Discovering Interpretable Representations for Both Deep Generative and Discriminative Models

By Tameem Adel, Zoubin Ghahramani, Adrian Weller. 2018

96 citations recorded by the source corpus.

Subfields: Generative Models, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

In Safety, Robustness & Interpretability

Citations in the corpus

Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, background): “Koh et al. (2020), Losch et al. (2019) and Adel et al. (2018) try to directly learn a concept classifier in the latent space.”
On the Transfer of Disentangled Representations in Realistic Settings (2020, background): “Disentangled representations hold the promise of generalization to unseen scenarios (Higgins et al., 2017b), increased interpretability (Adel et al., 2018; Higgins et al., 2018) and faster learning on downstream tasks (van Steenkiste et al., 2019; Locatello et al., 2019a).”
Weakly-Supervised Disentanglement Without Compromises (2020, background): “A recent line of work argued that representations which are disentangled offer useful properties such as interpretability (Adel et al., 2018; Bengio et al., 2013; Higgins et al., 2017a), predictive performance (Locatello et al.”
On the Fairness of Disentangled Representations (2019, background): “Indeed, isolating each independent factor of variation into the independent components of a representation vector should make it both interpretable and simplify downstream prediction tasks [6, 7, 29, 35, 56, 58, 60, 74, 82, 87, 89, 90, 1, 28].”
Concept whitening for interpretable image recognition (2020, methodology): “ay by normalizing ﬂows and maximizing the mutual information between the transformed representation and side information provided by human users. Although side information could also include concepts,Adel et al. (2018) query for the side information by active learning which is different from ours. W”
Explainable machine learning in deployment (2019, background): “Ideally, if Organization B had guarantees about the disentanglement of data generating factors [3], they would be able to understand which factors (descriptors) play a role in downstream classification.”
Disentangling Factors of Variation Using Few Labels (2019, background): “Exploiting relational information or knowledge of the effect of the factors of variation have both been qualitatively studied to learn disentangled representations (Hinton et al., 2011; Cohen & Welling, 2014; Karaletsos et al., 2015; Goroshin et al., 2015; Whitney et al., 2016; Fraccaro et al., 2017”
XOC: Explainable Observer-Classifier for Explainable Binary Decisions (2019, background): “Work on introspective explanations includes, e.g., visualizing features (Springenberg et al., 2014; Zhou et al., 2016; Selvaraju et al., 2017), saliency maps (Simonyan et al., 2013), interpretable features (Adel et al., 2018), and modular networks (Andreas et al., 2016).”
The Role and Limits of Principles in AI Ethics: Towards a Focus on Tensions (2019, background): “For example, acknowledging that there is currently a tradeoff between performance and interpretability in state-of-the-art machine learning systems has motivated technical research that attempts to reduce or eliminate this trade-off [2].”
Towards Nonlinear Disentanglement in Natural Data with Temporal Sparse Coding (2020, background): “A disentangled representation would be valuable for a wide variety of machine learning applications, including sample efﬁciency for downstream tasks [8, 26], fairness [27, 28] and interpretability [1, 29, 30].”
Transparency: Motivations and Challenges (2017, background): “ng an action which could feasibly be taken by the individual to change the decision. Other approaches seek to identify interpretable representations in order to help understand how an algorithm works [11,29,1]. We have not provided an exhaustive survey of interpretability methods, but hope that it i”
Transparency: Motivations and Challenges (2019, background): “Other approaches seek to identify interpretable representations in order to help understand how an algorithm works [11, 29, 1].”
Is Independence all you need? On the Generalization of Representations Learned from Correlated Data (2020, background): “Due to the induced structure, disentangled representations promise generalization to unseen scenarios (Higgins et al., 2017b), increased interpretability (Adel et al., 2018; Higgins et al., 2018) and faster learning on downstream tasks (van Steenkiste et al., 2019; Locatello et al., 2019a).”
Disentangling Factors of Variations Using Few Labels (2020, background): “Exploiting relational information or knowledge of the effect of the factors of variation have both been qualitatively studied to learn disentangled representations (Hinton et al., 2011; Cohen & Welling, 2014; Karaletsos et al., 2015; Goroshin et al., 2015; Whitney et al., 2016; Fraccaro et al., 2017”