“Improving Trustworthiness of Deep Learning via Inspectable and Robust Representations,” Universität des Saarlandes, Saarbrücken, 2025.
Max Losch (PhD Student)
Personal Information
Research Interests
- Computer Vision
- Machine Learning
Education
- Ph.D. student, Computer Science, Max-Planck-Institute for Informatics, Saarbrücken, Germany, (2017 - present)
- Research Scientist, Brain & Cognition, University of Amsterdam, The Netherlands, 2015-2017
- M.Sc., Machine Learning, Kungliga Tekniska Högskolan, Stockholm, Sweden, 2013-2015
- B.Sc., Computer Science, Freie Universität Berlin, Germany, 2009-2012
Other
See my Google Scholar profile.
Publications
2025
2024
- “On Adversarial Training without Perturbing all Examples,” in The Twelfth International Conference on Learning Representations (ICLR 2024), Vienna, Austria, 2024.
2023
- “Certified Robust Models with Slack Control and Large Lipschitz Constants,” in Pattern Recognition (DAGM GCPR 2023), Heidelberg, Germany, 2023.
2021
- “Semantic Bottlenecks: Quantifying and Improving Inspectability of Deep Representations,” International Journal of Computer Vision, vol. 129, 2021.
2020
- “Analyzing the Dependency of ConvNets on Spatial Information,” in Pattern Recognition (GCPR 2020), Tübingen, Germany, 2021.
- “Semantic Bottlenecks: Quantifying & Improving Inspectability of Deep Representations,” in Pattern Recognition (GCPR 2020), Tübingen, Germany, 2021.
- “Analyzing the Dependency of ConvNets on Spatial Information,” 2020. [Online]. Available: https://arxiv.org/abs/2002.01827.more
Abstract
Intuitively, image classification should profit from using spatial
information. Recent work, however, suggests that this might be overrated in
standard CNNs. In this paper, we are pushing the envelope and aim to further
investigate the reliance on spatial information. We propose spatial shuffling
and GAP+FC to destroy spatial information during both training and testing
phases. Interestingly, we observe that spatial information can be deleted from
later layers with small performance drops, which indicates spatial information
at later layers is not necessary for good performance. For example, test
accuracy of VGG-16 only drops by 0.03% and 2.66% with spatial information
completely removed from the last 30% and 53% layers on CIFAR100, respectively.
Evaluation on several object recognition datasets (CIFAR100, Small-ImageNet,
ImageNet) with a wide range of CNN architectures (VGG16, ResNet50, ResNet152)
shows an overall consistent pattern.
2019
- “Interpretability Beyond Classification Output: Semantic Bottleneck Networks,” 2019. [Online]. Available: http://arxiv.org/abs/1907.10882.more
Abstract
Today's deep learning systems deliver high performance based on end-to-end
training. While they deliver strong performance, these systems are hard to
interpret. To address this issue, we propose Semantic Bottleneck Networks
(SBN): deep networks with semantically interpretable intermediate layers that
all downstream results are based on. As a consequence, the analysis on what the
final prediction is based on is transparent to the engineer and failure cases
and modes can be analyzed and avoided by high-level reasoning. We present a
case study on street scene segmentation to demonstrate the feasibility and
power of SBN. In particular, we start from a well performing classic deep
network which we adapt to house a SB-Layer containing task related semantic
concepts (such as object-parts and materials). Importantly, we can recover
state of the art performance despite a drastic dimensionality reduction from
1000s (non-semantic feature) to 10s (semantic concept) channels. Additionally
we show how the activations of the SB-Layer can be used for both the
interpretation of failure cases of the network as well as for confidence
prediction of the resulting output. For the first time, e.g., we show
interpretable segmentation results for most predictions at over 99% accuracy.