Maurice Weber

Last Name

Weber

First Name

Maurice

ORCID

0000-0002-4176-8222

Show all metadata

Search Results

Publications 1 - 6 of 6

Certifying Out-of-Domain Generalization for Blackbox Functions
Weber, Maurice; Li, Linyi; Wang, Boxin; et al. (2022)
Proceedings of Machine Learning Research ~ Proceedings of the 39th International Conference on Machine Learning
Certifying the robustness of model performance under bounded data distribution drifts has recently attracted intensive interest under the umbrella of distributional robustness. However, existing techniques either make strong assumptions on the model class and loss functions that can be certified, such as smoothness expressed via Lipschitz continuity of gradients, or require to solve complex optimization problems. As a result, the wider application of these techniques is currently limited by its scalability and flexibility — these techniques often do not scale to large-scale datasets with modern deep neural networks or cannot handle loss functions which may be non-smooth such as the 0-1 loss. In this paper, we focus on the problem of certifying distributional robustness for blackbox models and bounded loss functions, and propose a novel certification framework based on the Hellinger distance. Our certification technique scales to ImageNet-scale datasets, complex models, and a diverse set of loss functions. We then focus on one specific application enabled by such scalability and flexibility, i.e., certifying out-of-domain generalization for large neural networks and loss functions such as accuracy and AUC. We experimentally validate our certification method on a number of datasets, ranging from ImageNet, where we provide the first non-vacuous certified out-of-domain generalization, to smaller classification tasks where we are able to compare with the state-of-the-art and show that our method performs considerably better.
RAB: Provable Robustness Against Backdoor Attacks
Weber, Maurice; Xu, Xiaojun; Karlas, Bojan; et al. (2023)
IEEE Symposium on Security and Privacy
Recent studies have shown that deep neural networks (DNNs) are vulnerable to adversarial attacks, including evasion and backdoor (poisoning) attacks. On the defense side, there have been intensive efforts on improving both empirical and provable robustness against evasion attacks; however, the provable robustness against backdoor attacks still remains largely unexplored. In this paper, we focus on certifying the machine learning model robustness against general threat models, especially backdoor attacks. We first provide a unified framework via randomized smoothing techniques and show how it can be instantiated to certify the robustness against both evasion and backdoor attacks. We then propose the first robust training process, RAB, to smooth the trained model and certify its robustness against backdoor attacks. We theoretically prove the robustness bound for machine learning models trained with RAB and prove that our robustness bound is tight. In addition, we theoretically show that it is possible to train the robust smoothed models efficiently for simple models such as Knearest neighbor classifiers, and we propose an exact smoothtraining algorithm that eliminates the need to sample from a noise distribution for such models. Empirically, we conduct comprehensive experiments for different machine learning (ML) models such as DNNs, support vector machines, and K-NN models on MNIST, CIFAR-10, and ImageNette datasets and provide the first benchmark for certified robustness against backdoor attacks. In addition, we evaluate K-NN models on a spambase tabular dataset to demonstrate the advantages of the proposed exact algorithm. Both the theoretic analysis and the comprehensive evaluation on diverse ML models and datasets shed light on further robust learning strategies against general training time attacks.
Probabilistic Robustness Guarantees for Machine Learning Systems
Weber, Maurice (2023)
Certifying Some Distributional Fairness with Subpopulation Decomposition
Kang, Mintong; Li, Linyi; Weber, Maurice; et al. (2022)
Advances in Neural Information Processing Systems 35
Extensive efforts have been made to understand and improve the fairness of machine learning models based on different fairness measurement metrics, especially in high-stakes domains such as medical insurance, education, and hiring decisions. However, there is a lack of certified fairness on the end-to-end performance of an ML model. In this paper, we first formulate the certified fairness of an ML model trained on a given data distribution as an optimization problem based on the model performance loss bound on a fairness constrained distribution, which is within bounded distributional distance with the training distribution. We then propose a general fairness certification framework and instantiate it for both sensitive shifting and general shifting scenarios. In particular, we propose to solve the optimization problem by decomposing the original data distribution into analytical subpopulations and proving the convexity of the sub-problems to solve them. We evaluate our certified fairness on six real-world datasets and show that our certification is tight in the sensitive shifting scenario and provides non-trivial certification under general shifting. Our framework is flexible to integrate additional non-skewness constraints and we show that it provides even tighter certification under different real-world scenarios. We also compare our certified fairness bound with adapted existing distributional robustness bounds on Gaussian data and demonstrate that our method is significantly tighter.
Toward reliability in the NISQ era: Robust interval guarantee for quantum measurements on approximate states
Weber, Maurice; Anand, Abhinav; Cervera-Lierta, Alba; et al. (2022)
Physical Review Research
Near-term quantum computation holds potential across multiple application domains. However, imperfect preparation and evolution of states due to algorithmic and experimental shortcomings, characteristic in the near-term implementation, would typically result in measurement outcomes deviating from the ideal setting. It is thus crucial for any near-term application to quantify and bound these output errors. We address this need by deriving robustness intervals which are guaranteed to contain the output in the ideal setting. The first type of interval is based on formulating robustness bounds as semidefinite programs, and uses only the first moment and the fidelity to the ideal state. Furthermore, we consider higher statistical moments of the observable and generalize bounds for pure states based on the non-negativity of Gram matrices to mixed states, thus enabling their applicability in the NISQ era where noisy scenarios are prevalent. Finally, we demonstrate our results in the context of the variational quantum eigensolver (VQE) on noisy and noiseless simulations.
A deep learning approach for automated scoring of the Rey-Osterrieth complex figure
Langer, Nicolas; Weber, Maurice; Hebling Vieira, Bruno; et al. (2024)
eLife
Memory deficits are a hallmark of many different neurological and psychiatric conditions. The Rey-Osterrieth complex figure (ROCF) is the state-of-the-art assessment tool for neuropsychologists across the globe to assess the degree of non-verbal visual memory deterioration. To obtain a score, a trained clinician inspects a patient's ROCF drawing and quantifies deviations from the original figure. This manual procedure is time-consuming, slow and scores vary depending on the clinician's experience, motivation, and tiredness. Here, we leverage novel deep learning architectures to automatize the rating of memory deficits. For this, we collected more than 20k hand-drawn ROCF drawings from patients with various neurological and psychiatric disorders as well as healthy participants. Unbiased ground truth ROCF scores were obtained from crowdsourced human intelligence. This dataset was used to train and evaluate a multihead convolutional neural network. The model performs highly unbiased as it yielded predictions very close to the ground truth and the error was similarly distributed around zero. The neural network outperforms both online raters and clinicians. The scoring system can reliably identify and accurately score individual figure elements in previously unseen ROCF drawings, which facilitates explainability of the AI-scoring system. To ensure generalizability and clinical utility, the model performance was successfully replicated in a large independent prospective validation study that was pre-registered prior to data collection. Our AI-powered scoring system provides healthcare institutions worldwide with a digital tool to assess objectively, reliably, and time-efficiently the performance in the ROCF test from hand-drawn images.

Publications 1 - 6 of 6

Maurice Weber

Last Name

First Name

ORCID

Organisational unit

Filters

Search Results