Matias Valdenegro-Toro

Alan Preciado-Grijalva, Bilal Wehbe, Miguel Bande Firvida et al.

Self-supervised learning has proved to be a powerful approach to learn image representations without the need of large labeled datasets. For underwater robotics, it is of great interest to design computer vision algorithms to improve perception capabilities such as sonar image classification. Due to the confidential nature of sonar imaging and the difficulty to interpret sonar images, it is challenging to create public large labeled sonar datasets to train supervised learning algorithms. In this work, we investigate the potential of three self-supervised learning methods (RotNet, Denoising Autoencoders, and Jigsaw) to learn high-quality sonar image representation without the need of human labels. We present pre-training and transfer learning results on real-life sonar image datasets. Our results indicate that self-supervised pre-training yields classification performance comparable to supervised pre-training in a few-shot transfer learning setup across all three methods. Code and self-supervised pre-trained models are be available at https://github.com/agrija9/ssl-sonar-images

Deepan Chakravarthi Padmanabhan, Paul G. Plöger, Octavio Arriaga et al.

State-of-the-art object detectors are treated as black boxes due to their highly non-linear internal computations. Even with unprecedented advancements in detector performance, the inability to explain how their outputs are generated limits their use in safety-critical applications. Previous work fails to produce explanations for both bounding box and classification decisions, and generally make individual explanations for various detectors. In this paper, we propose an open-source Detector Explanation Toolkit (DExT) which implements the proposed approach to generate a holistic explanation for all detector decisions using certain gradient-based explanation methods. We suggests various multi-object visualization methods to merge the explanations of multiple objects detected in an image as well as the corresponding detections in a single image. The quantitative evaluation show that the Single Shot MultiBox Detector (SSD) is more faithfully explained compared to other detectors regardless of the explanation methods. Both quantitative and human-centric evaluations identify that SmoothGrad with Guided Backpropagation (GBP) provides more trustworthy explanations among selected methods across all detectors. We expect that DExT will motivate practitioners to evaluate object detectors from the interpretability perspective by explaining both bounding box and classification decisions.

Ivo Pascal de Jong, Andreea Ioana Sburlea, Matias Valdenegro-Toro

Uncertainty Quantification (UQ) has gained traction in an attempt to improve the interpretability and robustness of machine learning predictions. Specifically (medical) biosignals such as electroencephalography (EEG), electrocardiography (ECG), electrooculography (EOG), and electromyography (EMG) could benefit from good UQ, since these suffer from a poor signal-to-noise ratio, and good human interpretability is pivotal for medical applications. In this paper, we review the state of the art of applying Uncertainty Quantification to Machine Learning tasks in the biosignal domain. We present various methods, shortcomings, uncertainty measures and theoretical frameworks that currently exist in this application domain. We address misconceptions in the field, provide recommendations for future work, and discuss gaps in the literature in relation to diagnostic implementations as well as control for prostheses or brain-computer interfaces. Overall it can be concluded that promising UQ methods are available, but that research is needed on how people and systems may interact with an uncertainty-model in a (clinical) environment

Ivo Pascal de Jong, Andreea Ioana Sburlea, Matthia Sabatelli et al.

Aleatoric (data) and epistemic (knowledge) uncertainty are textbook components of Uncertainty Quantification. Jointly estimating these components has been shown to be problematic and non-trivial. As a result, there are multiple ways to disentangle these uncertainties, but current methods to evaluate them are insufficient. We propose that aleatoric and epistemic uncertainty estimates should be orthogonally disentangled - meaning that each uncertainty is not affected by the other - a necessary condition that is often not met. We prove that orthogonality and consistency and necessary and sufficient criteria for disentanglement, and construct Uncertainty Disentanglement Error as a metric to measure these criteria, with further empirical evaluation showing that finetuned models give different orthogonality results than models trained from scratch and that UDE can be optimized for through dropout rate. We demonstrate a Deep Ensemble trained from scratch on ImageNet-1k with Information Theoretic disentangling achieves consistent and orthogonal estimates of epistemic uncertainty, but estimates of aleatoric uncertainty still fail on orthogonality.

Deepan Chakravarthi Padmanabhan, Paul G. Plöger, Octavio Arriaga et al.

Saliency methods are frequently used to explain Deep Neural Network-based models. Adebayo et al.'s work on evaluating saliency methods for classification models illustrate certain explanation methods fail the model and data randomization tests. However, on extending the tests for various state of the art object detectors we illustrate that the ability to explain a model is more dependent on the model itself than the explanation method. We perform sanity checks for object detection and define new qualitative criteria to evaluate the saliency explanations, both for object classification and bounding box decisions, using Guided Backpropagation, Integrated Gradients, and their Smoothgrad versions, together with Faster R-CNN, SSD, and EfficientDet-D0, trained on COCO. In addition, the sensitivity of the explanation method to model parameters and data labels varies class-wise motivating to perform the sanity checks for each class. We find that EfficientDet-D0 is the most interpretable method independent of the saliency method, which passes the sanity checks with little problems.

Merlijn Quincent Mulder, Matias Valdenegro-Toro, Andreea Ioana Sburlea et al.

