Montreal, March 20, 2023 – Mila – Quebec Artificial Intelligence Institute and the United Nations Educational, Scientific and Cultural Organization (UNESCO) today unveiled at Mila a joint book on the urgent need for a better governance of artificial intelligence (AI) in the face of unprecedented technological change. The book, Missing Links in AI Governance, includes 18 articles on AI governance written by academics, civil society representatives, innovators and policy makers at a time when technological revolutions provide new scientific, economic and social opportunities while raising ethical questions and posing regulatory challenges. The book explores themes such as the influence of AI on indigenous and LGBTI communities, the necessary inclusion of Southern countries in global governance or the use of AI to support innovation for socially beneficial purposes. It maps out possible solutions to foster an AI development that is ethical, inclusive, and respectful of human rights. The authors also warn against the use of AI in potentially harmful contexts like autonomous weapons or the manipulation of digital content for social destabilization, deplore the increasing centralization of decision-making power in the development of AI systems and biases embedded in them, and the lack of transparency and accountability in the industry.

Foundation models (e.g. ChatGPT, StableDiffusion) pervasively influence society, warranting immediate social attention. While the models themselves garner much attention, to accurately characterize their impact, we must consider the broader sociotechnical ecosystem. We propose Ecosystem Graphs as a documentation framework to transparently centralize knowledge of this ecosystem. Ecosystem Graphs is composed of assets (datasets, models, applications) linked together by dependencies that indicate technical (e.g. how Bing relies on GPT-4) and social (e.g. how Microsoft relies on OpenAI) relationships. To supplement the graph structure, each asset is further enriched with fine-grained metadata (e.g. the license or training emissions). We document the ecosystem extensively at https://crfm.stanford.edu/ecosystem-graphs/. As of March 16, 2023, we annotate 262 assets (64 datasets, 128 models, 70 applications) from 63 organizations linked by 356 dependencies. We show Ecosystem Graphs functions as a powerful abstraction and interface for achieving the minimum transparency required to address myriad use cases. Therefore, we envision Ecosystem Graphs will be a community-maintained resource that provides value to stakeholders spanning AI researchers, industry professionals, social scientists, auditors and policymakers.

As we enter the exascale computing era, efficiently utilizing power and optimizing the performance of scientific applications under power and energy constraints has become critical and challenging. We propose a low-overhead autotuning framework to autotune performance and energy for various hybrid MPI/OpenMP scientific applications at large scales and to explore the tradeoffs between application runtime and power/energy for energy efficient application execution, then use this framework to autotune four ECP proxy applications -- XSBench, AMG, SWFFT, and SW4lite. Our approach uses Bayesian optimization with a Random Forest surrogate model to effectively search parameter spaces with up to 6 million different configurations on two large-scale production systems, Theta at Argonne National Laboratory and Summit at Oak Ridge National Laboratory. The experimental results show that our autotuning framework at large scales has low overhead and achieves good scalability. Using the proposed autotuning framework to identify the best configurations, we achieve up to 91.59% performance improvement, up to 21.2% energy savings, and up to 37.84% EDP improvement on up to 4,096 nodes.

Vision Transformers (ViTs) emerge to achieve impressive performance on many data-abundant computer vision tasks by capturing long-range dependencies among local features. However, under few-shot learning (FSL) settings on small datasets with only a few labeled data, ViT tends to overfit and suffers from severe performance degradation due to its absence of CNN-alike inductive bias. Previous works in FSL avoid such problem either through the help of self-supervised auxiliary losses, or through the dextile uses of label information under supervised settings. But the gap between self-supervised and supervised few-shot Transformers is still unfilled. Inspired by recent advances in self-supervised knowledge distillation and masked image modeling (MIM), we propose a novel Supervised Masked Knowledge Distillation model (SMKD) for few-shot Transformers which incorporates label information into self-distillation frameworks. Compared with previous self-supervised methods, we allow intra-class knowledge distillation on both class and patch tokens, and introduce the challenging task of masked patch tokens reconstruction across intra-class images. Experimental results on four few-shot classification benchmark datasets show that our method with simple design outperforms previous methods by a large margin and achieves a new start-of-the-art. Detailed ablation studies confirm the effectiveness of each component of our model. Code for this paper is available here: https://github.com/HL-hanlin/SMKD.

Post-processing typically takes the outputs of a Numerical Weather Prediction (NWP) model and applies linear statistical techniques to produce improve localized forecasts, by including additional observations, or determining systematic errors at a finer scale. In this pilot study, we investigate the benefits and challenges of using non-linear neural network (NN) based methods to post-process multiple weather features -- temperature, moisture, wind, geopotential height, precipitable water -- at 30 vertical levels, globally and at lead times up to 7 days. We show that we can achieve accuracy improvements of up to 12% (RMSE) in a field such as temperature at 850hPa for a 7 day forecast. However, we recognize the need to strengthen foundational work on objectively measuring a sharp and correct forecast. We discuss the challenges of using standard metrics such as root mean squared error (RMSE) or anomaly correlation coefficient (ACC) as we move from linear statistical models to more complex non-linear machine learning approaches for post-processing global weather forecasts.

