Company earnings calls are crucial events that provide transparency into a company’s financial health and prospects. Large language models (LLMs) offer a promising approach to automatically generate the first draft of earnings call scripts from financial data and past examples. We evaluate two methods: 1) Few-shot prompt engineering with a state-of-the-art model, and 2) Fine-tuning a language model on earnings call transcript data. Our results indicate both approaches can produce coherent scripts covering key metrics, updates, and guidance. However, there are trade-offs in comprehensiveness, hallucinations, writing style, ease of use, and cost. We discuss the pros and cons of each method to guide practitioners on effectively leveraging large language models for this financial communication task.

Accurately determining the indoor location of mobile devices has garnered great interest due to its significant challenge in locating sources due to non-line-of-sight propagation and multipath effects. To address this challenge, This paper proposes a new approach to indoor positioning that utilises channel state information (CSI) and machine learning (ML) techniques to improve Accuracy. The proposed method extracts subcarrier amplitude and phase differences from CSI data to create fingerprints, which are then clustered to identify the number of groups of data and split into two sub-databases using a threshold. The ML algorithms and network architecture are used to train both sub-databases of fingerprints. Experiments conducted in a standard indoor environment demonstrate the effectiveness of the proposed method.

The ease with which mobile money is used to facilitate cross-border payments presents a global threat to law enforcement in the fight against laundering and terrorist financing. This paper aims to use machine learning classifiers to predict transactions flagged as a fraud in mobile money transfers. Data for this paper came from real-time transactions that stimulate a  well-known mobile transfer fraud scheme. This paper uses logistic regression as the baseline model and compares it with ensembles and gradient descent models. The results indicate that the established logistic regression model did not perform too poorly compared to the other models. The random forest classifier had the most outstanding performance among all measures. The amount of money transferred was the top feature to predict money laundering transactions through mobile money transfers. These findings suggest that more research is needed to improve the logistic regression model. The random forest classifier should be further explored as a potential tool for law enforcement and financial institutions to detect money laundering activities in mobile money transfers.

Patents show how technology evolves in most scientific fields over time. The best way to use this valuable knowledge base is to use efficient and effective information retrieval and searches for related prior art. Patent classification, i.e., assigning a patent to one or more predefined categories, is a fundamental step towards synthesizing the information content of an invention. To this end, architectures based on Transformers, especially those derived from the BERT family have already been proposed in the literature and they have shown remarkable results by setting a new state-of-the-art performance for the classification task. Here, we study how domain adaptation can push the performance boundaries in patent classification by rigorously evaluating and implementing a collection of recent transfer learning techniques, e.g., domain-adaptive pretraining and adapters. Our analysis shows how leveraging these advancements enables the development of state-of-the-art models with increased precision, recall, and F1-score. We base our evaluation on both standard patent classification datasets derived from patent offices-defined code hierarchies and more practical real-world use-case scenarios containing labels from the agrochemical industrial domain. The application of these domain adapted techniques to patent classification in a multilingual setting is also examined and evaluated.

DARPA formulated the Explainable Artificial Intelligence (XAI) program in 2015 with the goal to enable end users to better understand, trust, and effectively manage artificially intelligent systems. In 2017, the four-year XAI research program began. Now, as XAI comes to an end in 2021, it is time to reflect on what succeeded, what failed, and what was learned. This article summarizes the goals, organization, and research progress of the XAI Program.

We present a focused analysis of user studies in explainable artificial intelligence (XAI) entailing qualitative investigation. We draw on social science corpora to suggest ways for improving the rigor of studies where XAI researchers use observations, interviews, focus groups, and/or questionnaires to capture qualitative data. We contextualize the presentation of the XAI papers included in our analysis according to the components of rigor described in the qualitative research literature: 1) underlying theories or frameworks, 2) methodological approaches, 3) data collection methods, and 4) data analysis processes. The results of our analysis support calls from others in the XAI community advocating for collaboration with experts from social disciplines to bolster rigor and effectiveness in user studies.

One could be excused for assuming that deep learning had or will soon usurp all credible work in reasoning, artificial intelligence and statistics, but like most ‘meme’ class broad generalizations the concept does not hold up to scrutiny. Memes don’t generally matter since the experts will always know better but in the case of Bayesian software like Stan and PyMC3 even its developers and advocates bemoan the apparent dominance of deep learning as manifested in popular culture, breathtaking performance and most problematically from funding agency peer review that impacts our ability to further advance the field. The facts however do not support the assumed dominance of deep learning in science upon closer examination. This letter simply makes the argument by the crudest of possible metrics, citation count, that once Computer Science is subtracted, Bayesian software accounts for nearly a third of research citations. Stan and PyMC3 dominate some fields, PyTorch, Keras and TensorFlow dominate others with lots of variation in between. Bayesian and deep learning approaches are related but very different technologies in goals, implementation and applicability with little actual overlap so this is not a surprise. While deep learning is backed by industry behemoths (Google, Facebook) the Bayesian efforts are not and it would behoove funders to recognize the impact of Bayesian software given its centrality to science.