Machine Learning classifiers used in Brain-Computer Interfaces make classifications based on the distribution of data they were trained on. When they need to make inferences on samples that fall outside of this distribution, they can only make blind guesses. Instead of allowing random guesses, these Out-of-Distribution (OOD) samples should be detected and rejected. We study OOD detection in Motor Imagery BCIs by training a model on some classes and observing whether unfamiliar classes can be detected based on increased uncertainty. We test seven different OOD detection techniques and one more method that has been claimed to boost the quality of OOD detection. Our findings show that OOD detection for Brain-Computer Interfaces is more challenging than in other machine learning domains due to the high uncertainty inherent in classifying EEG signals. For many subjects, uncertainty for in-distribution classes can still be higher than for out-of-distribution classes. As a result, many OOD detection methods prove to be ineffective, though MC Dropout performed best. Additionally, we show that high in-distribution classification performance predicts high OOD detection performance, suggesting that improved accuracy can also lead to improved robustness. Our research demonstrates a setup for studying how models deal with unfamiliar EEG data and evaluates methods that are robust to these unfamiliar inputs. OOD detection can improve the overall safety and reliability of BCIs.

Matias Valdenegro-Toro, Matthia Sabatelli

Overfitting and generalization is an important concept in Machine Learning as only models that generalize are interesting for general applications. Yet some students have trouble learning this important concept through lectures and exercises. In this paper we describe common examples of students misunderstanding overfitting, and provide recommendations for possible solutions. We cover student misconceptions about overfitting, about solutions to overfitting, and implementation mistakes that are commonly confused with overfitting issues. We expect that our paper can contribute to improving student understanding and lectures about this important topic.

Levente Foldesi, Matias Valdenegro-Toro

Increasingly high-stakes decisions are made using neural networks in order to make predictions. Specifically, meteorologists and hedge funds apply these techniques to time series data. When it comes to prediction, there are certain limitations for machine learning models (such as lack of expressiveness, vulnerability of domain shifts and overconfidence) which can be solved using uncertainty estimation. There is a set of expectations regarding how uncertainty should ``behave". For instance, a wider prediction horizon should lead to more uncertainty or the model's confidence should be proportional to its accuracy. In this paper, different uncertainty estimation methods are compared to forecast meteorological time series data and evaluate these expectations. The results show how each uncertainty estimation method performs on the forecasting task, which partially evaluates the robustness of predicted uncertainty.

Martino Pelucchi, Matias Valdenegro-Toro

ChatGPT took the world by storm for its impressive abilities. Due to its release without documentation, scientists immediately attempted to identify its limits, mainly through its performance in natural language processing (NLP) tasks. This paper aims to join the growing literature regarding ChatGPT's abilities by focusing on its performance in high-resource languages and on its capacity to predict its answers' accuracy by giving a confidence level. The analysis of high-resource languages is of interest as studies have shown that low-resource languages perform worse than English in NLP tasks, but no study so far has analysed whether high-resource languages perform as well as English. The analysis of ChatGPT's confidence calibration has not been carried out before either and is critical to learn about ChatGPT's trustworthiness. In order to study these two aspects, five high-resource languages and two NLP tasks were chosen. ChatGPT was asked to perform both tasks in the five languages and to give a numerical confidence value for each answer. The results show that all the selected high-resource languages perform similarly and that ChatGPT does not have a good confidence calibration, often being overconfident and never giving low confidence values.

Lokesh Veeramacheneni, Matias Valdenegro-Toro

Safety-critical applications like autonomous driving use Deep Neural Networks (DNNs) for object detection and segmentation. The DNNs fail to predict when they observe an Out-of-Distribution (OOD) input leading to catastrophic consequences. Existing OOD detection methods were extensively studied for image inputs but have not been explored much for LiDAR inputs. So in this study, we proposed two datasets for benchmarking OOD detection in 3D semantic segmentation. We used Maximum Softmax Probability and Entropy scores generated using Deep Ensembles and Flipout versions of RandLA-Net as OOD scores. We observed that Deep Ensembles out perform Flipout model in OOD detection with greater AUROC scores for both datasets.

Matias Valdenegro-Toro, Daniel Saromo

Neural networks are ubiquitous in many tasks, but trusting their predictions is an open issue. Uncertainty quantification is required for many applications, and disentangled aleatoric and epistemic uncertainties are best. In this paper, we generalize methods to produce disentangled uncertainties to work with different uncertainty quantification methods, and evaluate their capability to produce disentangled uncertainties. Our results show that: there is an interaction between learning aleatoric and epistemic uncertainty, which is unexpected and violates assumptions on aleatoric uncertainty, some methods like Flipout produce zero epistemic uncertainty, aleatoric uncertainty is unreliable in the out-of-distribution setting, and Ensembles provide overall the best disentangling quality. We also explore the error produced by the number of samples hyper-parameter in the sampling softmax function, recommending N > 100 samples. We expect that our formulation and results help practitioners and researchers choose uncertainty methods and expand the use of disentangled uncertainties, as well as motivate additional research into this topic.

Mariela De Lucas Álvarez, Jichen Guo, Raul Domínguez et al.

Terrain Classification is an essential task in space exploration, where unpredictable environments are difficult to observe using only exteroceptive sensors such as vision. Implementing Neural Network classifiers can have high performance but can be deemed untrustworthy as they lack transparency, which makes them unreliable for taking high-stakes decisions during mission planning. We address this by proposing Neural Networks with Uncertainty Quantification in Terrain Classification. We enable our Neural Networks with Monte Carlo Dropout, DropConnect, and Flipout in time series-capable architectures using only proprioceptive data as input. We use Bayesian Optimization with Hyperband for efficient hyperparameter optimization to find optimal models for trustworthy terrain classification.

