In this paper we present a novel mechanism to get explanations that allow to better understand network predictions when dealing with sequential data. Specifically, we adopt memory-based networks — Differential Neural Computers — to exploit their capability of storing data in memory and reusing it for inference. By tracking both the memory access at prediction time, and the information stored by the network at each step of the input sequence, we can retrieve the most relevant input steps associated to each prediction. We validate our approach (1) on a modified T-maze, which is a non-Markovian discrete control task evaluating an algorithm’s ability to correlate events far apart in history, and (2) on the Story Cloze Test, which is a commonsense reasoning framework for evaluating story understanding that requires a system to choose the correct ending to a four-sentence story. Our results show that we are able to explain agent’s decisions in (1) and to reconstruct the most relevant sentences used by the network to select the story ending in (2). Additionally, we show not only that by removing those sentences the network prediction changes, but also that the same are sufficient to reproduce the inference.

The use and creation of machine-learning-based solutions to solve problems or reduce their computational costs are becoming increasingly widespread in many domains. Deep Learning plays a large part in this growth. However, it has drawbacks such as a lack of explainability and behaving as a black-box model. During the last few years, Visual Analytics has provided several proposals to cope with these drawbacks, supporting the emerging eXplainable Deep Learning field. This survey aims to (i) systematically report the contributions of Visual Analytics for eXplainable Deep Learning; (ii) spot gaps and challenges; (iii) serve as an anthology of visual analytical solutions ready to be exploited and put into operation by the Deep Learning community (architects, trainers and end users) and (iv) prove the degree of maturity, ease of integration and results for specific domains. The survey concludes by identifying future research challenges and bridging activities that are helpful to strengthen the role of Visual Analytics as effective support for eXplainable Deep Learning and to foster the adoption of Visual Analytics solutions in the eXplainable Deep Learning community. An interactive explorable version of this survey is available online at https:// awarediag-sapienza.github.io/ VA4XDL.

No abstract

Deep neural networks are the driving force of the recent explosion of machine learning applications in everyday life. However, they usually require a lot of training data to work well, and they act as black-boxes, making predictions without any explanation about them. This paper presents Memory Wrap, a module (i.e, a set of layers) that can be added to deep learning models to improve their performance and interpretability in settings where few data are available. Memory Wrap adopts a sparse content-attention mechanism between the input and some memories of past training samples. We show that adding Memory Wrap to standard deep neural networks improves their performance when they learn from a limited set of data, and allows them to reach comparable performance when they learn from the full dataset. We discuss how the analysis of its structure and content-attention weights helps to get insights about its decision process and makes their predictions more interpretable, compared to the same networks without Memory Wrap. We test our approach on image classification tasks using several networks on three different datasets, namely CIFAR10, SVHN, and CINIC10.

The recent success of deep learning models in solving complex problems and in different domains has increased interest in understanding what they learn. Therefore, different approaches have been employed to explain these models, one of which uses human-understandable concepts as explanations. Two examples of methods that use this approach are Network Dissection [5] and Compositional explanations [23]. The former explains units using atomic concepts, while the latter makes explanations more expressive, replacing atomic concepts with logical forms. While intuitively, logical forms are more informative than atomic concepts, it is not clear how to quantify this improvement, and their evaluation is often based on the same metric that is optimized during the search-process and on the usage of hyper-parameters to be tuned. In this paper, we propose to use as evaluation metric the Detection Accuracy, which measures units' consistency of detection of their assigned explanations. We show that this metric (1) evaluates explanations of different lengths effectively, (2) can be used as a stopping criterion for the compositional explanation search, eliminating the explanation length hyper-parameter, and (3) exposes new specialized units whose length 1 explanations are the perceptual abstractions of their longer explanations. * Contact Author Preprint. Under review.