The massive use of machine learning models, particularly neural networks, has raised serious concerns about their environmental impact. Indeed, over the last few years we have seen an explosion in the computing costs associated with training and deploying these systems. It is, therefore, crucial to understand their energy requirements in order to better integrate them into the evaluation of models, which has so far focused mainly on performance. In this paper, we study several neural network architectures that are key components of sound event detection systems, using an audio tagging task as an example. We measure the energy consumption for training and testing small to large architectures and establish complex relationships between the energy consumption, the number of floating-point operations, the number of parameters, and the GPU/memory utilization.

This paper presents an adaptive visual servoing framework for robotic on-orbit servicing (OOS), specifically designed for capturing tumbling satellites. The vision-guided robotic system is capable of selecting optimal control actions in the event of partial or complete vision system failure, particularly in the short term. The autonomous system accounts for physical and operational constraints, executing visual servoing tasks to minimize a cost function. A hierarchical control architecture is developed, integrating a variant of the Iterative Closest Point (ICP) algorithm for image registration, a constrained noise-adaptive Kalman filter, fault detection and recovery logic, and a constrained optimal path planner. The dynamic estimator provides real-time estimates of unknown states and uncertain parameters essential for motion prediction, while ensuring consistency through a set of inequality constraints. It also adjusts the Kalman filter parameters adaptively in response to unexpected vision errors. In the event of vision system faults, a recovery strategy is activated, guided by fault detection logic that monitors the visual feedback via the metric fit error of image registration. The estimated/predicted pose and parameters are subsequently fed into an optimal path planner, which directs the robot's end-effector to the target's grasping point. This process is subject to multiple constraints, including acceleration limits, smooth capture, and line-of-sight maintenance with the target. Experimental results demonstrate that the proposed visual servoing system successfully captured a free-floating object, despite complete occlusion of the vision system.

This paper introduces a novel approach to interactive robots by leveraging the form-factor of cards to create thin robots equipped with vibrational capabilities for locomotion and haptic feedback. The system is composed of flat-shaped robots with on-device sensing and wireless control, which offer lightweight portability and scalability. This research introduces a hardware prototype. Applications include augmented card playing, educational tools, and assistive technology, which showcase CARDinality's versatility in tangible interaction.

Reliability analysis is a formidable task, particularly in systems with a large number of stochastic parameters. Conventional methods for quantifying reliability often rely on extensive simulations or experimental data, which can be costly and time-consuming, especially when dealing with systems governed by complex physical laws which necessitates computationally intensive numerical methods such as finite element or finite volume techniques. On the other hand, surrogate-based methods offer an efficient alternative for computing reliability by approximating the underlying model from limited data. Neural operators have recently emerged as effective surrogates for modelling physical systems governed by partial differential equations. These operators can learn solutions to PDEs for varying inputs and parameters. Here, we investigate the efficacy of the recently developed physics-informed wavelet neural operator in solving reliability analysis problems. In particular, we investigate the possibility of using physics-informed operator for solving high-dimensional reliability analysis problems, while bypassing the need for any simulation. Through four numerical examples, we illustrate that physics-informed operator can seamlessly solve high-dimensional reliability analysis problems with reasonable accuracy, while eliminating the need for running expensive simulations.

The uncertainty inherent in the environmental transition model of Reinforcement Learning (RL) necessitates a careful balance between exploration and exploitation to optimize the use of computational resources for accurately estimating an agent's expected reward. Achieving balance in control systems is particularly challenging in scenarios with sparse rewards. However, given the extensive prior knowledge available for many environments, it is redundant to begin learning from scratch in such settings. To address this, we introduce \textbf{L}anguage \textbf{M}odel \textbf{G}uided \textbf{T}rade-offs (i.e., \textbf{LMGT}), a novel, sample-efficient framework that leverages the comprehensive prior knowledge embedded in Large Language Models (LLMs) and their adeptness at processing non-standard data forms, such as wiki tutorials. LMGT proficiently manages the exploration-exploitation trade-off by employing reward shifts guided by LLMs, which direct agents' exploration endeavors, thereby improving sample efficiency. We have thoroughly tested LMGT across various RL tasks and deployed it in industrial-grade RL recommendation systems, where it consistently outperforms baseline methods. The results indicate that our framework can significantly reduce the time cost required during the training phase in RL.