Kumud Lakara, Matias Valdenegro-Toro

Trusting the predictions of deep learning models in safety critical settings such as the medical domain is still not a viable option. Distentangled uncertainty quantification in the field of medical imaging has received little attention. In this paper, we study disentangled uncertainties in image to image translation tasks in the medical domain. We compare multiple uncertainty quantification methods, namely Ensembles, Flipout, Dropout, and DropConnect, while using CycleGAN to convert T1-weighted brain MRI scans to T2-weighted brain MRI scans. We further evaluate uncertainty behavior in the presence of out of distribution data (Brain CT and RGB Face Images), showing that epistemic uncertainty can be used to detect out of distribution inputs, which should increase reliability of model outputs.

Matias Valdenegro-Toro, Radina Stoykova

The AI act is the European Union-wide regulation of AI systems. It includes specific provisions for general-purpose AI models which however need to be further interpreted in terms of technical standards and state-of-art studies to ensure practical compliance solutions. This paper examines the AI act requirements for providers and deployers of general-purpose AI and further proposes uncertainty estimation as a suitable measure for legal compliance and quality assurance in training of such models. We argue that uncertainty estimation should be a required component for deploying models in the real world, and under the EU AI Act, it could fulfill several requirements for transparency, accuracy, and trustworthiness. However, generally using uncertainty estimation methods increases the amount of computation, producing a dilemma, as computation might go over the threshold ($10^{25}$ FLOPS) to classify the model as a systemic risk system which bears more regulatory burden.

Deepak Singh, Matias Valdenegro-Toro

Accurate detection and segmentation of marine debris is important for keeping the water bodies clean. This paper presents a novel dataset for marine debris segmentation collected using a Forward Looking Sonar (FLS). The dataset consists of 1868 FLS images captured using ARIS Explorer 3000 sensor. The objects used to produce this dataset contain typical house-hold marine debris and distractor marine objects (tires, hooks, valves,etc), divided in 11 classes plus a background class. Performance of state of the art semantic segmentation architectures with a variety of encoders have been analyzed on this dataset and presented as baseline results. Since the images are grayscale, no pretrained weights have been used. Comparisons are made using Intersection over Union (IoU). The best performing model is Unet with ResNet34 backbone at 0.7481 mIoU. The dataset is available at https://github.com/mvaldenegro/marine-debris-fls-datasets/

Octavio Arriaga, Matias Valdenegro-Toro, Paul Plöger

In this paper we propose an implement a general convolutional neural network (CNN) building framework for designing real-time CNNs. We validate our models by creating a real-time vision system which accomplishes the tasks of face detection, gender classification and emotion classification simultaneously in one blended step using our proposed CNN architecture. After presenting the details of the training procedure setup we proceed to evaluate on standard benchmark sets. We report accuracies of 96% in the IMDB gender dataset and 66% in the FER-2013 emotion dataset. Along with this we also introduced the very recent real-time enabled guided back-propagation visualization technique. Guided back-propagation uncovers the dynamics of the weight changes and evaluates the learned features. We argue that the careful implementation of modern CNN architectures, the use of the current regularization methods and the visualization of previously hidden features are necessary in order to reduce the gap between slow performances and real-time architectures. Our system has been validated by its deployment on a Care-O-bot 3 robot used during RoboCup@Home competitions. All our code, demos and pre-trained architectures have been released under an open-source license in our public repository.

Tobias Groot, Matias Valdenegro-Toro

Language and Vision-Language Models (LLMs/VLMs) have revolutionized the field of AI by their ability to generate human-like text and understand images, but ensuring their reliability is crucial. This paper aims to evaluate the ability of LLMs (GPT4, GPT-3.5, LLaMA2, and PaLM 2) and VLMs (GPT4V and Gemini Pro Vision) to estimate their verbalized uncertainty via prompting. We propose the new Japanese Uncertain Scenes (JUS) dataset, aimed at testing VLM capabilities via difficult queries and object counting, and the Net Calibration Error (NCE) to measure direction of miscalibration. Results show that both LLMs and VLMs have a high calibration error and are overconfident most of the time, indicating a poor capability for uncertainty estimation. Additionally we develop prompts for regression tasks, and we show that VLMs have poor calibration when producing mean/standard deviation and 95% confidence intervals.

Mirko Borszukovszki, Ivo Pascal de Jong, Matias Valdenegro-Toro

To leverage the full potential of Large Language Models (LLMs) it is crucial to have some information on their answers' uncertainty. This means that the model has to be able to quantify how certain it is in the correctness of a given response. Bad uncertainty estimates can lead to overconfident wrong answers undermining trust in these models. Quite a lot of research has been done on language models that work with text inputs and provide text outputs. Still, since the visual capabilities have been added to these models recently, there has not been much progress on the uncertainty of Visual Language Models (VLMs). We tested three state-of-the-art VLMs on corrupted image data. We found that the severity of the corruption negatively impacted the models' ability to estimate their uncertainty and the models also showed overconfidence in most of the experiments.

Ahmed Faisal Abdelrahman, Matias Valdenegro-Toro, Maren Bennewitz et al.

