN considerably outperforms conventional defense algorithms both in terms of word error rate and sentence level recognition accuracy.

Mohammad Esmaeilpour, Raymel Alfonso Sallo, Olivier St-Georges et al.

In this paper we propose a conditioning trick, called difference departure from normality, applied on the generator network in response to instability issues during GAN training. We force the generator to get closer to the departure from normality function of real samples computed in the spectral domain of Schur decomposition. This binding makes the generator amenable to truncation and does not limit exploring all the possible modes. We slightly modify the BigGAN architecture incorporating residual network for synthesizing 2D representations of audio signals which enables reconstructing high quality sounds with some preserved phase information. Additionally, the proposed conditional training scenario makes a trade-off between fidelity and variety for the generated spectrograms. The experimental results on UrbanSound8k and ESC-50 environmental sound datasets and the Mozilla common voice dataset have shown that the proposed GAN configuration with the conditioning trick remarkably outperforms baseline architectures, according to three objective metrics: inception score, Frechet inception distance, and signal-to-noise ratio.

Mohammad Esmaeilpour, Raymel Alfonso Sallo, Olivier St-Georges et al.

In this paper we address the instability issue of generative adversarial network (GAN) by proposing a new similarity metric in unitary space of Schur decomposition for 2D representations of audio and speech signals. We show that encoding departure from normality computed in this vector space into the generator optimization formulation helps to craft more comprehensive spectrograms. We demonstrate the effectiveness of binding this metric for enhancing stability in training with less mode collapse compared to baseline GANs. Experimental results on subsets of UrbanSound8k and Mozilla common voice datasets have shown considerable improvements on the quality of the generated samples measured by the Fréchet inception distance. Moreover, reconstructed signals from these samples, have achieved higher signal to noise ratio compared to regular LS-GANs.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

In this paper, we investigate the impact of different standard environmental sound representations (spectrograms) on the recognition performance and adversarial attack robustness of a victim residual convolutional neural network. Averaged over various experiments on three benchmarking environmental sound datasets, we found the ResNet-18 model outperforms other deep learning architectures such as GoogLeNet and AlexNet both in terms of classification accuracy and the number of training parameters. Therefore we set this model as our front-end classifier for subsequent investigations. Herein, we measure the impact of different settings required for generating more informative mel-frequency cepstral coefficient (MFCC), short-time Fourier transform (STFT), and discrete wavelet transform (DWT) representations on our front-end model. This measurement involves comparing the classification performance over the adversarial robustness. On the balance of average budgets allocated by adversary and the cost of attack, we demonstrate an inverse relationship between recognition accuracy and model robustness against six attack algorithms. Moreover, our experimental results show that while the ResNet-18 model trained on DWT spectrograms achieves the highest recognition accuracy, attacking this model is relatively more costly for the adversary compared to other 2D representations.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

Adversarial attacks have always been a serious threat for any data-driven model. In this paper, we explore subspaces of adversarial examples in unitary vector domain, and we propose a novel detector for defending our models trained for environmental sound classification. We measure chordal distance between legitimate and malicious representation of sounds in unitary space of generalized Schur decomposition and show that their manifolds lie far from each other. Our front-end detector is a regularized logistic regression which discriminates eigenvalues of legitimate and adversarial spectrograms. The experimental results on three benchmarking datasets of environmental sounds represented by spectrograms reveal high detection rate of the proposed detector for eight types of adversarial attacks and outperforms other detection approaches.

Karl Michel Koerich, Mohammad Esmaeilpour, Sajjad Abdoli et al.

This paper shows the susceptibility of spectrogram-based audio classifiers to adversarial attacks and the transferability of such attacks to audio waveforms. Some commonly used adversarial attacks to images have been applied to Mel-frequency and short-time Fourier transform spectrograms, and such perturbed spectrograms are able to fool a 2D convolutional neural network (CNN). Such attacks produce perturbed spectrograms that are visually imperceptible by humans. Furthermore, the audio waveforms reconstructed from the perturbed spectrograms are also able to fool a 1D CNN trained on the original audio. Experimental results on a dataset of western music have shown that the 2D CNN achieves up to 81.87% of mean accuracy on legitimate examples and such performance drops to 12.09% on adversarial examples. Likewise, the 1D CNN achieves up to 78.29% of mean accuracy on original audio samples and such performance drops to 27.91% on adversarial audio waveforms reconstructed from the perturbed spectrograms.

