Quasars experiencing strong lensing offer unique viewpoints on subjects related to the cosmic expansion rate, the dark matter profile within the foreground deflectors, and the quasar host galaxies. Unfortunately, identifying them in astronomical images is challenging since they are overwhelmed by the abundance of non-lenses. To address this, we have developed a novel approach by ensembling cutting-edge convolutional networks (CNNs) -- for instance, ResNet, Inception, NASNet, MobileNet, EfficientNet, and RegNet -- along with vision transformers (ViTs) trained on realistic galaxy-quasar lens simulations based on the Hyper Suprime-Cam (HSC) multiband images. While the individual model exhibits remarkable performance when evaluated against the test dataset, achieving an area under the receiver operating characteristic curve of $>$97.3% and a median false positive rate of 3.6%, it struggles to generalize in real data, indicated by numerous spurious sources picked by each classifier. A significant improvement is achieved by averaging these CNNs and ViTs, resulting in the impurities being downsized by factors up to 50. Subsequently, combining the HSC images with the UKIRT, VISTA, and unWISE data, we retrieve approximately 60 million sources as parent samples and reduce this to 892,609 after employing a photometry preselection to discover $z>1.5$ lensed quasars with Einstein radii of $\theta_\mathrm{E}<5$ arcsec. Afterward, the ensemble classifier indicates 3080 sources with a high probability of being lenses, for which we visually inspect, yielding 210 prevailing candidates awaiting spectroscopic confirmation. These outcomes suggest that automated deep learning pipelines hold great potential in effectively detecting strong lenses in vast datasets with minimal manual visual inspection involved.

In response to the COVID-19 pandemic and the potential threat of future epidemics caused by novel viruses, we developed a flexible framework for modeling disease intervention effects. This tool is intended to aid decision makers at multiple levels as they compare possible responses to emerging epidemiological threats for optimal control and reduction of harm. The framework is specifically designed to be both scalable and modular, allowing it to model a variety of population levels, viruses, testing methods and strategies--including pooled testing--and intervention strategies. In this paper, we provide an overview of this framework and examine the impact of different intervention strategies and their impact on infection dynamics.

Backdoor (Trojan) attacks are an important type of adversarial exploit against deep neural networks (DNNs), wherein a test instance is (mis)classified to the attacker's target class whenever the attacker's backdoor trigger is present. In this paper, we reveal and analyze an important property of backdoor attacks: a successful attack causes an alteration in the distribution of internal layer activations for backdoor-trigger instances, compared to that for clean instances. Even more importantly, we find that instances with the backdoor trigger will be correctly classified to their original source classes if this distribution alteration is corrected. Based on our observations, we propose an efficient and effective method that achieves post-training backdoor mitigation by correcting the distribution alteration using reverse-engineered triggers. Notably, our method does not change any trainable parameters of the DNN, but achieves generally better mitigation performance than existing methods that do require intensive DNN parameter tuning. It also efficiently detects test instances with the trigger, which may help to catch adversarial entities in the act of exploiting the backdoor.

Language models are useful adjuncts to optical models for producing accurate optical character recognition (OCR) results. One factor which limits the power of language models in this context is the existence of many specialized domains with language statistics very different from those implied by a general language model - think of checks, medical prescriptions, and many other specialized document classes. This paper introduces an algorithm for efficiently generating and attaching a domain specific word based language model at run time to a general language model in an OCR system. In order to best use this model the paper also introduces a modified CTC beam search decoder which effectively allows hypotheses to remain in contention based on possible future completion of vocabulary words. The result is a substantial reduction in word error rate in recognizing material from specialized domains.

Parallel robots (PRs) allow for higher speeds in human-robot collaboration due to their lower moving masses but are more prone to unintended contact. For a safe reaction, knowledge of the location and force of a collision is useful. A novel algorithm for collision isolation and identification with proprioceptive information for a real PR is the scope of this work. To classify the collided body, the effects of contact forces at the links and platform of the PR are analyzed using a kinetostatic projection. This insight enables the derivation of features from the line of action of the estimated external force. The significance of these features is confirmed in experiments for various load cases. A feedforward neural network (FNN) classifies the collided body based on these physically modeled features. Generalization with the FNN to 300k load cases on the whole robot structure in other joint angle configurations is successfully performed with a collision-body classification accuracy of 84% in the experiments. Platform collisions are isolated and identified with an explicit solution, while a particle filter estimates the location and force of a contact on a kinematic chain. Updating the particle filter with estimated external joint torques leads to an isolation error of less than 3cm and an identification error of 4N in a real-world experiment.