Neuromorphic computing mimics computational principles of the brain in $\textit{silico}$ and motivates research into event-based vision and spiking neural networks (SNNs). Event cameras (ECs) exclusively capture local intensity changes and offer superior power consumption, response latencies, and dynamic ranges. SNNs replicate biological neuronal dynamics and have demonstrated potential as alternatives to conventional artificial neural networks (ANNs), such as in reducing energy expenditure and inference time in visual classification. Nevertheless, these novel paradigms remain scarcely explored outside the domain of aerial robots. To investigate the utility of brain-inspired sensing and data processing, we developed a neuromorphic approach to obstacle avoidance on a camera-equipped manipulator. Our approach adapts high-level trajectory plans with reactive maneuvers by processing emulated event data in a convolutional SNN, decoding neural activations into avoidance motions, and adjusting plans using a dynamic motion primitive. We conducted experiments with a Kinova Gen3 arm performing simple reaching tasks that involve obstacles in sets of distinct task scenarios and in comparison to a non-adaptive baseline. Our neuromorphic approach facilitated reliable avoidance of imminent collisions in simulated and real-world experiments, where the baseline consistently failed. Trajectory adaptations had low impacts on safety and predictability criteria. Among the notable SNN properties were the correlation of computations with the magnitude of perceived motions and a robustness to different event emulation methods. Tests with a DAVIS346 EC showed similar performance, validating our experimental event emulation. Our results motivate incorporating SNN learning, utilizing neuromorphic processors, and further exploring the potential of neuromorphic methods.

Maniraj Sai Adapa, Marco Zullich, Matias Valdenegro-Toro

Deep Learning-based image super-resolution (SR) has been gaining traction with the aid of Generative Adversarial Networks. Models like SRGAN and ESRGAN are constantly ranked between the best image SR tools. However, they lack principled ways for estimating predictive uncertainty. In the present work, we enhance these models using Monte Carlo-Dropout and Deep Ensemble, allowing the computation of predictive uncertainty. When coupled with a prediction, uncertainty estimates can provide more information to the model users, highlighting pixels where the SR output might be uncertain, hence potentially inaccurate, if these estimates were to be reliable. Our findings suggest that these uncertainty estimates are decently calibrated and can hence fulfill this goal, while providing no performance drop with respect to the corresponding models without uncertainty estimation.

Ömer Tarik Özyilmaz, Matt Coler, Matias Valdenegro-Toro

Although commercial Arabic automatic speech recognition (ASR) systems support Modern Standard Arabic (MSA), they struggle with dialectal speech. We investigate the effect of fine-tuning OpenAI's Whisper on five major Arabic dialects (Gulf, Levantine, Iraqi, Egyptian, Maghrebi) using Mozilla Common Voice for MSA and the MASC dataset for dialectal speech. We evaluate MSA training size effects, benefits of pre-training on MSA data, and dialect-specific versus dialect-pooled models. We find that small amounts of MSA fine-tuning data yield substantial improvements for smaller models, matching larger non-fine-tuned models. While MSA pre-training shows minimal benefit, suggesting limited shared features between MSA and dialects, our dialect-pooled models perform comparably to dialect-specific ones. This indicates that pooling dialectal data, when properly balanced, can help address data scarcity in low-resource ASR without significant performance loss.

Josh Bruegger, Diana Ioana Catana, Vanja Macovaz et al.

The attribution of the author of an art piece is typically a laborious manual process, usually relying on subjective evaluations of expert figures. However, there are some situations in which quantitative features of the artwork can support these evaluations. The extraction of these features can sometimes be automated, for instance, with the use of Machine Learning (ML) techniques. An example of these features is represented by repeated, mechanically impressed patterns, called punches, present chiefly in 13th and 14th-century panel paintings from Tuscany. Previous research in art history showcased a strong connection between the shapes of punches and specific artists or workshops, suggesting the possibility of using these quantitative cues to support the attribution. In the present work, we first collect a dataset of large-scale images of these panel paintings. Then, using YOLOv10, a recent and popular object detection model, we train a ML pipeline to perform object detection on the punches contained in the images. Due to the large size of the images, the detection procedure is split across multiple frames by adopting a sliding-window approach with overlaps, after which the predictions are combined for the whole image using a custom non-maximal suppression routine. Our results indicate how art historians working in the field can reliably use our method for the identification and extraction of punches.

Miro Miranda, Deepak Pathak, Patrick Helber et al.

Crop yield prediction requires substantial data to train scalable models. However, creating yield prediction datasets is constrained by high acquisition costs, heterogeneous data quality, and data privacy regulations. Consequently, existing datasets are scarce, low in quality, or limited to regional levels or single crop types, hindering the development of scalable data-driven solutions. In this work, we release YieldSAT, a large, high-quality, and multimodal dataset for high-resolution crop yield prediction. YieldSAT spans various climate zones across multiple countries, including Argentina, Brazil, Uruguay, and Germany, and includes major crop types, including corn, rapeseed, soybeans, and wheat, across 2,173 expert-curated fields. In total, over 12.2 million yield samples are available, each with a spatial resolution of 10 m. Each field is paired with multispectral satellite imagery, resulting in 113,555 labeled satellite images, complemented by auxiliary environmental data. We demonstrate the potential of large-scale and high-resolution crop yield prediction as a pixel regression task by comparing various deep learning models and data fusion architectures. Furthermore, we highlight open challenges arising from severe distribution shifts in the ground truth data under real-world conditions. To mitigate this, we explore a domain-informed Deep Ensemble approach that exhibits significant performance gains. The dataset is available at https://yieldsat.github.io/.

Stefania Radu, Marco Zullich, Matias Valdenegro-Toro

Pixel-based language models aim to solve the vocabulary bottleneck problem in language modeling, but the challenge of uncertainty quantification remains open. The novelty of this work consists of analysing uncertainty and confidence in pixel-based language models across 18 languages and 7 scripts, all part of 3 semantically challenging tasks. This is achieved through several methods such as Monte Carlo Dropout, Transformer Attention, and Ensemble Learning. The results suggest that pixel-based models underestimate uncertainty when reconstructing patches. The uncertainty is also influenced by the script, with Latin languages displaying lower uncertainty. The findings on ensemble learning show better performance when applying hyperparameter tuning during the named entity recognition and question-answering tasks across 16 languages.

