PROBABILISTIC AI

Research Group

The Probabilistic AI Research Group (PAIR)

Despite the groundbreaking progress in deep learning and AI, these technologies often struggle to quantify uncertainties and lack interpretability. This limitation is evident in issues like hallucinations in large language models, where generated outputs can be plausible yet incorrect. Such challenges are critical in various contexts, including load forecasting and medical diagnostics.

Example Applications

Creating AI models that are precise and understandable for medical professionals, particularly in medical imaging.

Developing AI systems that make transparent and understandable decisions, enhancing trust and accountability.

Improving production planning by including uncertainty in the prediction.

Knowing When You Don’t Know:

Developing systems that can recognize and appropriately respond when they encounter scenarios outside their training data, such as:

 

Industrial Maintenance: Detecting when machinery data indicates an unknown issue, knowing when you don’t know the specifics, and triggering a manual inspection before failure occurs.

 

Customer Service: Knowing when you don’t know how to adequately address a customer query, and escalating it to a human representative when the chatbot or automated system cannot provide a satisfactory solution.

Probabilistic AI addresses these challenges by modeling uncertainties, making predictions more robust, interpretable, and reliable. By incorporating probabilistic methods with deep learning, we can enhance the capability of AI systems to handle real-world phenomena with inherent uncertainties.

 

The Probabilistic AI Research (PAIR) Group at TIDIT focuses on the innovative fusion of probabilistic methods with deep learning. Our mission is to develop cutting-edge techniques for data analysis and predictive modeling. This approach is crucial for managing uncertainties in real-world scenarios.

Leadership

TOBI DEFINED

Research Group Leader

Prof. Dr. Oliver Dürr

Co-Lead

Members

Postdoc

PhD Candidate

Research Assistant

Latest Publications

  • B Sick, T Hathorn, O Dürr - 2021
    Abstract: We present a deep transformation model for probabilistic regression. Deep learning is known for outstandingly accurate predictions on complex data but in regression tasks it is predominantly used to just predict a single number. This ignores the non-deterministic character of most tasks. Especially if crucial decisions are based on the pre...

  • L Kook, L Herzog, T Hothorn, O Dürr, B Sick - Pattern Recognition - 2022•Elsevier
    Abstract: Outcomes with a natural order commonly occur in prediction problems and often the available input data are a mixture of complex data like images and tabular predictors. Deep Learning (DL) models are state-of-the-art for image classification tasks but frequently treat ordinal outcomes as unorde...

  • O Dürr, B Sick, E Murina - 2020
    Abstract: Probabilistic Deep Learning is a hands-on guide to the principles that support neural networks. Learn to improve network performance with the right distribution for different data types, and discover Bayesian variants that can state their own uncertainty to increase accuracy. This book provides easy-to-apply code and uses popular framewo...

  • O Dürr, S Hörtling, D Dold, I Kovylov, B Sick - AStA Advances in Statistical Analysis, 2024
    Abstract: Black-box variational inference (BBVI) is a technique to approximate the posterior of Bayesian models by optimization. Similar to MCMC, the user only needs to specify the model; then, the inference procedure is done automatically. In contrast to MCMC, BBVI scales to many observatio...

  • M Arpogaus, M Voss, B Sick, M Nigge-Uricher, O Dürr - IEEE Transactions on Smart Grid, 2023
    Abstract: The transition to a fully renewable energy grid requires better forecasting of demand at the low-voltage level to increase efficiency and ensure reliable control. However, high fluctuations and increasing electrification cause huge forecast variability, not reflected in traditional...

  • L Herzog, L Kook, A Götschi, K Petermann, M Hänsel, J Hamann, O Dürr, S Wegener, B Sick - Biometrical Journal, 2023
    Abstract: In many medical applications, interpretable models with high prediction performance are sought. Often, those models are required to handle semistructured data like tabular and image data. We show how to apply deep transformation models (DTMs) for distributio...

  • L Kook, PFM Baumann, O Dürr, B Sick, D Rügamer - arXiv preprint arXiv:2211.13665, 2022
    Abstract: Contemporary empirical applications frequently require flexible regression models for complex response types and large tabular or non-tabular, including image or text, data. Classical regression models either break down under the computational load of processing such data or require add...

  • C Berlin, S Adomeit, P Grover, M Dreischarf, H Halm, O Dürr, P Obid - Global Spine Journal, 2024
    Study design: Retrospective, mono-centric cohort research study. Objectives: The purpose of this study is to validate a novel artificial intelligence (AI)-based algorithm against human-generated ground truth for radiographic parameters of adolescent idiopathic scoliosis (AIS). Methods:...

  • O Dürr, PY Fan, ZX Yin - IEEE Sensors Journal, 2023
    Abstract: This study aims to investigate the utilization of Bayesian techniques for the calibration of micro-electro-mechanical system (MEMS) accelerometers. These devices have garnered substantial interest in various practical applications and typically require calibration through error-correcting functions. The parameters of the...

  • D Dold, D Rügamer, B Sick, O Dürr - International Conference on Artificial Intelligence and Statistics, 2024
    Abstract: Semi-structured regression models enable the joint modeling of interpretable structured and complex unstructured feature effects. The structured model part is inspired by statistical models and can be used to infer the input-output relationship for features of part...