Alexandre Reeberg de Mello, Marcelo Ricardo Stemmer, Alessandro Lameiras Koerich

In this paper we present an incremental variant of the Twin Support Vector Machine (TWSVM) called Fuzzy Bounded Twin Support Vector Machine (FBTWSVM) to deal with large datasets and learning from data streams. We combine the TWSVM with a fuzzy membership function, so that each input has a different contribution to each hyperplane in a binary classifier. To solve the pair of quadratic programming problems (QPPs) we use a dual coordinate descent algorithm with a shrinking strategy, and to obtain a robust classification with a fast training we propose the use of a Fourier Gaussian approximation function with our linear FBTWSVM. Inspired by the shrinking technique, the incremental algorithm re-utilizes part of the training method with some heuristics, while the decremental procedure is based on a scored window. The FBTWSVM is also extended for multi-class problems by combining binary classifiers using a Directed Acyclic Graph (DAG) approach. Moreover, we analyzed the theoretical foundations properties of the proposed approach and its extension, and the experimental results on benchmark datasets indicate that the FBTWSVM has a fast training and retraining process while maintaining a robust classification performance.

Kelly Lais Wiggers, Alceu de Souza Britto Junior, Alessandro Lameiras Koerich et al.

This paper describes two approaches for content-based image retrieval and pattern spotting in document images using deep learning. The first approach uses a pre-trained CNN model to cope with the lack of training data, which is fine-tuned to achieve a compact yet discriminant representation of queries and image candidates. The second approach uses a Siamese Convolution Neural Network trained on a previously prepared subset of image pairs from the ImageNet dataset to provide the similarity-based feature maps. In both methods, the learned representation scheme considers feature maps of different sizes which are evaluated in terms of retrieval performance. A robust experimental protocol using two public datasets (Tobacoo-800 and DocExplore) has shown that the proposed methods compare favorably against state-of-the-art document image retrieval and pattern spotting methods.

Mohammed Senoussaoui, Patrick Cardinal, Alessandro Lameiras Koerich

In this paper we present a novel approach for extracting a Bag-of-Words (BoW) representation based on a Neural Network codebook. The conventional BoW model is based on a dictionary (codebook) built from elementary representations which are selected randomly or by using a clustering algorithm on a training dataset. A metric is then used to assign unseen elementary representations to the closest dictionary entries in order to produce a histogram. In the proposed approach, an autoencoder (AE) encompasses the role of both the dictionary creation and the assignment metric. The dimension of the encoded layer of the AE corresponds to the size of the dictionary and the output of its neurons represents the assignment metric. Experimental results for the continuous emotion prediction task on the AVEC 2017 audio dataset have shown an improvement of the Concordance Correlation Coefficient (CCC) from 0.225 to 0.322 for arousal dimension and from 0.244 to 0.368 for valence dimension relative to the conventional BoW version implemented in a baseline system.

Alexandre Reeberg Mello, Jonathan de Matos, Marcelo R. Stemmer et al.

In this paper, we propose the use of a black-box optimization method called deterministic Mesh Adaptive Direct Search (MADS) algorithm with orthogonal directions (Ortho-MADS) for the selection of hyperparameters of Support Vector Machines with a Gaussian kernel. Different from most of the methods in the literature that exploit the properties of the data or attempt to minimize the accuracy of a validation dataset over the first quadrant of (C, gamma), the Ortho-MADS provides convergence proof. We present the MADS, followed by the Ortho-MADS, the dynamic stopping criterion defined by the MADS mesh size and two different search strategies (Nelder-Mead and Variable Neighborhood Search) that contribute to a competitive convergence rate as well as a mechanism to escape from undesired local minima. We have investigated the practical selection of hyperparameters for the Support Vector Machine with a Gaussian kernel, i.e., properly choose the hyperparameters gamma (bandwidth) and C (trade-off) on several benchmark datasets. The experimental results have shown that the proposed approach for hyperparameter tuning consistently finds comparable or better solutions, when using a common configuration, than other methods. We have also evaluated the accuracy and the number of function evaluations of the Ortho-MADS with the Nelder-Mead search strategy and the Variable Neighborhood Search strategy using the mesh size as a stopping criterion, and we have achieved accuracy that no other method for hyperparameters optimization could reach.