Iris Kamsteeg, Juan Cardenas-Cartagena, Floris van Beers et al.

This research aims to explore the intersection of Large Language Models and confidence calibration in Entity Matching. To this end, we perform an empirical study to compare baseline RoBERTa confidences for an Entity Matching task against confidences that are calibrated using Temperature Scaling, Monte Carlo Dropout and Ensembles. We use the Abt-Buy, DBLP-ACM, iTunes-Amazon and Company datasets. The findings indicate that the proposed modified RoBERTa model exhibits a slight overconfidence, with Expected Calibration Error scores ranging from 0.0043 to 0.0552 across datasets. We find that this overconfidence can be mitigated using Temperature Scaling, reducing Expected Calibration Error scores by up to 23.83%.

Leonidas Zotos, Ivo Pascal de Jong, Matias Valdenegro-Toro et al.

Estimating the difficulty of exam questions is essential for developing good exams, but professors are not always good at this task. We compare various Large Language Model-based methods with three professors in their ability to estimate what percentage of students will give correct answers on True/False exam questions in the areas of Neural Networks and Machine Learning. Our results show that the professors have limited ability to distinguish between easy and difficult questions and that they are outperformed by directly asking Gemini 2.5 to solve this task. Yet, we obtained even better results using uncertainties of the LLMs solving the questions in a supervised learning setting, using only 42 training samples. We conclude that supervised learning using LLM uncertainty can help professors better estimate the difficulty of exam questions, improving the quality of assessment.

Joris Suurmeijer, Ivo Pascal de Jong, Matias Valdenegro-Toro et al.

Brain-computer interfaces (BCIs) turn brain signals into functionally useful output, but they are not always accurate. A good Machine Learning classifier should be able to indicate how confident it is about a given classification, by giving a probability for its classification. Standard classifiers for Motor Imagery BCIs do give such probabilities, but research on uncertainty quantification has been limited to Deep Learning. We compare the uncertainty quantification ability of established BCI classifiers using Common Spatial Patterns (CSP-LDA) and Riemannian Geometry (MDRM) to specialized methods in Deep Learning (Deep Ensembles and Direct Uncertainty Quantification) as well as standard Convolutional Neural Networks (CNNs). We found that the overconfidence typically seen in Deep Learning is not a problem in CSP-LDA and MDRM. We found that MDRM is underconfident, which we solved by adding Temperature Scaling (MDRM-T). CSP-LDA and MDRM-T give the best uncertainty estimates, but Deep Ensembles and standard CNNs give the best classifications. We show that all models are able to separate between easy and difficult estimates, so that we can increase the accuracy of a Motor Imagery BCI by rejecting samples that are ambiguous.

Michal Tešnar, Bilal Wehbe, Matias Valdenegro-Toro

Machine learning proves effective in constructing dynamics models from data, especially for underwater vehicles. Continuous refinement of these models using incoming data streams, however, often requires storage of an overwhelming amount of redundant data. This work investigates the use of uncertainty in the selection of data points to rehearse in online learning when storage capacity is constrained. The models are learned using an ensemble of multilayer perceptrons as they perform well at predicting epistemic uncertainty. We present three novel approaches: the Threshold method, which excludes samples with uncertainty below a specified threshold, the Greedy method, designed to maximize uncertainty among the stored points, and Threshold-Greedy, which combines the previous two approaches. The methods are assessed on data collected by an underwater vehicle Dagon. Comparison with baselines reveals that the Threshold exhibits enhanced stability throughout the learning process and also yields a model with the least cumulative testing loss. We also conducted detailed analyses on the impact of model parameters and storage size on the performance of the models, as well as a comparison of three different uncertainty estimation methods.

Matias Valdenegro-Toro, Deepan Chakravarthi Padmanabhan, Deepak Singh et al.

Sonar sensing is fundamental for underwater robotics, but limited by capabilities of AI systems, which need large training datasets. Public data in sonar modalities is lacking. This paper presents the Marine Debris Forward-Looking Sonar datasets, with three different settings (watertank, turntable, flooded quarry) increasing dataset diversity and multiple computer vision tasks: object classification, object detection, semantic segmentation, patch matching, and unsupervised learning. We provide full dataset description, basic analysis and initial results for some tasks. We expect the research community will benefit from this dataset, which is publicly available at https://doi.org/10.5281/zenodo.15101686

Matias Valdenegro-Toro, Marco Zullich

Inputs to machine learning models can have associated noise or uncertainties, but they are often ignored and not modelled. It is unknown if Bayesian Neural Networks and their approximations are able to consider uncertainty in their inputs. In this paper we build a two input Bayesian Neural Network (mean and standard deviation) and evaluate its capabilities for input uncertainty estimation across different methods like Ensembles, MC-Dropout, and Flipout. Our results indicate that only some uncertainty estimation methods for approximate Bayesian NNs can model input uncertainty, in particular Ensembles and Flipout.

Eric Brouwer, Jan Erik van Woerden, Gertjan Burghouts et al.

