Search

In practice, training using federated learning can be orders of magnitude slower than standard centralized training. This severely limits the amount of experimentation and tuning that can be done, making it challenging to obtain good performance on a given task. Server-side proxy data can be used to run training simulations, for instance for hyperparameter tuning. This can greatly speed up the training pipeline by reducing the number of tuning runs to be performed overall on the true clients. However, it is challenging to ensure that these simulations accurately reflect the dynamics of the real federated training. In particular, the proxy data used for simulations often comes as a single centralized dataset without a partition into distinct clients, and partitioning this data in a naive way can lead to simulations that poorly reflect real federated training. In this paper we address the challenge of how to partition centralized data in a way that reflects the statistical heterogeneity of the true federated clients. We propose a fully federated, theoretically justified, algorithm that efficiently learns the distribution of the true clients and observe improved server-side simulations when using the inferred distribution to create simulated clients from the centralized data.

We present PeFLL, a new personalized federated learning algorithm that improves over the state-of-the-art in three aspects: 1) it produces more accurate models, especially in the low-data regime, and not only for clients present during its training phase, but also for any that may emerge in the future; 2) it reduces the amount of on-client computation and client-server communication by providing future clients with ready-to-use personalized models that require no additional finetuning or optimization; 3) it comes with theoretical guarantees that establish generalization from the observed clients to future ones. At the core of PeFLL lies a learning-to-learn approach that jointly trains an embedding network and a hypernetwork. The embedding network is used to represent clients in a latent descriptor space in a way that reflects their similarity to each other. The hypernetwork takes as input such descriptors and outputs the parameters of fully personalized client models. In combination, both networks constitute a learning algorithm that achieves state-of-the-art performance in several personalized federated learning benchmarks.

We present Cross-Client Label Propagation(XCLP), a new method for transductive federated learning. XCLP estimates a data graph jointly from the data of multiple clients and computes labels for the unlabeled data by propagating label information across the graph. To avoid clients having to share their data with anyone, XCLP employs two cryptographically secure protocols: secure Hamming distance computation and secure summation. We demonstrate two distinct applications of XCLP within federated learning. In the first, we use it in a one-shot way to predict labels for unseen test points. In the second, we use it to repeatedly pseudo-label unlabeled training data in a federated semi-supervised setting. Experiments on both real federated and standard benchmark datasets show that in both applications XCLP achieves higher classification accuracy than alternative approaches.

There is a need for techniques for efficient and accurate measurement of the impedance of rechargeable batteries at extra-low frequencies (ELFs, typically below 10uHz), as these reflect real usage and cycling patterns, and their importance in fractional battery circuit modeling is becoming increasingly apparent. Major impediments include the time required to perform such measurements, and `drift' in impedance values when measurements are taken from the same battery at different times. Moreover, commercial impedance analyzers are generally unable to measure at frequencies of the order of microhertz. We describe here our use of programmable two-quadrant power supplies to deliver multiple small signal measurement tones in the presence of large signal `working' currents, and our use of these data to generate impedance measurements with good precision and in reasonable time., Comment: 8 pages, 13 figures, IEEE format

Exploration class missions will have communication latency requiring crew members to make decisions more autonomously, with less support from ground personnel. Therefore, new software is needed to provide crew the ability to not only visualize their countermeasures data, but to also derive comprehensive, nuanced, and multidisciplinary insights regarding the health and performance informatics throughout a mission. The Multidisciplinary Analytics, Visualization, and Reporting Interface for Integrated Countermeasures (MAVRIIC) software seeks to establish a centralized approach to countermeasure data visualization that will facilitate a more holistic understanding of crew members' well-being and performance and enable the development of informed and autonomous decision-making support systems aligned with the evolving requirements of Exploration missions. The culmination of MAVRIIC phase 1 (end of FY23) marked the release of a full-stack, cloud-based ground tool displaying visualizations of in-flight exercise data, exercise ground testing, and exercise MEDB reports. Phase 2 of MAVRIIC (FY24) focuses on beginning the expansion of data contents to domains outside of exercise, including functional fitness, sensorimotor, food/nutrition, radiation, bone, and cardio/vision. Phase 3 and phase 4 will comprise enhanced analytics integration and initiation of flight tool development for infusion into the Artemis Program and eventually the Mars Transit Habitat.

Batteries experience capacity offset where available charge depends on the rate at which this charge is drawn. In this work we analyze the capacity offset of a 4.8 A h lithium nickel-cobalt-aluminium oxide battery using an equivalent circuit model of a fractional capacitor in series with a resistor. In this case, the available charge, in theory, becomes infinite in the limit of infinitesimal rate. We show that the fractional properties of the capacitor can be extracted from the charge against rate plot. We then use a network of RC elements to represent the fractional capacitor in order to simulate the data with Matlab. We find that the fractional exponent alpha obtained in this way, 0.971, agrees with that obtained in a more traditional manner from an impedance versus frequency plot, although the fractional capacity does not. Such an approach demonstrates the importance of a fractional description for capacity offset even when an element is nearly a pure capacitor and is valuable for predictions of state-of-charge when low currents are drawn., Comment: 15 pages, 5 figures

We use weights on objects in an abelian category to define what we call a path metric. We introduce three special classes of weight: those compatible with short exact sequences; those induced by their path metric; and those which bound their path metric. We prove that these conditions are in fact equivalent, and call such weights exact. As a special case of a path metric, we obtain a distance for generalized persistence modules whose indexing category is a measure space. We use this distance to define Wasserstein distances, which coincide with the previously defined Wasserstein distances for one-parameter persistence modules. For one-parameter persistence modules, we also describe maps to and from an interval module, and we give a matrix reduction for monomorphisms and epimorphisms., Comment: accepted to Journal of Applied and Computational Topology, 33 pages, v4: made changes suggested by referees

This project will leverage existing capabilities of an augmented reality (AR) sensorimotor assessment tool. Currently, the tool mimics the capabilities of validated 2-dimensional (2D) assessments. While this modality is well-aligned with seated interactions with flight-deck displays, extravehicular activity (EVA) tasks will require full body mobility wherein astronauts reach and bend in a 3-dimensional (3D) space. These types of motions, naturally including head movements, are known to elicit experiences of disorientation. Thus, sensorimotor assessments that include these stimulations and larger movements are important for understanding astronaut readiness prior to initial EVAs. The proposed effort will extend the current 2D eye-hand coordination task in AR to a 3D task designed to evaluate dynamic balance and eye-hand-body coordination. The new module will extend the range of motion and interaction space such that it better aligns with operational task needs for astronaut readiness evaluation.

One of the central notions to emerge from the study of persistent homology is that of interleaving distance. It has found recent applications in symplectic and contact geometry, sheaf theory, computational geometry, and phylogenetics. Here we present a general study of this topic. We define interleaving of functors with common codomain as solutions to an extension problem. In order to define interleaving distance in this setting we are led to categorical generalizations of Hausdorff distance, Gromov-Hausdorff distance, and the space of metric spaces. We obtain comparisons with previous notions of interleaving via the study of future equivalences. As an application we recover a definition of shift equivalences of discrete dynamical systems., Comment: 35 pages, v3: changed title and added references to uses of interleaving (Section 1.3)