About
For a number of different applications, it’s really important that neural networks don’t make flawed predictions. As a PhD student in the Oxford Robotics Institute, I’ve spent the last few years researching uncertainty estimation for deep learning to address this problem in the context of safe robot deployment. Over the course of my studies, my group and I have developed a number of methods that can train neural networks that make either accurate and confident estimates, or express high uncertainty (seen below).
More broadly, my main interest is the design of techniques that allow the robust application of deep learning to challenging real-world problems.
Code:
ue_testing - Benchmarking quality of neural network uncertainty estimates
gammassl - Training neural networks to estimate uncertainty from unlabelled data
Publications:
Masked Gamma-SSL: Learning Uncertainty Estimation via Masked Image Modeling
D. Williams, M. Gadd, P. Newman and D. De Martini
2024 IEEE International Conference on Robotics and Automation (ICRA)
Mitigating Distributional Shift in Semantic Segmentation via Uncertainty Estimation from Unlabelled Data
D. Williams, D. De Martini, M. Gadd and P. Newman
2024 IEEE Transactions on Robotics (T-RO)
Doppler-aware Odometry from FMCW Scanning Radar
F. Rennie, D. Williams, P. Newman and D. De Martini
2023 IEEE Intelligent Transportation Systems Conference (ITSC)
Fool Me Once: Robust Selective Segmentation via Out-of-Distribution Detection with Contrastive Learning
D. Williams, M. Gadd, D. De Martini and P. Newman
2021 IEEE International Conference on Robotics and Automation (ICRA)
Keep off the Grass: Permissible Driving Routes from Radar with Weak Audio Supervision
D. Williams, D. De Martini, M. Gadd, L. Marchegiani and P. Newman
2020 IEEE Intelligent Transportation Systems Conference (ITSC)
Thesis:
Chapter 2: Introduction to Semantic Segmentation
Chapter 3: Uncertainty Estimation in Deep Learning
Chapter 4: Model Evaluation and Datasets
Chapter 5: Learning OoD Detection from Large-Scale Datasets
Chapter 6: Learning Uncertainty Estimation from Uncurated Domain Data
Chapter 7: Learning Uncertainty Estimation with Masking & Foundation Models