Few-shot, fine-grained classification in computer vision poses significant challenges due to the need to differentiate subtle class distinctions with limited data. This paper presents a novel method that enhances the Contrastive Language-Image Pre-Training (CLIP) model through adaptive prompt tuning, guided by real-time visual inputs. Unlike existing techniques such as Context Optimization (CoOp) and Visual Prompt Tuning (VPT), which are constrained by static prompts or visual token reliance, the proposed approach leverages a cross-attention mechanism to dynamically refine text prompts for the image at hand. This enables an image-specific alignment of textual features with image patches extracted from the Vision Transformer, making the model more effective for datasets with high intra-class variance and low inter-class differences. The method is evaluated on several datasets, including CUBirds, Oxford Flowers, and FGVC Aircraft, showing significant performance gains over static prompt tuning approaches. To ensure these performance gains translate into trustworthy predictions, we integrate Monte-Carlo Dropout in our approach to improve the reliability of the model predictions and uncertainty estimates. This integration provides valuable insights into the model's predictive confidence, helping to identify when predictions can be trusted and when additional verification is necessary. This dynamic approach offers a robust solution, advancing the state-of-the-art for few-shot fine-grained classification.

Matias Valdenegro-Toro, Ivo Pascal de Jong, Marco Zullich

Modelling uncertainty in Machine Learning models is essential for achieving safe and reliable predictions. Most research on uncertainty focuses on output uncertainty (predictions), but minimal attention is paid to uncertainty at inputs. We propose a method for propagating uncertainty in the inputs through a Neural Network that is simultaneously able to estimate input, data, and model uncertainty. Our results show that this propagation of input uncertainty results in a more stable decision boundary even under large amounts of input noise than comparatively simple Monte Carlo sampling. Additionally, we discuss and demonstrate that input uncertainty, when propagated through the model, results in model uncertainty at the outputs. The explicit incorporation of input uncertainty may be beneficial in situations where the amount of input uncertainty is known, though good datasets for this are still needed.

Mihir Mulye, Matias Valdenegro-Toro

Explanation methods help understand the reasons for a model's prediction. These methods are increasingly involved in model debugging, performance optimization, and gaining insights into the workings of a model. With such critical applications of these methods, it is imperative to measure the uncertainty associated with the explanations generated by these methods. In this paper, we propose a pipeline to ascertain the explanation uncertainty of neural networks by combining uncertainty estimation methods and explanation methods. We use this pipeline to produce explanation distributions for the CIFAR-10, FER+, and California Housing datasets. By computing the coefficient of variation of these distributions, we evaluate the confidence in the explanation and determine that the explanations generated using Guided Backpropagation have low uncertainty associated with them. Additionally, we compute modified pixel insertion/deletion metrics to evaluate the quality of the generated explanations.

Matias Valdenegro-Toro, Mihir Mulye

Explanations for machine learning models can be hard to interpret or be wrong. Combining an explanation method with an uncertainty estimation method produces explanation uncertainty. Evaluating explanation uncertainty is difficult. In this paper we propose sanity checks for uncertainty explanation methods, where a weight and data randomization tests are defined for explanations with uncertainty, allowing for quick tests to combinations of uncertainty and explanation methods. We experimentally show the validity and effectiveness of these tests on the CIFAR10 and California Housing datasets, noting that Ensembles seem to consistently pass both tests with Guided Backpropagation, Integrated Gradients, and LIME explanations.

Prithviraj Manivannan, Ivo Pascal de Jong, Matias Valdenegro-Toro et al.

Uncertainty Quantification aims to determine when the prediction from a Machine Learning model is likely to be wrong. Computer Vision research has explored methods for determining epistemic uncertainty (also known as model uncertainty), which should correspond with generalisation error. These methods theoretically allow to predict misclassifications due to inter-subject variability. We applied a variety of Uncertainty Quantification methods to predict misclassifications for a Motor Imagery Brain Computer Interface. Deep Ensembles performed best, both in terms of classification performance and cross-subject Uncertainty Quantification performance. However, we found that standard CNNs with Softmax output performed better than some of the more advanced methods.

Ivo Pascal de Jong, Lüke Luna van den Wittenboer, Matias Valdenegro-Toro et al.

Public Motor Imagery-based brain-computer interface (BCI) datasets are being used to develop increasingly good classifiers. However, they usually follow discrete paradigms where participants perform Motor Imagery at regularly timed intervals. It is often unclear what changes may happen in the EEG patterns when users attempt to perform a control task with such a BCI. This may lead to generalisation errors. We demonstrate a new paradigm containing a standard calibration session and a novel BCI control session based on EMG. This allows us to observe similarities in sensorimotor rhythms, and observe the additional preparation effects introduced by the control paradigm. In the Movement Related Cortical Potentials we found large differences between the calibration and control sessions. We demonstrate a CSP-based Machine Learning model trained on the calibration data that can make surprisingly good predictions on the BCI-controlled driving data.

Tsegaye Misikir Tashu, Eduard-Raul Kontos, Matthia Sabatelli et al.

Recommendation systems, for documents, have become tools to find relevant content on the Web. However, these systems have limitations when it comes to recommending documents in languages different from the query language, which means they might overlook resources in non-native languages. This research focuses on representing documents across languages by using Transformer Leveraged Document Representations (TLDRs) that are mapped to a cross-lingual domain. Four multilingual pre-trained transformer models (mBERT, mT5 XLM RoBERTa, ErnieM) were evaluated using three mapping methods across 20 language pairs representing combinations of five selected languages of the European Union. Metrics like Mate Retrieval Rate and Reciprocal Rank were used to measure the effectiveness of mapped TLDRs compared to non-mapped ones. The results highlight the power of cross-lingual representations achieved through pre-trained transformers and mapping approaches suggesting a promising direction for expanding beyond language connections, between two specific languages.

