Google at NeurIPS 2022 – Google AI Blog


I Can't Believe It's Not Better: Understanding Deep Learning Through Empirical Falsification Workshop Organizers include: Javier Antorán Panelists include: Kevin Murphy

Interpretable Local Concept-based Explanation with Human Feedback to Predict All-cause Mortality

Journal of Artificial Intelligence Research

Machine learning models are incorporated in different fields and disciplines in which some of them require a high level of accountability and transparency, for example, the healthcare sector. With the General Data Protection Regulation (GDPR), the importance for plausibility and verifiability of the predictions made by machine learning models has become essential. A widely used category of explanation techniques attempts to explain models’ predictions by quantifying the importance score of each input feature. However, summarizing such scores to provide human-interpretable explanations is challenging. Another category of explanation techniques focuses on learning a domain representation in terms of high-level human-understandable concepts and then utilizing them to explain predictions. These explanations are hampered by how concepts are constructed, which is not intrinsically interpretable. To this end, we propose Concept-based Local Explanations with Feedback (CLEF), a novel local model agnostic explanation framework for learning a set of high-level transparent concept definitions in high-dimensional tabular data that uses clinician-labeled concepts rather than raw features. CLEF maps the raw input features to high-level intuitive concepts and then decompose the evidence of prediction of the instance being explained into concepts. In addition, the proposed framework generates counterfactual explanations, suggesting the minimum changes in the instance’s concept based explanation that will lead to a different prediction. We demonstrate with simulated user feedback on predicting the risk of mortality. Such direct feedback is more effective than other techniques, that rely on hand-labelled or automatically extracted concepts, in learning concepts that align with ground truth concept definitions.

Open-Source Toolkits To Develop Responsible AI Systems In 2023


Artificial intelligence is advancing rapidly, with recent estimates by PwC predicting that AI will contribute 13.7 trillion USD to the world economy by 2030. Consequently, there is a need for open-source toolkits that help developers build responsible AI systems. However, it is also critical to understand the implications of developing and deploying AI systems responsibly and ethically. The increased use of artificial intelligence (AI) has led to concerns about its potential impact on society. One way to mitigate these concerns is to develop responsible AI systems that consider the ethical principles of beneficence, non-maleficence, autonomy, and justice. There are several open-source toolkits available that can be used to develop responsible AI systems.

Linear Algebra for AI: NLP and ML Use Cases Simply Explained


Linear algebra is a mathematical discipline concerned with studying vector spaces and linear mappings between them [1]. It is essential in artificial intelligence implementations because it allows for unlocking meanings in high-dimensional data, a common use case pipeline (in AI). Applications for implementation include solving problems across many use cases in AI, including machine learning, deep learning, and natural language processing. Namely, it can be utilized to predict the behavior of neural networks, and it is also being used to improve the accuracy of deep learning models. Further, linear algebra provides a way to understand and visualize high-dimensional data, often used in natural language processing tasks.

Predicting Decisions in Language Based Persuasion Games

Journal of Artificial Intelligence Research

Sender-receiver interactions, and specifically persuasion games, are widely researched in economic modeling and artificial intelligence, and serve as a solid foundation for powerful applications. However, in the classic persuasion games setting, the messages sent from the expert to the decision-maker are abstract or well-structured application-specific signals rather than natural (human) language messages, although natural language is a very common communication signal in real-world persuasion setups. This paper addresses the use of natural language in persuasion games, exploring its impact on the decisions made by the players and aiming to construct effective models for the prediction of these decisions. For this purpose, we conduct an online repeated interaction experiment. At each trial of the interaction, an informed expert aims to sell an uninformed decision-maker a vacation in a hotel, by sending her a review that describes the hotel. While the expert is exposed to several scored reviews, the decision-maker observes only the single review sent by the expert, and her payoff in case she chooses to take the hotel is a random draw from the review score distribution available to the expert only. The expert’s payoff, in turn, depends on the number of times the decision-maker chooses the hotel. We also compare the behavioral patterns in this experiment to the equivalent patterns in similar experiments where the communication is based on the numerical values of the reviews rather than the reviews’ text, and observe substantial differences which can be explained through an equilibrium analysis of the game. We consider a number of modeling approaches for our verbal communication setup, differing from each other in the model type (deep neural network (DNN) vs. linear classifier), the type of features used by the model (textual, behavioral or both) and the source of the textual features (DNN-based vs. hand-crafted). Our results demonstrate that given a prefix of the interaction sequence, our models can predict the future decisions of the decision-maker, particularly when a sequential modeling approach and hand-crafted textual features are applied. Further analysis of the hand-crafted textual features allows us to make initial observations about the aspects of text that drive decision making in our setup.

