Nadia Bianchi‐Berthouze

Mohammad Mahdi Dehshibi, Temitayo Olugbade, Fernando Diaz-de-Maria et al.

There is a growing body of studies on applying deep learning to biometrics analysis. Certain circumstances, however, could impair the objective measures and accuracy of the proposed biometric data analysis methods. For instance, people with chronic pain (CP) unconsciously adapt specific body movements to protect themselves from injury or additional pain. Because there is no dedicated benchmark database to analyse this correlation, we considered one of the specific circumstances that potentially influence a person's biometrics during daily activities in this study and classified pain level and pain-related behaviour in the EmoPain database. To achieve this, we proposed a sparsely-connected recurrent neural networks (s-RNNs) ensemble with the gated recurrent unit (GRU) that incorporates multiple autoencoders using a shared training framework. This architecture is fed by multidimensional data collected from inertial measurement unit (IMU) and surface electromyography (sEMG) sensors. Furthermore, to compensate for variations in the temporal dimension that may not be perfectly represented in the latent space of s-RNNs, we fused hand-crafted features derived from information-theoretic approaches with represented features in the shared hidden state. We conducted several experiments which indicate that the proposed method outperforms the state-of-the-art approaches in classifying both pain level and pain-related behaviour.

Jitesh Joshi, Nadia Bianchi-Berthouze, Youngjun Cho

Segmentation of thermal facial images is a challenging task. This is because facial features often lack salience due to high-dynamic thermal range scenes and occlusion issues. Limited availability of datasets from unconstrained settings further limits the use of the state-of-the-art segmentation networks, loss functions and learning strategies which have been built and validated for RGB images. To address the challenge, we propose Self-Adversarial Multi-scale Contrastive Learning (SAM-CL) framework as a new training strategy for thermal image segmentation. SAM-CL framework consists of a SAM-CL loss function and a thermal image augmentation (TiAug) module as a domain-specific augmentation technique. We use the Thermal-Face-Database to demonstrate effectiveness of our approach. Experiments conducted on the existing segmentation networks (UNET, Attention-UNET, DeepLabV3 and HRNetv2) evidence the consistent performance gains from the SAM-CL framework. Furthermore, we present a qualitative analysis with UBComfort and DeepBreath datasets to discuss how our proposed methods perform in handling unconstrained situations.

Xijia Wei, Yuan Fang, Kevin Chetty et al.

Millimetre-wave (mmWave) radar offers a more privacy-preserving alternative to RGB-based human pose estimation. However, existing methods typically rely on pre-extracted intermediate representations such as sparse point clouds or spectrogram images, where the rich spatiotemporal information naturally present in radar video streams is discarded for model learning, while such signal processing adds system complexity. In addition, existing solutions are mainly conducted in an end-to-end supervised manner without leveraging unlabelled raw video streams to learn generalized representations. In this study, we present MAEPose, a masked autoencoding-based human pose estimation approach that operates directly on mmWave spectrogram videos. MAEPose learns spatiotemporal motion-aware generalized representations from unlabelled radar video, and leverages its heatmap decoder for multi-frame pose estimation predictions. We evaluate it across three datasets based on leave-one-person-out cross-validation with rigorous statistical testing. MAEPose consistently outperforms state-of-the-art baselines by up to 22.1% in MPJPE p<0.05, and maintains robust accuracy under zero-shot bystander interference with only a 6.5% error increase. Ablation studies confirm that both the pre-training and the heatmap decoder contribute substantially, while modality analysis indicates that leveraging Range-Doppler video as input achieves better pose estimation performance than Range-Azimuth or their fusion, with lower computational cost.

Katherine Wang, Nadia Berthouze, Aneesha Singh

AI systems are increasingly embedded in multi-user social environments, yet most alignment frameworks conceptualize interaction as a dyadic relationship between a single user and an AI system. Livestreaming platforms challenge this assumption: interaction unfolds among streamers and audiences in real time, producing dynamic affective and social feedback loops. In this paper, we introduce the Triadic Loop, a conceptual framework that reconceptualizes alignment in AI co-hosted livestreaming as a temporally reinforced process of bidirectional adaptation among three actors: streamer $\leftrightarrow$ AI co-host, AI co-host $\leftrightarrow$ audience, and streamer $\leftrightarrow$ audience. Unlike instruction-following paradigms, bidirectional alignment requires each actor to continuously reshape the others, meaning misalignment in any sub-loop can destabilize the broader system. Drawing on literature from multi-party interaction, collaborative AI, and relational agents, we articulate how AI co-hosts function not only as mediators but as performative participants and community members shaping collective meaning-making. We further propose "strategic misalignment" as a mechanism for sustaining community engagement and introduce three relational evaluation constructs grounded in established instruments. The framework contributes a model of dynamic multi-party alignment, an account of cross-loop reinforcement, and design implications for AI co-hosts that sustain social coherence in participatory media environments.

