skip to main content

Title: Learning Interpretable Representations with Informative Entanglements

Learning interpretable representations in an unsupervised setting is an important yet a challenging task. Existing unsupervised interpretable methods focus on extracting independent salient features from data. However they miss out the fact that the entanglement of salient features may also be informative. Acknowledging these entanglements can improve the interpretability, resulting in extraction of higher quality and a wider variety of salient features. In this paper, we propose a new method to enable Generative Adversarial Networks (GANs) to discover salient features that may be entangled in an informative manner, instead of extracting only disentangled features. Specifically, we propose a regularizer to punish the disagreement between the extracted feature interactions and a given dependency structure while training. We model these interactions using a Bayesian network, estimate the maximum likelihood parameters and calculate a negative likelihood score to measure the disagreement. Upon qualitatively and quantitatively evaluating the proposed method using both synthetic and real-world datasets, we show that our proposed regularizer guides GANs to learn representations with disentanglement scores competing with the state-of-the-art, while extracting a wider variety of salient features.

; ; ; ;
Award ID(s):
1763620 1652107
Publication Date:
Journal Name:
Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence
Page Range or eLocation-ID:
1970 to 1976
Sponsoring Org:
National Science Foundation
More Like this
  1. Disentangled generative models map a latent code vector to a target space, while enforcing that a subset of the learned latent codes are interpretable and associated with distinct properties of the target distribution. Recent advances have been dominated by Variational AutoEncoder (VAE)-based methods, while training disentangled generative adversarial networks (GANs) remains challenging. In this work, we show that the dominant challenges facing disentangled GANs can be mitigated through the use of self-supervision. We make two main contributions: first, we design a novel approach for training disentangled GANs with self-supervision. We propose contrastive regularizer, which is inspired by a natural notion of disentanglement: latent traversal. This achieves higher disentanglement scores than state-of-the-art VAE- and GAN-based approaches. Second, we propose an unsupervised model selection scheme called ModelCentrality, which uses generated synthetic samples to compute the medoid (multi-dimensional generalization of median) of a collection of models. The current common practice of hyper-parameter tuning requires using ground-truths samples, each labelled with known perfect disentangled latent codes. As real datasets are not equipped with such labels, we propose an unsupervised model selection scheme and show that it finds a model close to the best one, for both VAEs and GANs. Combining contrastive regularization with ModelCentrality,more »we improve upon the state-of-the-art disentanglement scores significantly, without accessing the supervised data.« less
  2. The label noise transition matrix, denoting the transition probabilities from clean labels to noisy labels, is crucial for designing statistically robust solutions. Existing estimators for noise transition matrices, e.g., using either anchor points or clusterability, focus on computer vision tasks that are relatively easier to obtain high-quality representations. We observe that tasks with lower-quality features fail to meet the anchor-point or clusterability condition, due to the coexistence of both uninformative and informative representations. To handle this issue, we propose a generic and practical information-theoretic approach to down-weight the less informative parts of the lower-quality features. This improvement is crucial to identifying and estimating the label noise transition matrix. The salient technical challenge is to compute the relevant information-theoretical metrics using only noisy labels instead of clean ones. We prove that the celebrated f-mutual information measure can often preserve the order when calculated using noisy labels. We then build our transition matrix estimator using this distilled version of features. The necessity and effectiveness of the proposed method are also demonstrated by evaluating the estimation error on a varied set of tabular data and text classification tasks with lower-quality features.
  3. Luigi Martelli, Pier (Ed.)
    Abstract Motivation Recent technological advances produce a wealth of high-dimensional descriptions of biological processes, yet extracting meaningful insight and mechanistic understanding from these data remains challenging. For example, in developmental biology, the dynamics of differentiation can now be mapped quantitatively using single-cell RNA sequencing, yet it is difficult to infer molecular regulators of developmental transitions. Here, we show that discovering informative features in the data is crucial for statistical analysis as well as making experimental predictions. Results We identify features based on their ability to discriminate between clusters of the data points. We define a class of problems in which linear separability of clusters is hidden in a low-dimensional space. We propose an unsupervised method to identify the subset of features that define a low-dimensional subspace in which clustering can be conducted. This is achieved by averaging over discriminators trained on an ensemble of proposed cluster configurations. We then apply our method to single-cell RNA-seq data from mouse gastrulation, and identify 27 key transcription factors (out of 409 total), 18 of which are known to define cell states through their expression levels. In this inferred subspace, we find clear signatures of known cell types that eluded classification prior to discovery ofmore »the correct low-dimensional subspace. Availability and implementation Supplementary information Supplementary data are available at Bioinformatics online.« less
  4. Abstract

    Continuous seismograms contain a wealth of information with a large variety of signals with different origin. Identifying these signals is a crucial step in understanding physical geological objects. We propose a strategy to identify classes of signals in continuous single‐station seismograms in an unsupervised fashion. Our strategy relies on extracting meaningful waveform features based on a deep scattering network combined with an independent component analysis. Based on the extracted features, agglomerative clustering then groups these waveforms in a hierarchical fashion and reveals the process of clustering in a dendrogram. We use the dendrogram to explore the seismic data and identify different classes of signals. To test our strategy, we investigate a two‐day‐long seismogram collected in the vicinity of the North Anatolian Fault, Turkey. We analyze the automatically inferred clusters' occurrence rate, spectral characteristics, cluster size, and waveform and envelope characteristics. At a low level in the cluster hierarchy, we obtain three clusters related to anthropogenic and ambient seismic noise and one cluster related to earthquake activity. At a high level in the cluster hierarchy, we identify a seismic burst that includes around 200 events with similar waveforms and high‐frequent signals with correlating envelopes and an anthropogenic origin. The applicationmore »shows that the cluster hierarchy helps to identify particular families of signals and to extract subclusters for further analysis. This is valuable when certain types of signals, such as earthquakes, are under‐represented in the data. The proposed method may also successfully discover new types of signals since it is entirely data‐driven.

    « less
  5. Node embedding is the task of extracting concise and informative representations of certain entities that are connected in a network. Various real-world networks include information about both node connectivity and certain node attributes, in the form of features or time-series data. Modern representation learning techniques employ both the connectivity and attribute information of the nodes to produce embeddings in an unsupervised manner. In this context, deriving embeddings that preserve the geometry of the network and the attribute vectors would be highly desirable, as they would reflect both the topological neighborhood structure and proximity in feature space. While this is fairly straightforward to maintain when only observing the connectivity or attribute information of the network, preserving the geometry of both types of information is challenging. A novel tensor factorization approach for node embedding in attributed networks is proposed in this paper, that preserves the distances of both the connections and the attributes. Furthermore, an effective and lightweight algorithm is developed to tackle the learning task and judicious experiments with multiple state-of-the-art baselines suggest that the proposed algorithm offers significant performance improvements in downstream tasks.