Visually Grounded Models of Spoken Language: A Survey of Datasets, Architectures and Evaluation Techniques

Journal of Artificial Intelligence Research

This survey provides an overview of the evolution of visually grounded models of spoken language over the last 20 years. Such models are inspired by the observation that when children pick up a language, they rely on a wide range of indirect and noisy clues, crucially including signals from the visual modality co-occurring with spoken utterances. Several fields have made important contributions to this approach to modeling or mimicking the process of learning language: Machine Learning, Natural Language and Speech Processing, Computer Vision and Cognitive Science. The current paper brings together these contributions in order to provide a useful introduction and overview for practitioners in all these areas. We discuss the central research questions addressed, the timeline of developments, and the datasets which enabled much of this work. We then summarize the main modeling architectures and offer an exhaustive overview of the evaluation metrics and analysis techniques.

Can Machines Help Us Answering Question 16 in Datasheets, and In Turn Reflecting on Inappropriate Content? Artificial Intelligence

Large datasets underlying much of current machine learning raise serious issues concerning inappropriate content such as offensive, insulting, threatening, or might otherwise cause anxiety. This calls for increased dataset documentation, e.g., using datasheets. They, among other topics, encourage to reflect on the composition of the datasets. So far, this documentation, however, is done manually and therefore can be tedious and error-prone, especially for large image datasets. Here we ask the arguably "circular" question of whether a machine can help us reflect on inappropriate content, answering Question 16 in Datasheets. To this end, we propose to use the information stored in pre-trained transformer models to assist us in the documentation process. Specifically, prompt-tuning based on a dataset of socio-moral values steers CLIP to identify potentially inappropriate content, therefore reducing human labor. We then document the inappropriate images found using word clouds, based on captions generated using a vision-language model. The documentations of two popular, large-scale computer vision datasets -- ImageNet and OpenImages -- produced this way suggest that machines can indeed help dataset creators to answer Question 16 on inappropriate image content.

Private Adaptive Optimization with Side Information Machine Learning

Adaptive optimization methods have become the default solvers for many machine learning tasks. Unfortunately, the benefits of adaptivity may degrade when training with differential privacy, as the noise added to ensure privacy reduces the effectiveness of the adaptive preconditioner. To this end, we propose AdaDPS, a general framework that uses non-sensitive side information to precondition the gradients, allowing the effective use of adaptive methods in private settings. We formally show AdaDPS reduces the amount of noise needed to achieve similar privacy guarantees, thereby improving optimization performance. Empirically, we leverage simple and readily available side information to explore the performance of AdaDPS in practice, comparing to strong baselines in both centralized and federated settings. Our results show that AdaDPS improves accuracy by 7.7% (absolute) on average -- yielding state-of-the-art privacy-utility trade-offs on large-scale text and image benchmarks.

Latent gaze information in highly dynamic decision-tasks Artificial Intelligence

Digitization is penetrating more and more areas of life. Tasks are increasingly being completed digitally, and are therefore not only fulfilled faster, more efficiently but also more purposefully and successfully. The rapid developments in the field of artificial intelligence in recent years have played a major role in this, as they brought up many helpful approaches to build on. At the same time, the eyes, their movements, and the meaning of these movements are being progressively researched. The combination of these developments has led to exciting approaches. In this dissertation, I present some of these approaches which I worked on during my Ph.D. First, I provide insight into the development of models that use artificial intelligence to connect eye movements with visual expertise. This is demonstrated for two domains or rather groups of people: athletes in decision-making actions and surgeons in arthroscopic procedures. The resulting models can be considered as digital diagnostic models for automatic expertise recognition. Furthermore, I show approaches that investigate the transferability of eye movement patterns to different expertise domains and subsequently, important aspects of techniques for generalization. Finally, I address the temporal detection of confusion based on eye movement data. The results suggest the use of the resulting model as a clock signal for possible digital assistance options in the training of young professionals. An interesting aspect of my research is that I was able to draw on very valuable data from DFB youth elite athletes as well as on long-standing experts in arthroscopy. In particular, the work with the DFB data attracted the interest of radio and print media, namely DeutschlandFunk Nova and SWR DasDing. All resulting articles presented here have been published in internationally renowned journals or at conferences.

Calibrated Learning to Defer with One-vs-All Classifiers Machine Learning

The learning to defer (L2D) framework has the potential to make AI systems safer. For a given input, the system can defer the decision to a human if the human is more likely than the model to take the correct action. We study the calibration of L2D systems, investigating if the probabilities they output are sound. We find that Mozannar & Sontag's (2020) multiclass framework is not calibrated with respect to expert correctness. Moreover, it is not even guaranteed to produce valid probabilities due to its parameterization being degenerate for this purpose. We propose an L2D system based on one-vs-all classifiers that is able to produce calibrated probabilities of expert correctness. Furthermore, our loss function is also a consistent surrogate for multiclass L2D, like Mozannar & Sontag's (2020). Our experiments verify that not only is our system calibrated, but this benefit comes at no cost to accuracy. Our model's accuracy is always comparable (and often superior) to Mozannar & Sontag's (2020) model's in tasks ranging from hate speech detection to galaxy classification to diagnosis of skin lesions.