Automatic video understanding is becoming more important for applications where real-time performance is crucial and compute is limited. Yet, accurate solutions so far have been computationally intensive. We propose efficient models for videos - Tiny Video Networks - which are video architectures, automatically designed to comply with fast runtimes and, at the same time are effective at video recognition tasks. The Tiny Video Networks run at faster-than-real-time speeds and demonstrate strong performance across several video benchmarks. These models not only provide new tools for real-time video applications, but also enable fast research and development in video understanding. Code and models are available.

Attention maps, a popular heatmap-based explanation method for Visual Question Answering (VQA), are supposed to help users understand the model by highlighting portions of the image/question used by the model to infer answers. However, we see that users are often misled by current attention map visualizations that point to relevant regions despite the model producing an incorrect answer. Hence, we propose Error Maps that clarify the error by highlighting image regions where the model is prone to err. Error maps can indicate when a correctly attended region may be processed incorrectly leading to an incorrect answer, and hence, improve users’ understanding of those cases. To evaluate our new explanations, we further introduce a metric that simulates users’ interpretation of explanations to evaluate their potential helpfulness to understand model correctness. We finally conduct user studies to see that our new explanations help users understand model correctness better than baselines by an expected 30% and that our proxy helpfulness metrics correlate strongly (rho>0.97) with how well users can predict model correctness.

The goal of continual learning (CL) is to learn a sequence of tasks without suffering from the phenomenon of catastrophic forgetting. Previous work has shown that leveraging memory in the form of a replay buffer can reduce performance degradation on prior tasks. We hypothesize that forgetting can be further reduced when the model is encouraged to remember the evidence for previously made decisions. As a first step towards exploring this hypothesis, we propose a simple novel training paradigm, called Remembering for the Right Reasons (RRR), that additionally stores visual model explanations for each example in the buffer and ensures the model has “the right reasons” for its predictions by encouraging its explanations to remain consistent with those used to make decisions at training time. Without this constraint, there is a drift in explanations and increase in forgetting as conventional continual learning algorithms learn new tasks. We demonstrate how RRR can be easily added to any memory or regularization-based approach and results in reduced forgetting, and more importantly, improved model explanations. We have evaluated our approach in the standard and few-shot settings and observed a consistent improvement across various CL approaches using different architectures and techniques to generate model explanations and demonstrated our approach showing a promising connection between explainability and continual learning. Our code is available at \url{https://github.com/SaynaEbrahimi/Remembering-for-the-Right-Reasons}

In the domain of Visual Question Answering (VQA), studies have shown improvement in users’ mental model of the VQA system when they are exposed to examples of how these systems answer certain Image-Question (IQ) pairs. In this work, we show that showing controlled counterfactual image-question examples are more effective at improving the mental model of users as compared to simply showing random examples. We compare a generative approach and a retrieval-based approach to show counterfactual examples. We use recent advances in generative adversarial networks (GANs) to generate counterfactual images by deleting and inpainting certain regions of interest in the image. We then expose users to changes in the VQA system’s answer on those altered images. To select the region of interest for inpainting, we experiment with using both human-annotated attention maps and a fully automatic method that uses the VQA system’s attention values. Finally, we test the user’s mental model by asking them to predict the model’s performance on a test counterfactual image. We note an overall improvement in users’ accuracy to predict answer change when shown counterfactual explanations. While realistic retrieved counterfactuals obviously are the most effective at improving the mental model, we show that a generative approach can also be equally effective.

COGLE (COmmon Ground Learning and Explanation) is an explainable artificial intelligence (XAI) system for autonomous drones that deliver supplies in mountainous areas to field units. The drone missions have risks that vary with topography, flight decisions, and mission goals in a simulated environment. Users must determine which AI-controlled drone is better for a mission. Narrative explanations identify the advantages of a drone’s plan (“What?”) and reasons that the better drone is able to do them (“Why?”). Visual explanations highlight risks from obstacles that users may have overlooked (“Where?”). A model induction user study showed that post-decision explanations produced a small effect on the participants’ abilities to identify the better of two imperfect drones and their plans for a mission, but they did not teach participants to judge the multiple success factors in complex missions as well as the AI pilots. In a decision support variation of the task, users would receive pre-decision explanations to help them to decide when to trust the XAI’s decision. In a fielded XAI application, every drone available for a mission may lack some competencies. We created a proof-of-concept demonstration of automatic ways to combine knowledge from multiple imperfect AIs to get better solutions that the individual AIs do not find on their own. This paper reports on the research challenges, technical approach, and findings of the project and also reflects on the multidisciplinary journey that we took.

