Diverse Projection Ensembles for Distributional Reinforcement Learning

Moritz A. Zanger, et al. The Twelfth International Conference on Learning Representations (ICLR 2024). May 07, 2024. [PDF]


In contrast to classical reinforcement learning (RL), distributional RL algorithms aim to learn the distribution of returns rather than their expected value. Since the nature of the return distribution is generally unknown a priori or arbitrarily complex, a common approach finds approximations within a set of representable, parametric distributions. Typically, this involves a projection of the unconstrained distribution onto the set of simplified distributions. We argue that this projection step entails a strong inductive bias when coupled with neural networks and gradient descent, thereby profoundly impacting the generalization behavior of learned models. In order to facilitate reliable uncertainty estimation through diversity, we study the combination of several different projections and representations in a distributional ensemble. We establish theoretical properties of such projection ensembles and derive an algorithm that uses ensemble disagreement, measured by the average 1-Wasserstein distance, as a bonus for deep exploration. We evaluate our algorithm on the behavior suite benchmark and VizDoom and find that diverse projection ensembles lead to significant performance improvements over existing methods on a variety of tasks with the most pronounced gains in directed exploration problems.

On the Estimation of Image-matching Uncertainty in Visual Place Recognition

Mubariz Zaffar, Liangliang Nan and Julian F. P. Kooij. Accepted for 2024 Conference on Computer Vision and Pattern Recognition. Mar 31, 2024. [PDF]


In Visual Place Recognition (VPR) the pose of a query image is estimated by comparing the image to a map of reference images with known reference poses. As is typical for image retrieval problems, a feature extractor maps the query and reference images to a feature space, where a nearest neighbor search is then performed. However, till recently little attention has been given to quantifying the confidence that a retrieved reference image is a correct match. Highly certain but incorrect retrieval can lead to catastrophic failure of VPR-based localization pipelines. This work compares for the first time the main approaches for estimating the image-matching uncertainty, including the traditional retrieval-based uncertainty estimation, more recent data-driven aleatoric uncertainty estimation, and the compute-intensive geometric verification. We further formulate a simple baseline method, ''SUE'', which unlike the other methods considers the freely-available poses of the reference images in the map. Our experiments reveal that a simple L2-distance between the query and reference descriptors is already a better estimate of image-matching uncertainty than current data-driven approaches. SUE outperforms the other efficient uncertainty estimation methods, and its uncertainty estimates complement the computationally expensive geometric verification approach. Future works for uncertainty estimation in VPR should consider the baselines discussed in this work.

Interval Reduced Order Surrogate Modelling Framework for Uncertainty Quantification

Ghifari A. Faza, Keivan Shariatmadar, Hans Hallez and David Moens. AIAA Scitech 2024 Forum. Jan 4, 2024. [PDF]


Surrogate models are widely used in the engineering community to approximate costly and large evaluation processes, such as difficult experiments or expensive simulations. This paper presents a non-intrusive framework for epistemic surrogate modelling, which is based on proper orthogonal decomposition (POD) and polynomial chaos expansion (PCE) for interval observations. In physical systems modelling, it is important to consider both aleatoric and epistemic uncertainty by constructing an uncertainty model to be included in the surrogate model. However, existing frameworks have a major limitation in that they can only handle scalar data observations and are not designed for non-deterministic observations like interval data. In many applications, the observed data can be inherently non-scalar due to various factors, such as observation uncertainties, conflicting data, or summarized data. In our proposed framework, we integrate POD for interval data with PCE for interval observations. Firstly, we employ interval POD to obtain an optimally reduced-order basis from the full-order snapshot. Then, we approximate this reduced-order basis using a non-intrusive interval PCE method. Allowing non-scalar data, such as intervals, is advantageous as it takes into account more information in the physical system modelling.


Random-Set Convolutional Neural Network (RS-CNN) for Epistemic Deep Learning

Shireen Kudukkil Manchingal et al. Preprint. Jul 11, 2023. [PDF]


Machine learning is increasingly deployed in safety-critical domains where robustness against adversarial attacks is crucial and erroneous predictions could lead to potentially catastrophic consequences. This highlights the need for learning systems to be equipped with the means to determine a model's confidence in its prediction and the epistemic uncertainty associated with it, 'to know when a model does not know'. In this paper, we propose a novel Random-Set Convolutional Neural Network (RS-CNN) for classification which predicts belief functions rather than probability vectors over the set of classes, using the mathematics of random sets, i.e., distributions over the power set of the sample space. Based on the epistemic deep learning approach, random-set models are capable of representing the 'epistemic' uncertainty induced in machine learning by limited training sets. We estimate epistemic uncertainty by approximating the size of credal sets associated with the predicted belief functions, and experimentally demonstrate how our approach outperforms competing uncertainty-aware approaches in a classical evaluation setting. The performance of RS-CNN is best demonstrated on OOD samples where it manages to capture the true prediction while standard CNNs fail.