Graph neural networks (GNNs) are increasingly used in critical human applications for predicting node labels in attributed graphs. Their ability to aggregate features from nodes' neighbors for accurate classification also has the capacity to exacerbate existing biases in data or to introduce new ones towards members from protected demographic groups. Thus, it is imperative to quantify how GNNs may be biased and to what extent their harmful effects may be mitigated. To this end, we propose two new GNN-agnostic interventions namely, (i) PFR-AX which decreases the separability between nodes in protected and non-protected groups, and (ii) PostProcess which updates model predictions based on a blackbox policy to minimize differences between error rates across demographic groups. Through a large set of experiments on four datasets, we frame the efficacies of our approaches (and three variants) in terms of their algorithmic fairness-accuracy tradeoff and benchmark our results against three strong baseline interventions on three state-of-the-art GNN models. Our results show that no single intervention offers a universally optimal tradeoff, but PFR-AX and PostProcess provide granular control and improve model confidence when correctly predicting positive outcomes for nodes in protected groups.

This dissertation explores the impact of bias in deep neural networks and presents methods for reducing its influence on model performance. The first part begins by categorizing and describing potential sources of bias and errors in data and models, with a particular focus on bias in machine learning pipelines. The next chapter outlines a taxonomy and methods of Explainable AI as a way to justify predictions and control and improve the model. Then, as an example of a laborious manual data inspection and bias discovery process, a skin lesion dataset is manually examined. A Global Explanation for the Bias Identification method is proposed as an alternative semi-automatic approach to manual data exploration for discovering potential biases in data. Relevant numerical methods and metrics are discussed for assessing the effects of the identified biases on the model. Whereas identifying errors and bias is critical, improving the model and reducing the number of flaws in the future is an absolute priority. Hence, the second part of the thesis focuses on mitigating the influence of bias on ML models. Three approaches are proposed and discussed: Style Transfer Data Augmentation, Targeted Data Augmentations, and Attribution Feedback. Style Transfer Data Augmentation aims to address shape and texture bias by merging a style of a malignant lesion with a conflicting shape of a benign one. Targeted Data Augmentations randomly insert possible biases into all images in the dataset during the training, as a way to make the process random and, thus, destroy spurious correlations. Lastly, Attribution Feedback is used to fine-tune the model to improve its accuracy by eliminating obvious mistakes and teaching it to ignore insignificant input parts via an attribution loss. The goal of these approaches is to reduce the influence of bias on machine learning models, rather than eliminate it entirely.

Many inductive logic programming approaches struggle to learn programs from noisy data. To overcome this limitation, we introduce an approach that learns minimal description length programs from noisy data, including recursive programs. Our experiments on several domains, including drug design, game playing, and program synthesis, show that our approach can outperform existing approaches in terms of predictive accuracies and scale to moderate amounts of noise.

According to various estimates, brown dwarfs (BD) should account for up to 25 percent of all objects in the Galaxy. However, few of them are discovered and well-studied, both individually and as a population. Homogeneous and complete samples of brown dwarfs are needed for these kinds of studies. Due to their weakness, spectral studies of brown dwarfs are rather laborious. For this reason, creating a significant reliable sample of brown dwarfs, confirmed by spectroscopic observations, seems unattainable at the moment. Numerous attempts have been made to search for and create a set of brown dwarfs using their colours as a decision rule applied to a vast amount of survey data. In this work, we use machine learning methods such as Random Forest Classifier, XGBoost, SVM Classifier and TabNet on PanStarrs DR1, 2MASS and WISE data to distinguish L and T brown dwarfs from objects of other spectral and luminosity classes. The explanation of the models is discussed. We also compare our models with classical decision rules, proving their efficiency and relevance.

ModSecurity is widely recognized as the standard open-source Web Application Firewall (WAF), maintained by the OWASP Foundation. It detects malicious requests by matching them against the Core Rule Set, identifying well-known attack patterns. Each rule in the CRS is manually assigned a weight, based on the severity of the corresponding attack, and a request is detected as malicious if the sum of the weights of the firing rules exceeds a given threshold. In this work, we show that this simple strategy is largely ineffective for detecting SQL injection (SQLi) attacks, as it tends to block many legitimate requests, while also being vulnerable to adversarial SQLi attacks, i.e., attacks intentionally manipulated to evade detection. To overcome these issues, we design a robust machine learning model, named AdvModSec, which uses the CRS rules as input features, and it is trained to detect adversarial SQLi attacks. Our experiments show that AdvModSec, being trained on the traffic directed towards the protected web services, achieves a better trade-off between detection and false positive rates, improving the detection rate of the vanilla version of ModSecurity with CRS by 21%. Moreover, our approach is able to improve its adversarial robustness against adversarial SQLi attacks by 42%, thereby taking a step forward towards building more robust and trustworthy WAFs.