  • O Dürr, PY Fan, ZX Yin - IEEE Sensors Journal, 2023
    Abstract: This study aims to investigate the utilization of Bayesian techniques for the calibration of micro-electro-mechanical system (MEMS) accelerometers. These devices have garnered substantial interest in various practical applications and typically require calibration through error-correcting functions. The parameters of the...

Latest Publications:

Latest Publications:

Latest Publications:

  • B Sick, T Hathorn, O Dürr - 2021
    Abstract: We present a deep transformation model for probabilistic regression. Deep learning is known for outstandingly accurate predictions on complex data but in regression tasks it is predominantly used to just predict a single number. This ignores the non-deterministic character of most tasks. Especially if crucial decisions are based on the pre...

  • L Kook, L Herzog, T Hothorn, O Dürr, B Sick - Pattern Recognition - 2022•Elsevier
    Abstract: Outcomes with a natural order commonly occur in prediction problems and often the available input data are a mixture of complex data like images and tabular predictors. Deep Learning (DL) models are state-of-the-art for image classification tasks but frequently treat ordinal outcomes as unorde...

  • O Dürr, B Sick, E Murina - 2020
    Abstract: Probabilistic Deep Learning is a hands-on guide to the principles that support neural networks. Learn to improve network performance with the right distribution for different data types, and discover Bayesian variants that can state their own uncertainty to increase accuracy. This book provides easy-to-apply code and uses popular framewo...

  • O Dürr, S Hörtling, D Dold, I Kovylov, B Sick - AStA Advances in Statistical Analysis, 2024
    Abstract: Black-box variational inference (BBVI) is a technique to approximate the posterior of Bayesian models by optimization. Similar to MCMC, the user only needs to specify the model; then, the inference procedure is done automatically. In contrast to MCMC, BBVI scales to many observatio...

  • M Arpogaus, M Voss, B Sick, M Nigge-Uricher, O Dürr - IEEE Transactions on Smart Grid, 2023
    Abstract: The transition to a fully renewable energy grid requires better forecasting of demand at the low-voltage level to increase efficiency and ensure reliable control. However, high fluctuations and increasing electrification cause huge forecast variability, not reflected in traditional...

  • L Herzog, L Kook, A Götschi, K Petermann, M Hänsel, J Hamann, O Dürr, S Wegener, B Sick - Biometrical Journal, 2023
    Abstract: In many medical applications, interpretable models with high prediction performance are sought. Often, those models are required to handle semistructured data like tabular and image data. We show how to apply deep transformation models (DTMs) for distributio...

  • L Kook, PFM Baumann, O Dürr, B Sick, D Rügamer - arXiv preprint arXiv:2211.13665, 2022
    Abstract: Contemporary empirical applications frequently require flexible regression models for complex response types and large tabular or non-tabular, including image or text, data. Classical regression models either break down under the computational load of processing such data or require add...

  • C Berlin, S Adomeit, P Grover, M Dreischarf, H Halm, O Dürr, P Obid - Global Spine Journal, 2024
    Study design: Retrospective, mono-centric cohort research study. Objectives: The purpose of this study is to validate a novel artificial intelligence (AI)-based algorithm against human-generated ground truth for radiographic parameters of adolescent idiopathic scoliosis (AIS). Methods:...

  • O Dürr, PY Fan, ZX Yin - IEEE Sensors Journal, 2023
    Abstract: This study aims to investigate the utilization of Bayesian techniques for the calibration of micro-electro-mechanical system (MEMS) accelerometers. These devices have garnered substantial interest in various practical applications and typically require calibration through error-correcting functions. The parameters of the...

  • D Dold, D Rügamer, B Sick, O Dürr - International Conference on Artificial Intelligence and Statistics, 2024
    Abstract: Semi-structured regression models enable the joint modeling of interpretable structured and complex unstructured feature effects. The structured model part is inspired by statistical models and can be used to infer the input-output relationship for features of part...

  • O Dürr, PY Fan, ZX Yin - IEEE Sensors Journal, 2023
    Abstract: This study aims to investigate the utilization of Bayesian techniques for the calibration of micro-electro-mechanical system (MEMS) accelerometers. These devices have garnered substantial interest in various practical applications and typically require calibration through error-correcting functions. The parameters of the...

Example Applications:

Interpretable Models for Medical Applications:

Creating AI models that are precise and understandable for medical professionals, particularly in medical imaging.

Responsible AI through Causal Models:

Developing AI systems that make transparent and understandable decisions, enhancing trust and accountability.

Planning with Uncertainties:

Improving production planning by including uncertainty in the prediction.

Uncertainty Awareness

Knowing When You Don’t Know:

Developing systems that can recognize and appropriately respond when they encounter scenarios outside their training data, such as:

 

Industrial Maintenance: Detecting when machinery data indicates an unknown issue, knowing when you don’t know the specifics, and triggering a manual inspection before failure occurs.