E-MCTS: Deep Exploration in Model-Based Reinforcement Learning by Planning with Epistemic Uncertainty

Yaniv Oren et al. 16th European Workshop on Reinforcement Learning. 2023. Aug 30, 2023. [PDF]


One of the most well-studied and highly performing planning approaches used in Model-Based Reinforcement Learning (MBRL) is Monte-Carlo Tree Search (MCTS). Key challenges of MCTS-based MBRL methods remain dedicated deep exploration and reliability in the face of the unknown, and both challenges can be alleviated through principled epistemic uncertainty estimation in the predictions of MCTS. We present two main contributions: First, we develop methodology to propagate epistemic uncertainty in MCTS, enabling agents to estimate the epistemic uncertainty in their predictions. Second, we utilize the propagated uncertainty for a novel deep exploration algorithm by explicitly planning to explore. We incorporate our approach into variations of MCTS-based MBRL approaches with learned and provided models, and empirically show deep exploration through successful epistemic uncertainty estimation achieved by our approach. We compare to a non-planning-based deep-exploration baseline, and demonstrate that planning with epistemic MCTS significantly outperforms non-planning based exploration in the investigated setting.

ROAD-R: the autonomous driving dataset with logical requirements

Eleonora Giunchiglia et al. Machine Learning. May 1, 2023. [PDF]


Neural networks have proven to be very powerful at computer vision tasks. However, they often exhibit unexpected behaviors, acting against background knowledge about the problem at hand. This calls for models (i) able to learn from requirements expressing such background knowledge, and (ii) guaranteed to be compliant with the requirements themselves. Unfortunately, the development of such models is hampered by the lack of real-world datasets equipped with formally specified requirements. In this paper, we introduce the ROad event Awareness Dataset with logical Requirements (ROAD-R), the first publicly available dataset for autonomous driving with requirements expressed as logical constraints. Given ROAD-R, we show that current state-of-the-art models often violate its logical constraints, and that it is possible to exploit them to create models that (i) have a better performance, and (ii) are guaranteed to be compliant with the requirements themselves.


An Introduction to Optimization under Uncertainty--A Short Survey

Shariatmadar, Keivan, et al. Preprint. Dec 1, 2022. [PDF]


Optimization equips engineers and scientists in a variety of fields with the ability to transcribe their problems into a generic formulation and receive optimal solutions with relative ease. Industries ranging from aerospace to robotics continue to benefit from advancements in optimization theory and the associated algorithmic developments. Nowadays, optimization is used in real time on autonomous systems acting in safety-critical situations, such as self-driving vehicles. It has become increasingly more important to produce robust solutions by incorporating uncertainty into optimization programs. This paper provides a short survey about the state of the art in optimization under uncertainty. The paper begins with a brief overview of the main classes of optimization without uncertainty. The rest of the paper focuses on the different methods for handling both aleatoric and epistemic uncertainty. Many of the applications discussed in this paper are within the domain of control. The goal of this survey paper is to briefly touch upon the state of the art in a variety of different methods and refer the reader to other literature for more in-depth treatments of the topics discussed here.

Epistemic Deep Learning

Manchingal, Shireen Kudukkil and Cuzzolin, Fabio. Presented at ICML 2022 Workshop on Distribution-free Uncertainty Quantification. Jul 23, 2022. [PDF]


The belief function approach to uncertainty quantification as proposed in the Demspter-Shafer theory of evidence is established upon the general mathematical models for set-valued observations, called random sets. Set-valued predictions are the most natural representations of uncertainty in machine learning. In this paper, we introduce a concept called epistemic deep learning based on the random-set interpretation of belief functions to model epistemic learning in deep neural networks. We propose a novel random-set convolutional neural network for classification that produces scores for sets of classes by learning set-valued ground truth representations. We evaluate different formulations of entropy and distance measures for belief functions as viable loss functions for these random-set networks. We also discuss methods for evaluating the quality of epistemic predictions and the performance of epistemic random-set neural networks. We demonstrate through experiments that the epistemic approach produces better performance results when compared to traditional approaches of estimating uncertainty. 