Shaoduo Gan, Akhil Mathur, Anton Isopoussu et al.

Breakthroughs in unsupervised domain adaptation (uDA) can help in adapting models from a label-rich source domain to unlabeled target domains. Despite these advancements, there is a lack of research on how uDA algorithms, particularly those based on adversarial learning, can work in distributed settings. In real-world applications, target domains are often distributed across thousands of devices, and existing adversarial uDA algorithms -- which are centralized in nature -- cannot be applied in these settings. To solve this important problem, we introduce FRuDA: an end-to-end framework for distributed adversarial uDA. Through a careful analysis of the uDA literature, we identify the design goals for a distributed uDA system and propose two novel algorithms to increase adaptation accuracy and training efficiency of adversarial uDA in distributed settings. Our evaluation of FRuDA with five image and speech datasets show that it can boost target domain accuracy by up to 50% and improve the training efficiency of adversarial uDA by at least 11 times.

Chongyang Wang, Yuan Gao, Chenyou Fan et al.

The annotation of domain experts is important for some medical applications where the objective ground truth is ambiguous to define, e.g., the rehabilitation for some chronic diseases, and the prescreening of some musculoskeletal abnormalities without further medical examinations. However, improper uses of the annotations may hinder developing reliable models. On one hand, forcing the use of a single ground truth generated from multiple annotations is less informative for the modeling. On the other hand, feeding the model with all the annotations without proper regularization is noisy given existing disagreements. For such issues, we propose a novel Learning to Agreement (Learn2Agree) framework to tackle the challenge of learning from multiple annotators without objective ground truth. The framework has two streams, with one stream fitting with the multiple annotators and the other stream learning agreement information between annotators. In particular, the agreement learning stream produces regularization information to the classifier stream, tuning its decision to be better in line with the agreement between annotators. The proposed method can be easily added to existing backbones, with experiments on two medical datasets showed better agreement levels with annotators.

M. M. N. Bieńkiewicz, A. Smykovskyi, T. Olugbade et al.

Our daily human life is filled with a myriad of joint action moments, be it children playing, adults working together (i.e., team sports), or strangers navigating through a crowd. Joint action brings individuals (and embodiment of their emotions) together, in space and in time. Yet little is known about how individual emotions propagate through embodied presence in a group, and how joint action changes individual emotion. In fact, the multi-agent component is largely missing from neuroscience-based approaches to emotion, and reversely joint action research has not found a way yet to include emotion as one of the key parameters to model socio-motor interaction. In this review, we first identify the gap and then stockpile evidence showing strong entanglement between emotion and acting together from various branches of sciences. We propose an integrative approach to bridge the gap, highlight five research avenues to do so in behavioral neuroscience and digital sciences, and address some of the key challenges in the area faced by modern societies.

Chongyang Wang, Yuan Gao, Akhil Mathur et al.

Protective behavior exhibited by people with chronic pain (CP) during physical activities is the key to understanding their physical and emotional states. Existing automatic protective behavior detection (PBD) methods rely on pre-segmentation of activities predefined by users. However, in real life, people perform activities casually. Therefore, where those activities present difficulties for people with chronic pain, technology-enabled support should be delivered continuously and automatically adapted to activity type and occurrence of protective behavior. Hence, to facilitate ubiquitous CP management, it becomes critical to enable accurate PBD over continuous data. In this paper, we propose to integrate human activity recognition (HAR) with PBD via a novel hierarchical HAR-PBD architecture comprising graph-convolution and long short-term memory (GC-LSTM) networks, and alleviate class imbalances using a class-balanced focal categorical-cross-entropy (CFCC) loss. Through in-depth evaluation of the approach using a CP patients' dataset, we show that the leveraging of HAR, GC-LSTM networks, and CFCC loss leads to clear increase in PBD performance against the baseline (macro F1 score of 0.81 vs. 0.66 and precision-recall area-under-the-curve (PR-AUC) of 0.60 vs. 0.44). We conclude by discussing possible use cases of the hierarchical architecture in CP management and beyond. We also discuss current limitations and ways forward.