Matias Valdenegro-Toro, Daniel Harnack, Hendrik Wöhrle

Modeling trajectories generated by robot joints is complex and required for high level activities like trajectory generation, clustering, and classification. Disentagled representation learning promises advances in unsupervised learning, but they have not been evaluated in robot-generated trajectories. In this paper we evaluate three disentangling VAEs ($β$-VAE, Decorr VAE, and a new $β$-Decorr VAE) on a dataset of 1M robot trajectories generated from a 3 DoF robot arm. We find that the decorrelation-based formulations perform the best in terms of disentangling metrics, trajectory quality, and correlation with ground truth latent features. We expect that these results increase the use of unsupervised learning in robot control.

Aaqib Parvez Mohammed, Matias Valdenegro-Toro

Reinforcement Learning (RL) based solutions are being adopted in a variety of domains including robotics, health care and industrial automation. Most focus is given to when these solutions work well, but they fail when presented with out of distribution inputs. RL policies share the same faults as most machine learning models. Out of distribution detection for RL is generally not well covered in the literature, and there is a lack of benchmarks for this task. In this work we propose a benchmark to evaluate OOD detection methods in a Reinforcement Learning setting, by modifying the physical parameters of non-visual standard environments or corrupting the state observation for visual environments. We discuss ways to generate custom RL environments that can produce OOD data, and evaluate three uncertainty methods for the OOD detection task. Our results show that ensemble methods have the best OOD detection performance with a lower standard deviation across multiple environments.

Matias Valdenegro-Toro

Uncertainty quantification in neural network promises to increase safety of AI systems, but it is not clear how performance might vary with the training set size. In this paper we evaluate seven uncertainty methods on Fashion MNIST and CIFAR10, as we sub-sample and produce varied training set sizes. We find that calibration error and out of distribution detection performance strongly depend on the training set size, with most methods being miscalibrated on the test set with small training sets. Gradient-based methods seem to poorly estimate epistemic uncertainty and are the most affected by training set size. We expect our results can guide future research into uncertainty quantification and help practitioners select methods based on their particular available data.

Adrian Lubitz, Matias Valdenegro-Toro, Frank Kirchner

Robots are becoming everyday devices, increasing their interaction with humans. To make human-machine interaction more natural, cognitive features like Visual Voice Activity Detection (VVAD), which can detect whether a person is speaking or not, given visual input of a camera, need to be implemented. Neural networks are state of the art for tasks in Image Processing, Time Series Prediction, Natural Language Processing and other domains. Those Networks require large quantities of labeled data. Currently there are not many datasets for the task of VVAD. In this work we created a large scale dataset called the VVAD-LRS3 dataset, derived by automatic annotations from the LRS3 dataset. The VVAD-LRS3 dataset contains over 44K samples, over three times the next competitive dataset (WildVVAD). We evaluate different baselines on four kinds of features: facial and lip images, and facial and lip landmark features. With a Convolutional Neural Network Long Short Term Memory (CNN LSTM) on facial images an accuracy of 92% was reached on the test set. A study with humans showed that they reach an accuracy of 87.93% on the test set.

Matias Valdenegro-Toro

Uncertainty in machine learning is not generally taught as general knowledge in Machine Learning course curricula. In this paper we propose a short curriculum for a course about uncertainty in machine learning, and complement the course with a selection of use cases, aimed to trigger discussion and let students play with the concepts of uncertainty in a programming setting. Our use cases cover the concept of output uncertainty, Bayesian neural networks and weight distributions, sources of uncertainty, and out of distribution detection. We expect that this curriculum and set of use cases motivates the community to adopt these important concepts into courses for safety in AI.

Mihir Patil, Bilal Wehbe, Matias Valdenegro-Toro

Docking control of an autonomous underwater vehicle (AUV) is a task that is integral to achieving persistent long term autonomy. This work explores the application of state-of-the-art model-free deep reinforcement learning (DRL) approaches to the task of AUV docking in the continuous domain. We provide a detailed formulation of the reward function, utilized to successfully dock the AUV onto a fixed docking platform. A major contribution that distinguishes our work from the previous approaches is the usage of a physics simulator to define and simulate the underwater environment as well as the DeepLeng AUV. We propose a new reward function formulation for the docking task, incorporating several components, that outperforms previous reward formulations. We evaluate proximal policy optimization (PPO), twin delayed deep deterministic policy gradients (TD3) and soft actor-critic (SAC) in combination with our reward function. Our evaluation yielded results that conclusively show the TD3 agent to be most efficient and consistent in terms of docking the AUV, over multiple evaluation runs it achieved a 100% success rate and episode return of 10667.1 +- 688.8. We also show how our reward function formulation improves over the state of the art.

Matias Valdenegro-Toro, Alan Preciado-Grijalva, Bilal Wehbe

Machine learning and neural networks are now ubiquitous in sonar perception, but it lags behind the computer vision field due to the lack of data and pre-trained models specifically for sonar images. In this paper we present the Marine Debris Turntable dataset and produce pre-trained neural networks trained on this dataset, meant to fill the gap of missing pre-trained models for sonar images. We train Resnet 20, MobileNets, DenseNet121, SqueezeNet, MiniXception, and an Autoencoder, over several input image sizes, from 32 x 32 to 96 x 96, on the Marine Debris turntable dataset. We evaluate these models using transfer learning for low-shot classification in the Marine Debris Watertank and another dataset captured using a Gemini 720i sonar. Our results show that in both datasets the pre-trained models produce good features that allow good classification accuracy with low samples (10-30 samples per class). The Gemini dataset validates that the features transfer to other kinds of sonar sensors. We expect that the community benefits from the public release of our pre-trained models and the turntable dataset.

