How do multi-modal generative models describe images of recent historical events and figures, whose legacies may be nuanced, multifaceted, or contested? This task necessitates not only accurate visual recognition, but also socio-cultural knowledge and cross-modal reasoning. To address this evaluation challenge, we introduce Century -- a novel dataset of sensitive historical images. This dataset consists of 1,500 images from recent history, created through an automated method combining knowledge graphs and language models with quality and diversity criteria created from the practices of museums and digital archives. We demonstrate through automated and human evaluation that this method produces a set of images that depict events and figures that are diverse across topics and represents all regions of the world. We additionally propose an evaluation framework for evaluating the historical contextualisation capabilities along dimensions of accuracy, thoroughness, and objectivity. We demonstrate this approach by using Century to evaluate four foundation models, scoring performance using both automated and human evaluation. We find that historical contextualisation of sensitive images poses a significant challenge for modern multi-modal foundation models, and offer practical recommendations for how developers can use Century to evaluate improvements to models and applications.

While text-to-image (T2I) generative models have become ubiquitous, they do not necessarily generate images that align with a given prompt. While many metrics and benchmarks have been proposed to evaluate T2I models and alignment metrics, the impact of the evaluation components (prompt sets, human annotations, evaluation task) has not been systematically measured. We find that looking at only *one slice of data*, i.e. one set of capabilities or human annotations, is not enough to obtain stable conclusions that generalise to new conditions or slices when evaluating T2I models or alignment metrics. We address this by introducing an evaluation suite of $>$100K annotations across four human annotation templates that comprehensively evaluates models' capabilities across a range of methods for gathering human annotations and comparing models. In particular, we propose (1) a carefully curated set of prompts -- *Gecko2K*; (2) a statistically grounded method of comparing T2I models; and (3) how to systematically evaluate metrics under three *evaluation tasks* -- *model ordering, pair-wise instance scoring, point-wise instance scoring*. Using this evaluation suite, we evaluate a wide range of metrics and find that a metric may do better in one setting but worse in another. As a result, we introduce a new, interpretable auto-eval metric that is consistently better correlated with human ratings than such existing metrics on our evaluation suite--across different human templates and evaluation settings--and on TIFA160.

The ability to properly benchmark model performance in the face of spurious correlations is important to both build better predictors and increase confidence that models are operating as intended. We demonstrate that characterizing (as opposed to simply quantifying) model mistakes across subgroups is pivotal to properly reflect model biases, which are ignored by standard metrics such as worst-group accuracy or accuracy gap. Inspired by the hypothesis testing framework, we introduce SkewSize, a principled and flexible metric that captures bias from mistakes in a model's predictions. It can be used in multi-class settings or generalised to the open vocabulary setting of generative models. SkewSize is an aggregation of the effect size of the interaction between two categorical variables: the spurious variable representing the bias attribute the model's prediction. We demonstrate the utility of SkewSize in multiple settings including: standard vision models trained on synthetic data, vision models trained on ImageNet, and large scale vision-and-language models from the BLIP-2 family. In each case, the proposed SkewSize is able to highlight biases not captured by other metrics, while also providing insights on the impact of recently proposed techniques, such as instruction tuning.

Text-to-image generative models are capable of producing high-quality images that often faithfully depict concepts described using natural language. In this work, we comprehensively evaluate a range of text-to-image models on numerical reasoning tasks of varying difficulty, and show that even the most advanced models have only rudimentary numerical skills. Specifically, their ability to correctly generate an exact number of objects in an image is limited to small numbers, it is highly dependent on the context the number term appears in, and it deteriorates quickly with each successive number. We also demonstrate that models have poor understanding of linguistic quantifiers (such as “few” or “as many as”), the concept of zero, and struggle with more advanced concepts such as fractional representations. We bundle prompts, generated images and human annotations into GeckoNum, a novel benchmark for evaluation of numerical reasoning.

Failures of fairness or robustness in machine learning predictive settings can be due to undesired dependencies between covariates, outcomes and auxiliary factors of variation. A common strategy to mitigate these failures is data balancing, which attempts to remove those undesired dependencies. In this work, we define conditions on the training distribution for data balancing to lead to fair or robust models. Our results display that in many cases, the balanced distribution does not correspond to selectively removing the undesired dependencies in a causal graph of the task, leading to multiple failure modes and even interference with other mitigation techniques such as regularization. Overall, our results highlight the importance of taking the causal graph into account before performing data balancing.