We study a user-guided approach for producing global explanations of deep networks for image recognition. The global explanations are produced with respect to a test data set and give the overall frequency of different “recognition reasons” across the data. Each reason corresponds to a small number of the most significant human-recognizable visual concepts used by the network. The key challenge is that the visual concepts cannot be predetermined and those concepts will often not correspond to existing vocabulary or have labelled data sets. We address this issue via an interactive-naming interface, which allows users to freely cluster significant image regions in the data into visually similar concepts. Our main contribution is a user study on two visual recognition tasks. The results show that the participants were able to produce a small number of visual concepts sufficient for explanation and that there was significant agreement among the concepts, and hence global explanations, produced by different participants.

This letter provides a retrospective analysis of our team’s research performed under the DARPA Explainable Artificial Intelligence (XAI) project. We began by exploring salience maps, English sentences, and lists of feature names for explaining the behavior of deep-learning-based discriminative systems, especially visual question answering systems. We demonstrated limited positive effects from statically presenting explanations along with system answers – for example when teaching people to identify bird species. Many XAI performers were getting better results when users interacted with explanations. This motivated us to evolve the notion of explanation as an interactive medium – usually, between humans and AI systems but sometimes within the software system. We realized that interacting via explanations could enable people to task and adapt ML agents. We added affordances for editing explanations and modified the ML system to act in accordance with the edits to produce an interpretable interface to the agent. Through this interface, editing an explanation can adapt a system’s performance to new, modified purposes. This deep tasking, wherein the agent knows its objective and the explanation for that objective will be critical to enable higher levels of autonomy.

We reflect on the progress in the area of Explainable AI (XAI) Program relative to previous work in the area of intelligent tutoring systems (ITS). A great deal was learned about explanation—and many challenges uncovered—in research that is directly relevant to XAI. We suggest opportunities for future XAI research deriving from ITS methods, as well as the challenges shared by both ITS and XAI in using AI to assist people in solving difficult problems effectively and efficiently.

Neural network architectures are achieving superhuman performance on an expanding range of tasks. To effectively and safely deploy these systems, their decision-making must to be understandable to a wide range of stakeholders. Methods to explain AI have been proposed to answer this challenge, but a lack of theory impedes the development of systematic abstractions which are necessary for cumulative knowledge gains. We propose Bayesian Teaching as a framework for unifying explainable AI (XAI) by integrating machine learning and human learning. Bayesian Teaching formalizes explanation as a communication act of an explainer to shift the beliefs of an explainee. This formalization decomposes any XAI method into four components: (1) the inference to be explained, (2) the explanatory medium, (3) the explainee model, and (4) the explainer model. The abstraction afforded by Bayesian Teaching to decompose any XAI method elucidates the invariances among them. The decomposition of XAI systems enables modular validation, as each of the first three components listed can be tested semi-independently. This decomposition also promotes generalization through recombination of components from different XAI systems, which facilitates the generation of novel variants. These new variants need not be evaluated one by one provided that each component has been validated, leading to an exponential decrease in development time. Finally, by making the goal of explanation explicit, Bayesian Teaching helps developers to assess how suitable an XAI system is for its intended real-world use case. Thus, Bayesian Teaching provides a theoretical framework that encourages systematic, scientific investigation of XAI.

How would you describe the features that a deep learning model composes if you were restricted to measuring observable behaviours? Explainable artificial intelligence (XAI) methods rely on privileged access to model architecture and parameters that is not always feasible for most users, practitioners, and regulators. Inspired by cognitive psychology research on humans, we present a case for measuring response times (RTs) of a forward pass using only the system clock as a technique for XAI. Our method applies to the growing class of models that use input-adaptive dynamic inference and we also extend our approach to standard models that are converted to dynamic inference post hoc. The experimental logic is simple: If the researcher can contrive a stimulus set where variability among input features is tightly controlled, differences in response time for those inputs can be attributed to the way the model composes those features. First, we show that RT is sensitive to difficult, complex features by comparing RTs from ObjectNet and ImageNet. Next, we make specific a priori predictions about RT for abstract features present in the SCEGRAM dataset, where object recognition in humans depends on complex intra-scene object-object relationships. Finally, we show that RT profiles bear specificity for class identity, and therefore the features that define classes. These results cast light on the model’s feature space without opening the black box.

The ability to predict the onset of labour is seen to be an important tool in a clinical setting. Magnetomyography has shown promise in the area of labour imminency prediction, but its clinical application remains limited due to high resource consumption associated with its broad number of channels. In this study, five electrode channels, which account for 3.3% of the total, are used alongside a novel signal decomposition algorithm and low complexity classifiers (logistic regression and linear-SVM) to classify between labour imminency due within 0–48hrs and >48hrs. The results suggest that the parsimonious representation comprising of five electrode channels and novel signal decomposition method alongside the candidate classifiers could allow for greater affordability and hence clinical viability of the magnetomyography-based prediction model, which carries a good degree of model interpretability.