Machine Learning and Deep Learning
Machine learning and deep learning methods combine approaches from statistics and computer science to make predictions, perform classification or recognise patterns even in complex situations.
In many recent applications, underlying processes, relations, and mechanisms are not well understood, and classical modeling approaches (e.g., based on physical theory or classical statistics) are not applicable. However, if a sufficient amount of data is available, modern machine learning algorithms can still provide good predictions without explicit specification of a model.
At IDP, we supplement classical statistical methods (e.g., regression analysis), with random forests, boosting, support vector machines and deep neural networks (deep learning) to solve complex prediction and classification problems. So-called unsupervised learning algorithms or data mining methods can also discover new structures and patterns in data sets (e.g., cluster analysis).
Machine learning methods are also suitable for analyzing data that are not available in tabular form, e.g., text data or network structures. For digital images (e.g., in medical diagnostics), audio data, or similar signals, the latest generation of neural networks ("deep learning") yields results with low, previously unattained error rates.
- Deep-Learning based Classification of Historical Subtypes of Lung Tumors
Deep Learning methods (convolutional neural network) can distinguish different types of lung tumors from images of histological sections as accurate as a pathologist.
- Artificial Intelligence in Real-Time-Simulations
Deep learning methods (reinforcement learning algorithms with Q-learning) were used to find an optimal action strategy in a real-time simulation.
Conferences organized by IDP or co-organized by IDP members: