The Game Theory & Multi-Agent team at DeepMind studies several aspects of multi-agent learning ranging from computing approximations to fundamental concepts in game theory to simulating social dilemmas in rich spatial environments and training 3-d humanoids in difficult team coordination tasks. A signature aim of our group is to use the resources and expertise made available to us at DeepMind in deep reinforcement learning to explore multi-agent systems in complex environments and use these benchmarks to advance our understanding. Here, we summarise the recent work of our team and present a taxonomy that we feel highlights many important open challenges in multi-agent research.

What will it take for a car to be able to drive itself anywhere a human can? Ask autonomous vehicle experts this question and the answer invariably includes a discussion of geofencing. In the broadest sense, geofencing is simply a virtual boundary around a physical area. In the world of self-driving cars, it describes a crucial subset of the operational design domain -- the geographic region where the vehicle is functional. Reaching full Level 5 autonomy means removing the "fence" from geofenced autonomous cars. Experts say that will require artificial intelligence that can make abstractions, inferences, and become smarter as it is being used.

This entirely absorbing movie from Korean-American director Kogonada is adapted from a short story by Alexander Weinstein in his collection Children of the New World; it floats on a Zen updraft of wisdom and ideas. After Yang is an enigmatic sci-fi drama about a family of the future whose AI robot child (a "techno-sapien") bought by the parents to be a kindly big brother to their adopted Chinese daughter, goes wrong and cannot be fixed. There are touches of Philip K Dick and even Charlie Kaufman, and this is also a pregnant meditation on grief, loss, memory and consciousness. Colin Farrell and Jodie Turner-Smith play Jake and Kyra, who have an adopted daughter Mika (Malea Emma Tjandrawidjaja) – and also AI bot Yang (Justin H Min), a happy, calm, friendly and very lovable figure of Chinese appearance, programmed with Chinese memories and Chinese knowledge, whom the family has grown to love. When Yang goes into an unexplained coma, it is deeply upsetting: a bereavement-shock far beyond the malfunction of some gadget.

Originally published on Towards AI the World's Leading AI and Technology News and Media Company. If you are building an AI-related product or service, we invite you to consider becoming an AI sponsor. At Towards AI, we help scale AI and technology startups. Let us help you unleash your technology to the masses. One of the major components of the economy is the insurance industry, it protects against the financial risks that are present at all stages of people's lives and businesses.

Split learning and inference propose to run training/inference of a large model that is split across client devices and the cloud. However, such a model splitting imposes privacy concerns, because the activation flowing through the split layer may leak information about the clients' private input data. There is currently no good way to quantify how much private information is being leaked through the split layer, nor a good way to improve privacy up to the desired level. In this work, we propose to use Fisher information as a privacy metric to measure and control the information leakage. We show that Fisher information can provide an intuitive understanding of how much private information is leaking through the split layer, in the form of an error bound for an unbiased reconstruction attacker. We then propose a privacy-enhancing technique, ReFIL, that can enforce a user-desired level of Fisher information leakage at the split layer to achieve high privacy, while maintaining reasonable utility.

Cross-domain alignment plays a key role in tasks ranging from image-text retrieval to machine translation. The main objective is to associate related entities across different domains. Recently, purely unsupervised methods operating on monolingual embeddings have successfully been used to infer a bilingual lexicon without relying on supervision. However, current state-of-the art methods only focus on point vectors although distributional embeddings have proven to embed richer semantic information when representing words. This paper investigates a novel stochastic optimization approach for aligning word distributional embeddings. Our method builds upon techniques in optimal transport to resolve the cross-domain matching problem in a principled manner. We evaluate our method on the problem of unsupervised word translation, by aligning word embeddings trained on monolingual data. We present empirical evidence to demonstrate the validity of our approach to the bilingual lexicon induction task across several language pairs.

Structural decomposition methods, such as generalized hypertree decompositions, have been successfully used for solving constraint satisfaction problems (CSPs). As decompositions can be reused to solve CSPs with the same constraint scopes, investing resources in computing good decompositions is beneficial, even though the computation itself is hard. Unfortunately, current methods need to compute a completely new decomposition even if the scopes change only slightly. In this paper, we make the first steps toward solving the problem of updating the decomposition of a CSP $P$ so that it becomes a valid decomposition of a new CSP $P'$ produced by some modification of $P$. Even though the problem is hard in theory, we propose and implement a framework for effectively updating GHDs. The experimental evaluation of our algorithm strongly suggests practical applicability.

Clustered federated learning (FL) has been shown to produce promising results by grouping clients into clusters. This is especially effective in scenarios where separate groups of clients have significant differences in the distributions of their local data. Existing clustered FL algorithms are essentially trying to group together clients with similar distributions so that clients in the same cluster can leverage each other's data to better perform federated learning. However, prior clustered FL algorithms attempt to learn these distribution similarities indirectly during training, which can be quite time consuming as many rounds of federated learning may be required until the formation of clusters is stabilized. In this paper, we propose a new approach to federated learning that directly aims to efficiently identify distribution similarities among clients by analyzing the principal angles between the client data subspaces. Each client applies a truncated singular value decomposition (SVD) step on its local data in a single-shot manner to derive a small set of principal vectors, which provides a signature that succinctly captures the main characteristics of the underlying distribution. This small set of principal vectors is provided to the server so that the server can directly identify distribution similarities among the clients to form clusters. This is achieved by comparing the similarities of the principal angles between the client data subspaces spanned by those principal vectors. The approach provides a simple, yet effective clustered FL framework that addresses a broad range of data heterogeneity issues beyond simpler forms of Non-IIDness like label skews. Our clustered FL approach also enables convergence guarantees for non-convex objectives. Our code is available at https://github.com/MMorafah/PACFL.

Most of the popular Big Data analytics tools evolved to adapt their working environment to extract valuable information from a vast amount of unstructured data. The ability of data mining techniques to filter this helpful information from Big Data led to the term Big Data Mining. Shifting the scope of data from small-size, structured, and stable data to huge volume, unstructured, and quickly changing data brings many data management challenges. Different tools cope with these challenges in their own way due to their architectural limitations. There are numerous parameters to take into consideration when choosing the right data management framework based on the task at hand. In this paper, we present a comprehensive benchmark for two widely used Big Data analytics tools, namely Apache Spark and Hadoop MapReduce, on a common data mining task, i.e., classification. We employ several evaluation metrics to compare the performance of the benchmarked frameworks, such as execution time, accuracy, and scalability. These metrics are specialized to measure the performance for classification task. To the best of our knowledge, there is no previous study in the literature that employs all these metrics while taking into consideration task-specific concerns. We show that Spark is 5 times faster than MapReduce on training the model. Nevertheless, the performance of Spark degrades when the input workload gets larger. Scaling the environment by additional clusters significantly improves the performance of Spark. However, similar enhancement is not observed in Hadoop. Machine learning utility of MapReduce tend to have better accuracy scores than that of Spark, like around 3%, even in small size data sets.

This work explores the physics-driven machine learning technique Operator Inference (OpInf) for predicting the state of chaotic dynamical systems. OpInf provides a non-intrusive approach to infer approximations of polynomial operators in reduced space without having access to the full order operators appearing in discretized models. Datasets for the physics systems are generated using conventional numerical solvers and then projected to a low-dimensional space via Principal Component Analysis (PCA). In latent space, a least-squares problem is set to fit a quadratic polynomial operator, which is subsequently employed in a time-integration scheme in order to produce extrapolations in the same space. Once solved, the inverse PCA operation is applied to reconstruct the extrapolations in the original space. The quality of the OpInf predictions is assessed via the Normalized Root Mean Squared Error (NRMSE) metric from which the Valid Prediction Time (VPT) is computed. Numerical experiments considering the chaotic systems Lorenz 96 and the Kuramoto-Sivashinsky equation show promising forecasting capabilities of the OpInf reduced order models with VPT ranges that outperform state-of-the-art machine learning methods such as backpropagation and reservoir computing recurrent neural networks [1], as well as Markov neural operators [2].