Arka Mallick, Paul Plöger, Matias Valdenegro-Toro

Application of underwater robots are on the rise, most of them are dependent on sonar for underwater vision, but the lack of strong perception capabilities limits them in this task. An important issue in sonar perception is matching image patches, which can enable other techniques like localization, change detection, and mapping. There is a rich literature for this problem in color images, but for acoustic images, it is lacking, due to the physics that produce these images. In this paper we improve on our previous results for this problem (Valdenegro-Toro et al, 2017), instead of modeling features manually, a Convolutional Neural Network (CNN) learns a similarity function and predicts if two input sonar images are similar or not. With the objective of improving the sonar image matching problem further, three state of the art CNN architectures are evaluated on the Marine Debris dataset, namely DenseNet, and VGG, with a siamese or two-channel architecture, and contrastive loss. To ensure a fair evaluation of each network, thorough hyper-parameter optimization is executed. We find that the best performing models are DenseNet Two-Channel network with 0.955 AUC, VGG-Siamese with contrastive loss at 0.949 AUC and DenseNet Siamese with 0.921 AUC. By ensembling the top performing DenseNet two-channel and DenseNet-Siamese models overall highest prediction accuracy obtained is 0.978 AUC, showing a large improvement over the 0.91 AUC in the state of the art.

Matias Valdenegro-Toro

Neural networks are used for many real world applications, but often they have problems estimating their own confidence. This is particularly problematic for computer vision applications aimed at making high stakes decisions with humans and their lives. In this paper we make a meta-analysis of the literature, showing that most if not all computer vision applications do not use proper epistemic uncertainty quantification, which means that these models ignore their own limitations. We describe the consequences of using models without proper uncertainty quantification, and motivate the community to adopt versions of the models they use that have proper calibrated epistemic uncertainty, in order to enable out of distribution detection. We close the paper with a summary of challenges on estimating uncertainty for computer vision applications and recommendations.

Matthias Rosynski, Frank Kirchner, Matias Valdenegro-Toro

Deep Reinforcement Learning (DRL) connects the classic Reinforcement Learning algorithms with Deep Neural Networks. A problem in DRL is that CNNs are black-boxes and it is hard to understand the decision-making process of agents. In order to be able to use RL agents in highly dangerous environments for humans and machines, the developer needs a debugging tool to assure that the agent does what is expected. Currently, rewards are primarily used to interpret how well an agent is learning. However, this can lead to deceptive conclusions if the agent receives more rewards by memorizing a policy and not learning to respond to the environment. In this work, it is shown that this problem can be recognized with the help of gradient visualization techniques. This work brings some of the best-known visualization methods from the field of image classification to the area of Deep Reinforcement Learning. Furthermore, two new visualization techniques have been developed, one of which provides particularly good results. It is being proven to what extent the algorithms can be used in the area of Reinforcement learning. Also, the question arises on how well the DRL algorithms can be visualized across different environments with varying visualization techniques.

Octavio Arriaga, Matias Valdenegro-Toro

Evaluating difficulty and biases in machine learning models has become of extreme importance as current models are now being applied in real-world situations. In this paper we present a simple method for calculating a difficulty score based on the accumulation of losses for each sample during training. We call this the action score. Our proposed method does not require any modification of the model neither any external supervision, as it can be implemented as callback that gathers information from the training process. We test and analyze our approach in two different settings: image classification, and object detection, and we show that in both settings the action score can provide insights about model and dataset biases.

Lauren Michelle Pfeifer, Matias Valdenegro-Toro

Around the globe, ticks are the culprit of transmitting a variety of bacterial, viral and parasitic diseases. The incidence of tick-borne diseases has drastically increased within the last decade, with annual cases of Lyme disease soaring to an estimated 300,000 in the United States alone. As a result, more efforts in improving lesion identification approaches and diagnostics for tick-borne illnesses is critical. The objective for this study is to build upon the approach used by Burlina et al. by using a variety of convolutional neural network models to detect tick-borne skin lesions. We expanded the data inputs by acquiring images from Google in seven different languages to test if this would diversify training data and improve the accuracy of skin lesion detection. The final dataset included nearly 6,080 images and was trained on a combination of architectures (ResNet 34, ResNet 50, VGG 19, and Dense Net 121). We obtained an accuracy of 80.72% with our model trained on the DenseNet 121 architecture.

Mohandass Muthuraja, Octavio Arriaga, Paul Plöger et al.

Object detectors have improved considerably in the last years by using advanced CNN architectures. However, many detector hyper-parameters are generally manually tuned, or they are used with values set by the detector authors. Automatic Hyper-parameter optimization has not been explored in improving CNN-based object detectors hyper-parameters. In this work, we propose the use of Black-box optimization methods to tune the prior/default box scales in Faster R-CNN and SSD, using Bayesian Optimization, SMAC, and CMA-ES. We show that by tuning the input image size and prior box anchor scale on Faster R-CNN mAP increases by 2% on PASCAL VOC 2007, and by 3% with SSD. On the COCO dataset with SSD there are mAP improvement in the medium and large objects, but mAP decreases by 1% in small objects. We also perform a regression analysis to find the significant hyper-parameters to tune.