Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Active Learning Strategies for Predictions on Real World Tasks
To support our research efforts in the area of automated human behavior analysis we are looking for a motivated master student that is passionate about applying machine learning on real world tasks to do either a master thesis or semester project on the topic of Active Learning.
Keywords: Active Learning, Machine Learning, Deep Learning, ML, DL, Computer Vision
When training machine learning models to perform well on a task, one of the main bottlenecks is the effort involved in labeling large training sets that are representative of the task domain. Often, this involves manually labeling tens of thousands of data points, which can take weeks and months of time. One supervised learning approach that enables high test accuracies but with much smaller dataset sizes is active learning, which achieves an improved optimization by iteratively training a model and using the trained model in each iteration to choose from unlabeled source samples with the highest prediction uncertainty. Those samples with the highest model uncertainty can then be labeled. By iteratively teaching the model with the labeled samples it is most uncertain of a high accuracy can be achieved in less time and with less labeled data.
When training machine learning models to perform well on a task, one of the main bottlenecks is the effort involved in labeling large training sets that are representative of the task domain. Often, this involves manually labeling tens of thousands of data points, which can take weeks and months of time. One supervised learning approach that enables high test accuracies but with much smaller dataset sizes is active learning, which achieves an improved optimization by iteratively training a model and using the trained model in each iteration to choose from unlabeled source samples with the highest prediction uncertainty. Those samples with the highest model uncertainty can then be labeled. By iteratively teaching the model with the labeled samples it is most uncertain of a high accuracy can be achieved in less time and with less labeled data.
To support the development of a surgical process monitoring system that involves several machine learning components for tasks such as image segmentation, activity recognition and object pose estimation, your task will be to explore active learning strategies and implement them into existing workflows.
For this, you will work on the following topics:
1. Model uncertainty measures that expose problematic situations in the trained models (such as through Bayesian estimation, uncertainty sampling, etc.)
2. Machine Learning tasks: image segmentation, activity recognition and object pose estimation (PyTorch, TF, scikit-learn, etc.)
To support the development of a surgical process monitoring system that involves several machine learning components for tasks such as image segmentation, activity recognition and object pose estimation, your task will be to explore active learning strategies and implement them into existing workflows.
For this, you will work on the following topics:
1. Model uncertainty measures that expose problematic situations in the trained models (such as through Bayesian estimation, uncertainty sampling, etc.) 2. Machine Learning tasks: image segmentation, activity recognition and object pose estimation (PyTorch, TF, scikit-learn, etc.)
- Good programming skills in Python (or Java, C#, C, C++) - Previous experience with machine (deep) learning algorithms and frameworks (scikit-learn, PyTorch, TF, etc) - Methodical way of working - Ability to take ownership in shaping the direction of the project
As part of our research at the AR Lab within the Human Behavior Group we are working on automatically analyzing a user’s interaction with his environment in scenarios such as surgery or in industrial machine interactions. By collecting real-world datasets during those scenarios and using them for machine learning tasks such as activity recognition, object pose estimation or image segmentation we can gain an understanding of how a user performed during a given task. We can then utilize this information to provide the user with real-time feedback on his task using mixed reality devices, such as the Microsoft HoloLens, that can guide him and prevent him from doing mistakes.
As part of our research at the AR Lab within the Human Behavior Group we are working on automatically analyzing a user’s interaction with his environment in scenarios such as surgery or in industrial machine interactions. By collecting real-world datasets during those scenarios and using them for machine learning tasks such as activity recognition, object pose estimation or image segmentation we can gain an understanding of how a user performed during a given task. We can then utilize this information to provide the user with real-time feedback on his task using mixed reality devices, such as the Microsoft HoloLens, that can guide him and prevent him from doing mistakes.
- Master Thesis - Semester Project - Mix between research, conceptual work and coding - Autonomy in shaping the direction of the project
Sophokles Ktistakis <ktistaks@ethz.ch>
For applications please provide your CV.