In this contribution, we augment the metric learning setting by introducing a parametric pseudo-distance, trained jointly with the encoder. Several interpretations are thus drawn for the learned distance-like model’s output. We first show it approximates a likelihood ratio which can be used for hypothesis tests, and that it further induces a large divergence across the joint distributions of pairs of examples from the same and from different classes. Evaluation is performed under the verification setting consisting of determining whether sets of examples belong to the same class, even if such classes are novel and were never presented to the model during training. Empirical evaluation shows such method defines an end-to-end approach for the verification problem, able to attain better performance than simple scorers such as those based on cosine similarity and further outperforming widely used downstream classifiers. We further observe training is much simplified under the proposed approach compared to metric learning with actual distances, requiring no complex scheme to harvest pairs of examples.

Generative Adversarial Networks (GANs) can successfully approximate a probability distribution and produce realistic samples. However, open questions such as sufficient convergence conditions and mode collapse still persist. In this paper, we build on existing work in the area by proposing a novel framework for training the generator against an ensemble of discriminator networks, which can be seen as a one-student/multiple-teachers setting. We formalize this problem within the full-information adversarial bandit framework, where we evaluate the capability of an algorithm to select mixtures of discriminators for providing the generator with feedback during learning. To this end, we propose a reward function which reflects the progress made by the generator and dynamically update the mixture weights allocated to each discriminator. We also draw connections between our algorithm and stochastic optimization methods and then show that existing approaches using multiple discriminators in literature can be recovered from our framework. We argue that less expressive discriminators are smoother and have a general coarse grained view of the modes map, which enforces the generator to cover a wide portion of the data distribution support. On the other hand, highly expressive discriminators ensure samples quality. Finally, experimental results show that our approach improves samples quality and diversity over existing baselines by effectively learning a curriculum. These results also support the claim that weaker discriminators have higher entropy improving modes coverage.

Recent literature has demonstrated promising results for training Generative Adversarial Networks by employing a set of discriminators, in contrast to the traditional game involving one generator against a single adversary. Such methods perform single-objective optimization on some simple consolidation of the losses, e.g. an arithmetic average. In this work, we revisit the multiple-discriminator setting by framing the simultaneous minimization of losses provided by different models as a multi-objective optimization problem. Specifically, we evaluate the performance of multiple gradient descent and the hypervolume maximization algorithm on a number of different datasets. Moreover, we argue that the previously proposed methods and hypervolume maximization can all be seen as variations of multiple gradient descent in which the update direction can be computed efficiently. Our results indicate that hypervolume maximization presents a better compromise between sample quality and computational cost than previous methods.

Assessment of mental workload in real world conditions is key to ensure the performance of workers executing tasks which demand sustained attention. Previous literature has employed electroencephalography (EEG) to this end. However, EEG correlates of mental workload vary across subjects and physical strain, thus making it difficult to devise models capable of simultaneously presenting reliable performance across users. The field of domain adaptation (DA) aims at developing methods that allow for generalization across different domains by learning domain-invariant representations. Such DA methods, however, rely on the so-called covariate shift assumption, which typically does not hold for EEG-based applications. As such, in this paper we propose a way to measure the statistical (marginal and conditional) shift observed on data obtained from different users and use this measure to quantitatively assess the effectiveness of different adaptation strategies. In particular, we use EEG data collected from individuals performing a mental task while running in a treadmill and explore the effects of different normalization strategies commonly used to mitigate cross-subject variability. We show the effects that different normalization schemes have on statistical shifts and their relationship with the accuracy of mental workload prediction as assessed on unseen participants at train time.

Modern applications of artificial neural networks have yielded remarkable performance gains in a wide range of tasks. However, recent studies have discovered that such modelling strategy is vulnerable to Adversarial Examples, i.e. examples with subtle perturbations often too small and imperceptible to humans, but that can easily fool neural networks. Defense techniques against adversarial examples have been proposed, but ensuring robust performance against varying or novel types of attacks remains an open problem. In this work, we focus on the detection setting, in which case attackers become identifiable while models remain vulnerable. Particularly, we employ the decision layer of independently trained models as features for posterior detection. The proposed framework does not require any prior knowledge of adversarial examples generation techniques, and can be directly employed along with unmodified off-the-shelf models. Experiments on the standard MNIST and CIFAR10 datasets deliver empirical evidence that such detection approach generalizes well across not only different adversarial examples generation methods but also quality degradation attacks. Non-linear binary classifiers trained on top of our proposed features can achieve a high detection rate (>90%) in a set of white-box attacks and maintain such performance when tested against unseen attacks.

Stress is well known to have negative effects on health and workplace performance. Physiological sensing using wearables shows in turn great potential for realtime stress monitoring. While some off-the-shelf consumer products (e.g. smartwatches) already feature stress detection, there is still a pressing need to improve the robustness of these models in ecological settings where physical activity can hamper detection accuracy. In this paper, we show that using a multimodal physiological stress model can not only improve model accuracy, but can increase robustness to physical activity inference. To do so, we propose a video game based method to elicit emotional responses. More specifically, 48 participants played video games in which psychological stress and physical activity were jointly modulated. Physiological features showing robustness to stress are analyzed in order to guide further research.

Assessment of mental workload is crucial in safety-critical applications. Often, such applications require the user to be ambulant, such as first responders (e.g., paramedics, firefighters, or police officers). Typically, mental workload models have relied on electroencephalography (EEG) signals. EEGs, however, are known to be highly sensitive to movement artefacts, thus limited applications exist for ambulant users and studies have mostly occurred in controlled laboratory settings. In this paper, we explore the robustness of new non-linear features against movement artefacts and test their effectiveness in monitoring mental workload for ambulant users with the end goal of developing mitigation measures based on mental state of operators. To this end, an EEG experiment was conducted where mental workload and physical activity levels were modulated simultaneously and data was collected from 48 participants. Classical EEG features used for workload assessment, such as spectral power and amplitude/phase coherence, were used as benchmarks and compared against the proposed non-linear multi-scale permutation entropy features. Experimental results show the proposed features consistently outperforming the benchmark ones, thus high-lighting their robustness to movement artefacts.

Assessment of mental workload is crucial for applications which require constant attention and where conditions such as mental fatigue and drowsiness must be avoided. As such, electroencephalography (EEG) based mental workload models have been developed in the past. The majority of these models, however, have assumed individuals are not ambulant, thus bypassing the issue of movement-related EEG artefacts. While such models may be useful for a number of applications (e.g., operators are sitting), they may not apply in situations in which operators are performing their task under different physical activity levels. Representative examples can include first responders, such as paramedics, firefighters, or police officers. In this work, we take the first steps towards overcoming this limitation and present results of an experiment simultaneously eliciting increasing mental workload states at varying physical activity levels. EEG data from forty-seven participants was collected while they performed the NASA Revised Multi-Attribute Task Battery II (MATB-II) under three different activity level conditions (no, medium, high). In this study, we report the effects of activity on the noise-robustness and distribution of several spectral, amplitude/phase coherence, and amplitude modulation features, with the ultimate goal of deriving a feature set tailored towards automated workload assessment during physical activity. Preliminary results show spectral features acquired from the frontal area of the cortex as the most promising and that activity aware mental workload models should be developed.

Automated mental workload measurement is particularly important in safety-critical settings, such as in nuclear plants, aviation, air traffic control, shipping, and transportation, to name a few. As an example, recent statistics have suggested that 90% of the accidents in the transport industry are due to human factors. In this paper, we explore the potential of off-the-shelf wearable technologies in monitoring mental workload in real-time, thus potentially reducing the number of accidents due to human errors. Wearable technologies, while providing the user with ease-of-use, comfort, and portability, have several limitations, such as lower quality signal readings (e.g., due to dry electrodes) and smaller number of recording sites. Such limitations place a burden on the accuracy of existing mental workload models. To overcome this limitation, we propose the use of phase-amplitude and amplitude-amplitude coupling features computed from a portable commercial electroencephalography (EEG) device. Experiments with three different tasks, namely N-back, mental rotation and visual search, show the proposed features being significantly correlated with multiple dimensions of the widely-used NASA task load index test and providing complementary information to other conventional features.