Fuel cells using oxygen and glucose could power microscopic robots operating in blood vessels. Swarms of such robots can significantly reduce oxygen concentration, depending on the time between successive transits of the lung, hematocrit variation in vessels and tissue oxygen consumption. These factors differ among circulation paths through the body. This paper evaluates how these variations affect the minimum oxygen concentration due to robot consumption and where it occurs: mainly in moderate-sized veins toward the end of long paths prior to their merging with veins from shorter paths. This shows that tens of billions of robots can obtain hundreds of picowatts throughout the body with minor reduction in total oxygen. However, a trillion robots significantly deplete oxygen in some parts of the body. By storing oxygen or limiting their consumption in long circulation paths, robots can actively mitigate this depletion. The variation in behavior is illustrated in three cases: the portal system which involves passage through two capillary networks, the spleen whose slits significantly slow some of the flow, and large tissue consumption in coronary circulation.

Modern machine-learning techniques are generally considered data-hungry. However, this may not be the case for turbulence as each of its snapshots can hold more information than a single data file in general machine-learning applications. This study asks the question of whether nonlinear machine-learning techniques can effectively extract physical insights even from as little as a single snapshot of a turbulent vortical flow. As an example, we consider machine-learning-based super-resolution analysis that reconstructs a high-resolution field from low-resolution data for two-dimensional decaying turbulence. We reveal that a carefully designed machine-learning model trained with flow tiles sampled from only a single snapshot can reconstruct vortical structures across a range of Reynolds numbers. Successful flow reconstruction indicates that nonlinear machine-learning techniques can leverage scale-invariance properties to learn turbulent flows. We further show that training data of turbulent flows can be cleverly collected from a single snapshot by considering characteristics of rotation and shear tensors. The present findings suggest that embedding prior knowledge in designing a model and collecting data is important for a range of data-driven analyses for turbulent flows. More broadly, this work hopes to stop machine-learning practitioners from being wasteful with turbulent flow data.

In this work, we present a novel method for extensive multi-scale generative terrain modeling. At the core of our model is a cascade of superresolution diffusion models that can be combined to produce consistent images across multiple resolutions. Pairing this concept with a tiled generation method yields a scalable system that can generate thousands of square kilometers of realistic Earth surfaces at high resolution. We evaluate our method on a dataset collected from Bing Maps and show that it outperforms super-resolution baselines on the extreme super-resolution task of 1024x zoom. We also demonstrate its ability to create diverse and coherent scenes via an interactive gigapixel-scale generated map. Finally, we demonstrate how our system can be extended to enable novel content creation applications including controllable world generation and 3D scene generation.

Efficient load balancing is crucial in cloud computing environments to ensure optimal resource utilization, minimize response times, and prevent server overload. Traditional load balancing algorithms, such as round-robin or least connections, are often static and unable to adapt to the dynamic and fluctuating nature of cloud workloads. In this paper, we propose a novel adaptive load balancing framework using Reinforcement Learning (RL) to address these challenges. The RL-based approach continuously learns and improves the distribution of tasks by observing real-time system performance and making decisions based on traffic patterns and resource availability. Our framework is designed to dynamically reallocate tasks to minimize latency and ensure balanced resource usage across servers. Experimental results show that the proposed RL-based load balancer outperforms traditional algorithms in terms of response time, resource utilization, and adaptability to changing workloads. These findings highlight the potential of AI-driven solutions for enhancing the efficiency and scalability of cloud infrastructures.

This paper compared physics-informed neural network (PINN), conventional neural network (NN) and traditional numerical discretization methods on solving differential equations (DEs) through literature investigation and experimental validation. We focused on the soft-constrained PINN approach and formalized its mathematical framework and computational flow for solving Ordinary DEs and Partial DEs (ODEs/PDEs). The working mechanism and its accuracy and efficiency were experimentally verified by solving typical linear and non-linear oscillator ODEs. We demonstrate that the DeepXDE-based implementation of PINN is not only light code and efficient in training, but also flexible across CPU/GPU platforms. PINN greatly reduces the need for labeled data: when the nonlinearity of the ODE is weak, a very small amount of supervised training data plus a few unsupervised collocation points are sufficient to predict the solution; in the minimalist case, only one or two training points (with initial values) are needed for first- or second-order ODEs, respectively. We also find that, with the aid of collocation points and the use of physical information, PINN has the ability to extrapolate data outside the time domain of the training set, and especially is robust to noisy data, thus with enhanced generalization capabilities. Training is accelerated when the gains obtained along with the reduction in the amount of data outweigh the delay caused by the increase in the loss function terms. The soft-constrained PINN can easily impose a physical law (e.g., conservation of energy) constraint by adding a regularization term to the total loss function, thus improving the solution performance to ODEs that obey this physical law. Furthermore, PINN can also be used for stiff ODEs, PDEs, and other types of DEs, and is becoming a favorable catalyst for the era of Digital Twins.