Theory of Mind and Preference Learning at the Interface of Cognitive Science, Neuroscience, and AI: A Review

 Langley, Christelle, et al. Frontiers in Artificial Intelligence. Apr 05, 2022. [PDF]


Theory of Mind (ToM)—the ability of the human mind to attribute mental states to others—is a key component of human cognition. In order to understand other people's mental states or viewpoint and to have successful interactions with others within social and occupational environments, this form of social cognition is essential. The same capability of inferring human mental states is a prerequisite for artificial intelligence (AI) to be integrated into society, for example in healthcare and the motoring industry. Autonomous cars will need to be able to infer the mental states of human drivers and pedestrians to predict their behavior. In the literature, there has been an increasing understanding of ToM, specifically with increasing cognitive science studies in children and in individuals with Autism Spectrum Disorder. Similarly, with neuroimaging studies there is now a better understanding of the neural mechanisms that underlie ToM. In addition, new AI algorithms for inferring human mental states have been proposed with more complex applications and better generalisability. In this review, we synthesize the existing understanding of ToM in cognitive and neurosciences and the AI computational models that have been proposed. We focus on preference learning as an area of particular interest and the most recent neurocognitive and computational ToM models. We also discuss the limitations of existing models and hint at potential approaches to allow ToM models to fully express the complexity of the human mind in all its aspects, including values and preferences.

Road: The ROad event Awareness Dataset for autonomous Driving

 Singh, Gurkirt, et al. IEEE Transactions on Pattern Analysis and Machine Intelligence. Feb 14 2022. [PDF]


Humans drive in a holistic fashion which entails, in particular, understanding dynamic road events and their evolution. Injecting these capabilities in autonomous vehicles can thus take situational awareness and decision making closer to human-level performance. To this purpose, we introduce the ROad event Awareness Dataset (ROAD) for Autonomous Driving, to our knowledge the first of its kind. ROAD is designed to test an autonomous vehicles ability to detect road events, defined as triplets composed by an active agent, the action(s) it performs and the corresponding scene locations. ROAD comprises videos originally from the Oxford RobotCar Dataset annotated with bounding boxes showing the location in the image plane of each road event. We benchmark various detection tasks, proposing as a baseline a new incremental algorithm for online road event awareness termed 3D-RetinaNet. We also report the performance on the ROAD tasks of Slowfast and YOLOv5 detectors, as well as that of the winners of the ICCV2021 ROAD challenge, which highlight the challenges faced by situation awareness in autonomous driving. ROAD is designed to allow scholars to investigate exciting tasks such as complex (road) activity detection, future event anticipation and continual learning. The dataset is available at; the baseline can be found at

Vision in adverse weather: Augmentation using CycleGANs with various object detectors for robust perception in autonomous racing

Teeti, Izzeddin, et al. Not peer-viewed. Jan 11 2022. [PDF]


In an autonomous driving system, perception - identification of features and objects from the environment - is crucial. In autonomous racing, high speeds and small margins demand rapid and accurate detection systems. During the race, the weather can change abruptly, causing significant degradation in perception, resulting in ineffective manoeuvres. In order to improve detection in adverse weather, deep-learning-based models typically require extensive datasets captured in such conditions - the collection of which is a tedious, laborious, and costly process. However, recent developments in CycleGAN architectures allow the synthesis of highly realistic scenes in multiple weather conditions. To this end, we introduce an approach of using synthesised adverse condition datasets in autonomous racing (generated using CycleGAN) to improve the performance of four out of five state-of-the-art detectors by an average of 42.7 and 4.4 mAP percentage points in the presence of night-time conditions and droplets, respectively. Furthermore, we present a comparative analysis of five object detectors - identifying the optimal pairing of detector and training data for use during autonomous racing in challenging conditions.

The intersection probability: betting with probability intervals

Cuzzolin, Fabio. Not peer-viewed. Jan 5 2022. [PDF]


Probability intervals are an attractive tool for reasoning under uncertainty. Unlike belief functions, though, they lack a natural probability transformation to be used for decision making in a utility theory framework. In this paper we propose the use of the intersection probability, a transform derived originally for belief functions in the framework of the geometric approach to uncertainty, as the most natural such transformation. We recall its rationale and definition, compare it with other candidate representives of systems of probability intervals, discuss its credal rationale as focus of a pair of simplices in the probability simplex, and outline a possible decision making framework for probability intervals, analogous to the Transferable Belief Model for belief functions.