Akhil Mathur, Fahim Kawsar, Nadia Berthouze et al.

This paper introduces a new dataset, Libri-Adapt, to support unsupervised domain adaptation research on speech recognition models. Built on top of the LibriSpeech corpus, Libri-Adapt contains English speech recorded on mobile and embedded-scale microphones, and spans 72 different domains that are representative of the challenging practical scenarios encountered by ASR models. More specifically, Libri-Adapt facilitates the study of domain shifts in ASR models caused by a) different acoustic environments, b) variations in speaker accents, c) heterogeneity in the hardware and platform software of the microphones, and d) a combination of the aforementioned three shifts. We also provide a number of baseline results quantifying the impact of these domain shifts on the Mozilla DeepSpeech2 ASR model.

Ahmed Alqaraawi, Martin Schuessler, Philipp Weiß et al.

Convolutional neural networks (CNNs) offer great machine learning performance over a range of applications, but their operation is hard to interpret, even for experts. Various explanation algorithms have been proposed to address this issue, yet limited research effort has been reported concerning their user evaluation. In this paper, we report on an online between-group user study designed to evaluate the performance of "saliency maps" - a popular explanation algorithm for image classification applications of CNNs. Our results indicate that saliency maps produced by the LRP algorithm helped participants to learn about some specific image features the system is sensitive to. However, the maps seem to provide very limited help for participants to anticipate the network's output for new images. Drawing on our findings, we highlight implications for design and further research on explainable AI. In particular, we argue the HCI and AI communities should look beyond instance-level explanations.

Cong Bao, Zafeirios Fountas, Temitayo Olugbade et al.

We propose a novel neural network architecture, named the Global Workspace Network (GWN), which addresses the challenge of dynamic and unspecified uncertainties in multimodal data fusion. Our GWN is a model of attention across modalities and evolving through time, and is inspired by the well-established Global Workspace Theory from the field of cognitive science. The GWN achieved average F1 score of 0.92 for discrimination between pain patients and healthy participants and average F1 score = 0.75 for further classification of three pain levels for a patient, both based on the multimodal EmoPain dataset captured from people with chronic pain and healthy people performing different types of exercise movements in unconstrained settings. In these tasks, the GWN significantly outperforms the typical fusion approach of merging by concatenation. We further provide extensive analysis of the behaviour of the GWN and its ability to address uncertainties (hidden noise) in multimodal data.

Joy O. Egede, Siyang Song, Temitayo A. Olugbade et al.

The EmoPain 2020 Challenge is the first international competition aimed at creating a uniform platform for the comparison of machine learning and multimedia processing methods of automatic chronic pain assessment from human expressive behaviour, and also the identification of pain-related behaviours. The objective of the challenge is to promote research in the development of assistive technologies that help improve the quality of life for people with chronic pain via real-time monitoring and feedback to help manage their condition and remain physically active. The challenge also aims to encourage the use of the relatively underutilised, albeit vital bodily expression signals for automatic pain and pain-related emotion recognition. This paper presents a description of the challenge, competition guidelines, bench-marking dataset, and the baseline systems' architecture and performance on the three sub-tasks: pain estimation from facial expressions, pain recognition from multimodal movement, and protective movement behaviour detection.

Youngjun Cho, Nadia Bianchi-Berthouze

Thermal imaging-based physiological and affective computing is an emerging research area enabling technologies to monitor our bodily functions and understand psychological and affective needs in a contactless manner. However, up to recently, research has been mainly carried out in very controlled lab settings. As small size and even low-cost versions of thermal video cameras have started to appear on the market, mobile thermal imaging is opening its door to ubiquitous and real-world applications. Here we review the literature on the use of thermal imaging to track changes in physiological cues relevant to affective computing and the technological requirements set so far. In doing so, we aim to establish computational and methodological pipelines from thermal images of the human skin to affective states and outline the research opportunities and challenges to be tackled to make ubiquitous real-life thermal imaging-based affect monitoring a possibility.

Youngjun Cho, Nadia Bianchi-Berthouze, Manuel Oliveira et al.