Mohammed Senoussaoui, Patrick Cardinal, Najim Dehak et al.

Automatic measuring of speaker sincerity degree is a novel research problem in computational paralinguistics. This paper proposes covariance-based feature vectors to model speech and ensembles of support vector regressors to estimate the degree of sincerity of a speaker. The elements of each covariance vector are pairwise statistics between the short-term feature components. These features are used alone as well as in combination with the ComParE acoustic feature set. The experimental results on the development set of the Sincerity Speech Corpus using a cross-validation procedure have shown an 8.1% relative improvement in the Spearman's correlation coefficient over the baseline system.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

Adversarial audio attacks can be considered as a small perturbation unperceptive to human ears that is intentionally added to the audio signal and causes a machine learning model to make mistakes. This poses a security concern about the safety of machine learning models since the adversarial attacks can fool such models toward the wrong predictions. In this paper we first review some strong adversarial attacks that may affect both audio signals and their 2D representations and evaluate the resiliency of the most common machine learning model, namely deep learning models and support vector machines (SVM) trained on 2D audio representations such as short time Fourier transform (STFT), discrete wavelet transform (DWT) and cross recurrent plot (CRP) against several state-of-the-art adversarial attacks. Next, we propose a novel approach based on pre-processed DWT representation of audio signals and SVM to secure audio systems against adversarial attacks. The proposed architecture has several preprocessing modules for generating and enhancing spectrograms including dimension reduction and smoothing. We extract features from small patches of the spectrograms using speeded up robust feature (SURF) algorithm which are further used to generate a codebook using the K-Means++ algorithm. Finally, codewords are used to train a SVM on the codebook of the SURF-generated vectors. All these steps yield to a novel approach for audio classification that provides a good trade-off between accuracy and resilience. Experimental results on three environmental sound datasets show the competitive performance of proposed approach compared to the deep neural networks both in terms of accuracy and robustness against strong adversarial attacks.

Sajjad Abdoli, Patrick Cardinal, Alessandro Lameiras Koerich

In this paper, we present an end-to-end approach for environmental sound classification based on a 1D Convolution Neural Network (CNN) that learns a representation directly from the audio signal. Several convolutional layers are used to capture the signal's fine time structure and learn diverse filters that are relevant to the classification task. The proposed approach can deal with audio signals of any length as it splits the signal into overlapped frames using a sliding window. Different architectures considering several input sizes are evaluated, including the initialization of the first convolutional layer with a Gammatone filterbank that models the human auditory filter response in the cochlea. The performance of the proposed end-to-end approach in classifying environmental sounds was assessed on the UrbanSound8k dataset and the experimental results have shown that it achieves 89% of mean accuracy. Therefore, the propose approach outperforms most of the state-of-the-art approaches that use handcrafted features or 2D representations as input. Furthermore, the proposed approach has a small number of parameters compared to other architectures found in the literature, which reduces the amount of data required for training.

Mohammad Esmaeilpour, Patrick Cardinal, Alessandro Lameiras Koerich

In this paper we propose a novel environmental sound classification approach incorporating unsupervised feature learning from codebook via spherical $K$-Means++ algorithm and a new architecture for high-level data augmentation. The audio signal is transformed into a 2D representation using a discrete wavelet transform (DWT). The DWT spectrograms are then augmented by a novel architecture for cycle-consistent generative adversarial network. This high-level augmentation bootstraps generated spectrograms in both intra and inter class manners by translating structural features from sample to sample. A codebook is built by coding the DWT spectrograms with the speeded-up robust feature detector (SURF) and the K-Means++ algorithm. The Random Forest is our final learning algorithm which learns the environmental sound classification task from the clustered codewords in the codebook. Experimental results in four benchmarking environmental sound datasets (ESC-10, ESC-50, UrbanSound8k, and DCASE-2017) have shown that the proposed classification approach outperforms the state-of-the-art classifiers in the scope, including advanced and dense convolutional neural networks such as AlexNet and GoogLeNet, improving the classification rate between 3.51% and 14.34%, depending on the dataset.