YOLO-Z: Improving small object detection in YOLOv5 for autonomous vehicles

Benjumea, Aduen, et al. ICCV Workshop: The ROAD challenge: Event Detection for Situation Awareness in Autonomous Driving. Dec 23 2021. [PDF]


As autonomous vehicles and autonomous racing rise in popularity, so does the need for faster and more accurate detectors. While our naked eyes are able to extract contextual information almost instantly, even from far away, image resolution and computational resources limitations make detecting smaller objects (that is, objects that occupy a small pixel area in the input image) a genuinely challenging task for machines and a wide-open research field. This study explores how the popular YOLOv5 object detector can be modified to improve its performance in detecting smaller objects, with a particular application in autonomous racing. To achieve this, we investigate how replacing certain structural elements of the model (as well as their connections and other parameters) can affect performance and inference time. In doing so, we propose a series of models at different scales, which we name `YOLO-Z', and which display an improvement of up to 6.9% in mAP when detecting smaller objects at 50% IOU, at the cost of just a 3ms increase in inference time compared to the original YOLOv5. Our objective is to inform future research on the potential of adjusting a popular detector such as YOLOv5 to address specific tasks and provide insights on how specific changes can impact small object detection. Such findings, applied to the broader context of autonomous vehicles, could increase the amount of contextual information available to such systems.

DeepSmoke: Deep Learning Model for Smoke Detection and Segmentation in Outdoor Environments 

Khan, Salman, et al. Expert Systems with Applications. Volume 182. Nov 15 2021. [Final Accepted Version] [Not Open Access]


Fire disaster throughout the globe causes social, environmental, and economical damage, making its early detection and instant reporting essential for saving human lives and properties. Smoke detection plays a key role in early fire detection but majority of the existing methods are limited to either indoor or outdoor surveillance environments, with poor performance for hazy scenarios. In this paper, we present a Convolutional Neural Network (CNN)-based smoke detection and segmentation framework for both clear and hazy environments. Unlike existing methods, we employ an efficient CNN architecture, termed EfficientNet, for smoke detection with better accuracy. We also segment the smoke regions using DeepLabv3+, which is supported by effective encoders and decoders along with a pixel-wise classifier for optimum localization. Our smoke detection results evince a noticeable gain up to 3% in accuracy and a decrease of 0.46% in False Alarm Rate (FAR), while segmentation reports a significant increase of 2% and 1% in global accuracy and mean Intersection over Union (IoU) scores, respectively. This makes our method a best fit for smoke detection and segmentation in real-world surveillance settings. 

Multi-weather city: Adverse weather stacking for autonomous driving

Mușat, Valentina, et al. 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). Oct 11-17 2021. [PDF]


Autonomous vehicles make use of sensors to perceive the world around them, with heavy reliance on vision based sensors such as RGB cameras. Unfortunately, since these sensors are affected by adverse weather, perception pipelines require extensive training on visual data under harsh conditions in order to improve the robustness of downstream tasks - data that is difficult and expensive to acquire. Based on GAN and CycleGAN architectures, we propose an overall (modular) architecture for constructing datasets, which allows one to add, swap out and combine components in order to generate images with diverse weather conditions. Starting from a single dataset with ground-truth, we generate 7 versions of the same data in diverse weather, and propose an extension to augment the generated conditions, thus resulting in a total of 14 adverse weather conditions, requiring a single ground truth. We test the quality of the generated conditions both in terms of perceptual quality and suitability for training downstream tasks, using real world, out-of-distribution adverse weather extracted from various datasets. We show improvements in both object detection and instance segmentation across all conditions, in many cases exceeding 10 percentage points increase in AP, and provide the materials and instructions needed to re-construct the multi-weather dataset, based upon the original Cityscapes dataset.

A geometric approach to conditioning belief functions

Cuzzolin, Fabio. Not peer-viewed. Apr 21 2021. [PDF]


Conditioning is crucial in applied science when inference involving time series is involved. Belief calculus is an effective way of handling such inference in the presence of epistemic uncertainty -- unfortunately, different approaches to conditioning in the belief function framework have been proposed in the past, leaving the matter somewhat unsettled. Inspired by the geometric approach to uncertainty, in this paper we propose an approach to the conditioning of belief functions based on geometrically projecting them onto the simplex associated with the conditioning event in the space of all belief functions. We show here that such a geometric approach to conditioning often produces simple results with straightforward interpretations in terms of degrees of belief. This raises the question of whether classical approaches, such as for instance Dempster's conditioning, can also be reduced to some form of distance minimisation in a suitable space. The study of families of combination rules generated by (geometric) conditioning rules appears to be the natural prosecution of the presented research.

Uncertainty measures: The big picture

Cuzzolin, Fabio. Not peer-viewed. Apr 14 2021. [PDF]


Probability theory is far from being the most general mathematical theory of uncertainty. A number of arguments point at its inability to describe second-order ('Knightian') uncertainty. In response, a wide array of theories of uncertainty have been proposed, many of them generalisations of classical probability. As we show here, such frameworks can be organised into clusters sharing a common rationale, exhibit complex links, and are characterised by different levels of generality. Our goal is a critical appraisal of the current landscape in uncertainty theory.

More info: ResearchGate.