Automatically monitoring and quantifying stress-induced thermal dynamic information in real-world settings is an extremely important but challenging problem. In this paper, we explore whether we can use mobile thermal imaging to measure the rich physiological cues of mental stress that can be deduced from a person's nose temperature. To answer this question we build i) a framework for monitoring nasal thermal variable patterns continuously and ii) a novel set of thermal variability metrics to capture a richness of the dynamic information. We evaluated our approach in a series of studies including laboratory-based psychosocial stress-induction tasks and real-world factory settings. We demonstrate our approach has the potential for assessing stress responses beyond controlled laboratory settings.

Chongyang Wang, Min Peng, Temitayo A. Olugbade et al.

For people with chronic pain, the assessment of protective behavior during physical functioning is essential to understand their subjective pain-related experiences (e.g., fear and anxiety toward pain and injury) and how they deal with such experiences (avoidance or reliance on specific body joints), with the ultimate goal of guiding intervention. Advances in deep learning (DL) can enable the development of such intervention. Using the EmoPain MoCap dataset, we investigate how attention-based DL architectures can be used to improve the detection of protective behavior by capturing the most informative temporal and body configurational cues characterizing specific movements and the strategies used to perform them. We propose an end-to-end deep learning architecture named BodyAttentionNet (BANet). BANet is designed to learn temporal and bodily parts that are more informative to the detection of protective behavior. The approach addresses the variety of ways people execute a movement (including healthy people) independently of the type of movement analyzed. Through extensive comparison experiments with other state-of-the-art machine learning techniques used with motion capture data, we show statistically significant improvements achieved by using these attention mechanisms. In addition, the BANet architecture requires a much lower number of parameters than the state of the art for comparable if not higher performances.

Chongyang Wang, Temitayo A. Olugbade, Akhil Mathur et al.

In chronic pain rehabilitation, physiotherapists adapt physical activity to patients' performance based on their expression of protective behavior, gradually exposing them to feared but harmless and essential everyday activities. As rehabilitation moves outside the clinic, technology should automatically detect such behavior to provide similar support. Previous works have shown the feasibility of automatic protective behavior detection (PBD) within a specific activity. In this paper, we investigate the use of deep learning for PBD across activity types, using wearable motion capture and surface electromyography data collected from healthy participants and people with chronic pain. We approach the problem by continuously detecting protective behavior within an activity rather than estimating its overall presence. The best performance reaches mean F1 score of 0.82 with leave-one-subject-out cross validation. When protective behavior is modelled per activity type, performance is mean F1 score of 0.77 for bend-down, 0.81 for one-leg-stand, 0.72 for sit-to-stand, 0.83 for stand-to-sit, and 0.67 for reach-forward. This performance reaches excellent level of agreement with the average experts' rating performance suggesting potential for personalized chronic pain management at home. We analyze various parameters characterizing our approach to understand how the results could generalize to other PBD datasets and different levels of ground truth granularity.

Youngjun Cho, Simon J. Julier, Nadia Bianchi-Berthouze

Background: A smartphone is a promising tool for daily cardiovascular measurement and mental stress monitoring. A smartphone camera-based PhotoPlethysmoGraphy (PPG) and a low-cost thermal camera can be used to create cheap, convenient and mobile monitoring systems. However, to ensure reliable monitoring results, a person has to remain still for several minutes while a measurement is being taken. This is very cumbersome and makes its use in real-life mobile situations quite impractical. Objective: We propose a system which combines PPG and thermography with the aim of improving cardiovascular signal quality and capturing stress responses quickly. Methods: Using a smartphone camera with a low cost thermal camera added on, we built a novel system which continuously and reliably measures two different types of cardiovascular events: i) blood volume pulse and ii) vasoconstriction/dilation-induced temperature changes of the nose tip. 17 healthy participants, involved in a series of stress-inducing mental workload tasks, measured their physiological responses to stressors over a short window of time (20 seconds) immediately after each task. Participants reported their level of perceived mental stress using a 10-cm Visual Analogue Scale (VAS). We used normalized K-means clustering to reduce interpersonal differences in the self-reported ratings. For the instant stress inference task, we built novel low-level feature sets representing variability of cardiovascular patterns. We then used the automatic feature learning capability of artificial Neural Networks (NN) to improve the mapping between the extracted set of features and the self-reported ratings. We compared our proposed method with existing hand-engineered features-based machine learning methods. Results, Conclusions: ... due to limited space here, we refer to our manuscript.

Youngjun Cho, Nadia Bianchi-Berthouze, Nicolai Marquardt et al.