This article proposes a partial solution to these three issues, in the form of controlled generation of synthetic text with large language models. I provide a conceptual overview of text generation, guidance on when researchers should prefer different techniques for generating synthetic text, a discussion of ethics, and a simple technique for improving the quality of synthetic text. I demonstrate the usefulness of synthetic text with three applications: generating synthetic tweets describing the fighting in Ukraine, synthetic news articles describing specified political events for training an event detection system, and a multilingual corpus of populist manifesto statements for training a sentence-level populism classifier.

In order for artificial agents to successfully perform tasks in changing environments, they must be able to both detect and adapt to novelty. However, visual novelty detection research often only evaluates on repurposed datasets such as CIFAR-10 originally intended for object classification, where images focus on one distinct, well-centered object. New benchmarks are needed to represent the challenges of navigating the complex scenes of an open world. Our new NovelCraft dataset contains multimodal episodic data of the images and symbolic world-states seen by an agent completing a pogo stick assembly task within a modified Minecraft environment. In some episodes, we insert novel objects of varying size within the complex 3D scene that may impact gameplay. Our visual novelty detection benchmark finds that methods that rank best on popular area-under-the-curve metrics may be outperformed by simpler alternatives when controlling false positives matters most. Further multimodal novelty detection experiments suggest that methods that fuse both visual and symbolic information can improve time until detection as well as overall discrimination. Finally, our evaluation of recent generalized category discovery methods suggests that adapting to new imbalanced categories in complex scenes remains an exciting open problem.

With growing security and privacy concerns in the Smart Grid domain, intrusion detection on critical energy infrastructure has become a high priority in recent years. To remedy the challenges of privacy preservation and decentralized power zones with strategic data owners, Federated Learning (FL) has contemporarily surfaced as a viable privacy-preserving alternative which enables collaborative training of attack detection models without requiring the sharing of raw data. To address some of the technical challenges associated with conventional synchronous FL, this paper proposes FeDiSa, a novel Semi-asynchronous Federated learning framework for power system faults and cyberattack Discrimination which takes into account communication latency and stragglers. Specifically, we propose a collaborative training of deep auto-encoder by Supervisory Control and Data Acquisition sub-systems which upload their local model updates to a control centre, which then perform a semi-asynchronous model aggregation for a new global model parameters based on a buffer system and a preset cut-off time. Experiments on the proposed framework using publicly available industrial control systems datasets reveal superior attack detection accuracy whilst preserving data confidentiality and minimizing the adverse effects of communication latency and stragglers. Furthermore, we see a 35% improvement in training time, thus validating the robustness of our proposed method.

In nuclear isotope and chemistry laboratories, hot cells and gloveboxes provide scientists with a controlled and safe environment to perform experiments. Working on experiments in these isolated containment cells requires scientists to be physically present. For hot cell work today, scientists manipulate equipment and radioactive material inside through a bilateral mechanical control mechanism. Motions produced outside the cell with the master control levers are mechanically transferred to the internal grippers inside the shielded containment cell. There is a growing need to have the capability to conduct experiments within these cells remotely. A simple method to enable remote manipulations within hot cell and glovebox cells is to mount two robotic arms inside a box to mimic the motions of human hands. An AR application was built in this work to allow a user wearing a Microsoft HoloLens 2 headset to teleoperate dual arm manipulators by grasping robotic end-effector digital replicas in AR from a remote location. In addition to the real-time replica of the physical robotic arms in AR, the application enables users to view a live video stream attached to the robotic arms and parse a 3D point cloud of 3D objects in their remote AR environment for better situational awareness. This work also provides users with virtual fixture to assist in manipulation and other teleoperation tasks.

Automated decision-making systems, especially those based on natural language processing, are pervasive in our lives. They are not only behind the internet search engines we use daily, but also take more critical roles: selecting candidates for a job, determining suspects of a crime, diagnosing autism and more. Such automated systems make errors, which may be harmful in many ways, be it because of the severity of the consequences (as in health issues) or because of the sheer number of people they affect. When errors made by an automated system affect a population more than others, we call the system \textit{biased}. Most modern natural language technologies are based on artifacts obtained from enormous volumes of text using machine learning, namely language models and word embeddings. Since they are created by applying subsymbolic machine learning, mostly artificial neural networks, they are opaque and practically uninterpretable by direct inspection, thus making it very difficult to audit them. In this paper, we present a methodology that spells out how social scientists, domain experts, and machine learning experts can collaboratively explore biases and harmful stereotypes in word embeddings and large language models. Our methodology is based on the following principles: * focus on the linguistic manifestations of discrimination on word embeddings and language models, not on the mathematical properties of the models * reduce the technical barrier for discrimination experts%, be it social scientists, domain experts or other * characterize through a qualitative exploratory process in addition to a metric-based approach * address mitigation as part of the training process, not as an afterthought