 

Customer Service: Knowing when you don’t know how to adequately address a customer query, and escalating it to a human representative when the chatbot or automated system cannot provide a satisfactory solution.

Probabilistic AI addresses these challenges by modeling uncertainties, making predictions more robust, interpretable, and reliable. By incorporating probabilistic methods with deep learning, we can enhance the capability of AI systems to handle real-world phenomena with inherent uncertainties.

 

The Probabilistic AI Research (PAIR) Group at TIDIT focuses on the innovative fusion of probabilistic methods with deep learning. Our mission is to develop cutting-edge techniques for data analysis and predictive modeling. This approach is crucial for managing uncertainties in real-world scenarios.

Example Applications:

Creating AI models that are precise and understandable for medical professionals, particularly in medical imaging.

Developing AI systems that make transparent and understandable decisions, enhancing trust and accountability.

Improving production planning by including uncertainty in the prediction.

Knowing When You Don’t Know:

Developing systems that can recognize and appropriately respond when they encounter scenarios outside their training data, such as:

 

Industrial Maintenance: Detecting when machinery data indicates an unknown issue, knowing when you don’t know the specifics, and triggering a manual inspection before failure occurs.

 

Customer Service: Knowing when you don’t know how to adequately address a customer query, and escalating it to a human representative when the chatbot or automated system cannot provide a satisfactory solution.

The Probabilistic AI Research Group (PAIR)

Example Applications:

Creating AI models that are precise and understandable for medical professionals, particularly in medical imaging.

Developing AI systems that make transparent and understandable decisions, enhancing trust and accountability.

Improving production planning by including uncertainty in the prediction.

Knowing When You Don’t Know:

Developing systems that can recognize and appropriately respond when they encounter scenarios outside their training data, such as:

 

Industrial Maintenance: Detecting when machinery data indicates an unknown issue, knowing when you don’t know the specifics, and triggering a manual inspection before failure occurs.

 

Customer Service: Knowing when you don’t know how to adequately address a customer query, and escalating it to a human representative when the chatbot or automated system cannot provide a satisfactory solution.

Despite the groundbreaking progress in deep learning and AI, these technologies often struggle to quantify uncertainties and lack interpretability. This limitation is evident in issues like hallucinations in large language models, where generated outputs can be plausible yet incorrect. Such challenges are critical in various contexts, including load forecasting and medical diagnostics.

Probabilistic AI addresses these challenges by modeling uncertainties, making predictions more robust, interpretable, and reliable. By incorporating probabilistic methods with deep learning, we can enhance the capability of AI systems to handle real-world phenomena with inherent uncertainties.

 

The Probabilistic AI Research (PAIR) Group at TIDIT focuses on the innovative fusion of probabilistic methods with deep learning. Our mission is to develop cutting-edge techniques for data analysis and predictive modeling. This approach is crucial for managing uncertainties in real-world scenarios.

The Probabilistic AI Research Group (PAIR)

Despite the groundbreaking progress in deep learning and AI, these technologies often struggle to quantify uncertainties and lack interpretability. This limitation is evident in issues like hallucinations in large language models, where generated outputs can be plausible yet incorrect. Such challenges are critical in various contexts, including load forecasting and medical diagnostics.

Example Applications:

Creating AI models that are precise and understandable for medical professionals, particularly in medical imaging.

Developing AI systems that make transparent and understandable decisions, enhancing trust and accountability.

Improving production planning by including uncertainty in the prediction.

Knowing When You Don’t Know:

Developing systems that can recognize and appropriately respond when they encounter scenarios outside their training data, such as:

 

Industrial Maintenance: Detecting when machinery data indicates an unknown issue, knowing when you don’t know the specifics, and triggering a manual inspection before failure occurs.

 

Customer Service: Knowing when you don’t know how to adequately address a customer query, and escalating it to a human representative when the chatbot or automated system cannot provide a satisfactory solution.

Probabilistic AI addresses these challenges by modeling uncertainties, making predictions more robust, interpretable, and reliable. By incorporating probabilistic methods with deep learning, we can enhance the capability of AI systems to handle real-world phenomena with inherent uncertainties.

 

The Probabilistic AI Research (PAIR) Group at TIDIT focuses on the innovative fusion of probabilistic methods with deep learning. Our mission is to develop cutting-edge techniques for data analysis and predictive modeling. This approach is crucial for managing uncertainties in real-world scenarios.

By incorporating probabilistic methods with deep learning, we can enhance the capability of AI systems to handle real-world phenomena with inherent uncertainties.

Example Applications:

Creating AI models that are precise and understandable for medical professionals, particularly in medical imaging.

Developing AI systems that make transparent and understandable decisions, enhancing trust and accountability.

Improving production planning by including uncertainty in the prediction.

Knowing When You Don’t Know:

Developing systems that can recognize and appropriately respond when they encounter scenarios outside their training data, such as:

 

Industrial Maintenance: Detecting when machinery data indicates an unknown issue, knowing when you don’t know the specifics, and triggering a manual inspection before failure occurs.

 

Customer Service: Knowing when you don’t know how to adequately address a customer query, and escalating it to a human representative when the chatbot or automated system cannot provide a satisfactory solution.