We introduce Deep Thermal Imaging, a new approach for close-range automatic recognition of materials to enhance the understanding of people and ubiquitous technologies of their proximal environment. Our approach uses a low-cost mobile thermal camera integrated into a smartphone to capture thermal textures. A deep neural network classifies these textures into material types. This approach works effectively without the need for ambient light sources or direct contact with materials. Furthermore, the use of a deep learning network removes the need to handcraft the set of features for different materials. We evaluated the performance of the system by training it to recognise 32 material types in both indoor and outdoor environments. Our approach produced recognition accuracies above 98% in 14,860 images of 15 indoor materials and above 89% in 26,584 images of 17 outdoor materials. We conclude by discussing its potentials for real-time use in HCI applications and future directions.

Youngjun Cho, Andrea Bianchi, Nicolai Marquardt et al.

We present RealPen, an augmented stylus for capacitive tablet screens that recreates the physical sensation of writing on paper with a pencil, ball-point pen or marker pen. The aim is to create a more engaging experience when writing on touch surfaces, such as screens of tablet computers. This is achieved by re-generating the friction-induced oscillation and sound of a real writing tool in contact with paper. To generate realistic tactile feedback, our algorithm analyses the frequency spectrum of the friction oscillation generated when writing with traditional tools, extracts principal frequencies, and uses the actuator's frequency response profile for an adjustment weighting function. We enhance the realism by providing the sound feedback aligned with the writing pressure and speed. Furthermore, we investigated the effects of superposition and fluctuation of several frequencies on human tactile perception, evaluated the performance of RealPen, and characterized users' perception and preference of each feedback type.

Youngjun Cho, Nadia Bianchi-Berthouze, Simon J. Julier et al.

The ability of sensing breathing is becoming an increasingly important function for technology that aims at supporting both psychological and physical wellbeing. We demonstrate ThermSense, a new breathing sensing platform based on smartphone technology and low-cost thermal camera, which allows a user to measure his/her breathing pattern in a contact-free manner. With the designed key functions of Thermal Voxel Integration-based breathing estimation and respiration variability spectrogram (RVS, bi-dimensional representation of breathing dynamics), the developed platform provides scalability and flexibility for gathering respiratory physiological measurements ubiquitously. The functionality could be used for a variety of applications from stress monitoring to respiration training.

Youngjun Cho, Simon J. Julier, Nicolai Marquardt et al.

The ability to monitor respiratory rate is extremely important for medical treatment, healthcare and fitness sectors. In many situations, mobile methods, which allow users to undertake every day activities, are required. However, current monitoring systems can be obtrusive, requiring users to wear respiration belts or nasal probes. Recent advances in thermographic systems have shrunk their size, weight and cost, to the point where it is possible to create smart-phone based respiration rate monitoring devices that are not affected by lighting conditions. However, mobile thermal imaging is challenged in scenes with high thermal dynamic ranges. This challenge is further amplified by general problems such as motion artifacts and low spatial resolution, leading to unreliable breathing signals. In this paper, we propose a novel and robust approach for respiration tracking which compensates for the negative effects of variations in the ambient temperature and motion artifacts and can accurately extract breathing rates in highly dynamic thermal scenes. It has three main contributions. The first is a novel Optimal Quantization technique which adaptively constructs a color mapping of absolute temperature to improve segmentation, classification and tracking. The second is the Thermal Gradient Flow method that computes thermal gradient magnitude maps to enhance accuracy of the nostril region tracking. Finally, we introduce the Thermal Voxel method to increase the reliability of the captured respiration signals compared to the traditional averaging method. We demonstrate the extreme robustness of our system to track the nostril-region and measure the respiratory rate in high dynamic range scenes.

Adriana Hamacher, Nadia Bianchi-Berthouze, Anthony G. Pipe et al.

Strategies are necessary to mitigate the impact of unexpected behavior in collaborative robotics, and research to develop solutions is lacking. Our aim here was to explore the benefits of an affective interaction, as opposed to a more efficient, less error prone but non-communicative one. The experiment took the form of an omelet-making task, with a wide range of participants interacting directly with BERT2, a humanoid robot assistant. Having significant implications for design, results suggest that efficiency is not the most important aspect of performance for users; a personable, expressive robot was found to be preferable over a more efficient one, despite a considerable trade off in time taken to perform the task. Our findings also suggest that a robot exhibiting human-like characteristics may make users reluctant to 'hurt its feelings'; they may even lie in order to avoid this.