Human Activity Recognition (HAR) based on inertial data is an increasingly diffused task on embedded devices, from smartphones to ultra low-power sensors. Due to the high computational complexity of deep learning models, most embedded HAR systems are based on simple and not-so-accurate classic machine learning algorithms. This work bridges the gap between on-device HAR and deep learning, proposing a set of efficient one-dimensional Convolutional Neural Networks (CNNs) deployable on general purpose microcontrollers (MCUs). Our CNNs are obtained combining hyper-parameters optimization with sub-byte and mixed-precision quantization, to find good trade-offs between classification results and memory occupation. Moreover, we also leverage adaptive inference as an orthogonal optimization to tune the inference complexity at runtime based on the processed input, hence producing a more flexible HAR system. With experiments on four datasets, and targeting an ultra-low-power RISC-V MCU, we show that (i) We are able to obtain a rich set of Pareto-optimal CNNs for HAR, spanning more than 1 order of magnitude in terms of memory, latency and energy consumption; (ii) Thanks to adaptive inference, we can derive >20 runtime operating modes starting from a single CNN, differing by up to 10% in classification scores and by more than 3x in inference complexity, with a limited memory overhead; (iii) on three of the four benchmarks, we outperform all previous deep learning methods, reducing the memory occupation by more than 100x. The few methods that obtain better performance (both shallow and deep) are not compatible with MCU deployment. (iv) All our CNNs are compatible with real-time on-device HAR with an inference latency <16ms. Their memory occupation varies in 0.05-23.17 kB, and their energy consumption in 0.005 and 61.59 uJ, allowing years of continuous operation on a small battery supply.

Modern Internet of Things (IoT) environments are monitored via a large number of IoT enabled sensing devices, with the data acquisition and processing infrastructure setting restrictions in terms of computational power and energy resources. To alleviate this issue, sensors are often configured to operate at relatively low sampling frequencies, yielding a reduced set of observations. Nevertheless, this can hamper dramatically subsequent decision-making, such as forecasting. To address this problem, in this work we evaluate short-term forecasting in highly underdetermined cases, i.e., the number of sensor streams is much higher than the number of observations. Several statistical, machine learning and neural network-based models are thoroughly examined with respect to the resulting forecasting accuracy on five different real-world datasets. The focus is given on a unified experimental protocol especially designed for short-term prediction of multiple time series at the IoT edge. The proposed framework can be considered as an important step towards establishing a solid forecasting strategy in resource constrained IoT applications.

Satellite remote sensing presents a cost-effective solution for synoptic flood monitoring, and satellite-derived flood maps provide a computationally efficient alternative to numerical flood inundation models traditionally used. While satellites do offer timely inundation information when they happen to cover an ongoing flood event, they are limited by their spatiotemporal resolution in terms of their ability to dynamically monitor flood evolution at various scales. Constantly improving access to new satellite data sources as well as big data processing capabilities has unlocked an unprecedented number of possibilities in terms of data-driven solutions to this problem. Specifically, the fusion of data from satellites, such as the Copernicus Sentinels, which have high spatial and low temporal resolution, with data from NASA SMAP and GPM missions, which have low spatial but high temporal resolutions could yield high-resolution flood inundation at a daily scale. Here a Convolutional-Neural-Network is trained using flood inundation maps derived from Sentinel-1 Synthetic Aperture Radar and various hydrological, topographical, and land-use based predictors for the first time, to predict high-resolution probabilistic maps of flood inundation. The performance of UNet and SegNet model architectures for this task is evaluated, using flood masks derived from Sentinel-1 and Sentinel-2, separately with 95 percent-confidence intervals. The Area under the Curve (AUC) of the Precision Recall Curve (PR-AUC) is used as the main evaluation metric, due to the inherently imbalanced nature of classes in a binary flood mapping problem, with the best model delivering a PR-AUC of 0.85.

The ability to discover new transient candidates via image differencing without direct human intervention is an important task in observational astronomy. For these kind of image classification problems, machine Learning techniques such as Convolutional Neural Networks (CNNs) have shown remarkable success. In this work, we present the results of an automated transient candidate identification on images with CNNs for an extant dataset from the Dark Energy Survey Supernova program (DES-SN), whose main focus was on using Type Ia supernovae for cosmology. By performing an architecture search of CNNs, we identify networks that efficiently select non-artifacts (e.g. The CNNs also help us identify a subset of mislabeled images. Performing a relabeling of the images in this subset, the resulting classification with CNNs is significantly better than previous results, lowering the false positive rate by 27% at a fixed missed detection rate of 0.05. INTRODUCTION A major aspect of observational astronomy is the "survey" which involves the wholesale mapping of various regions of the sky to create catalogs which are subsequently mined for scientifically important astronomical objects. We refer to a transient candidate as the detection on a single image of a new or varying source with respect to a previously taken reference image, regardless of its astrophysical nature since at this stage its classification is unknown and will remain so until further data is taken (spectroscopy and/or additional photometry). Some examples of such transient candidates are solar system objects, supernovae, active galactic nuclei, variable stars, and neutron star mergers, etc. Since some of these events are quite rare and will fade rapidly, it is often important to trigger follow-up observations immediately to glean their underlying nature and discover new physics. Hence, identifying transient candidates in images quickly and efficiently is very important so as not to waste precious, and expensive, follow-up resources. For many years this process was conducted by manual inspection of images by humans.

Team's algorithm could lead to pivotal discovery of new materials for solar cells. It will be doing so for billions more years. Yet, we have only just begun tapping into that abundant, renewable source of energy at affordable cost. Solar absorbers are a material used to convert this energy into heat or electricity. Maria Chan, a scientist in the U.S. Department of Energy's (DOE) Argonne National Laboratory, has developed a machine learning method for screening many thousands of compounds as solar absorbers.

Bangladesh and India were hit in June this year by one of the worst floods ever seen in a century as a result of climate change, leaving millions in need of aid. Rising temperatures are also causing heatwaves, droughts, wildfires, floods and other catastrophic events worldwide, while underprivileged regions are paying the biggest price for the world's failure so far to reduce carbon emissions. Governments should not need further wake-up calls to act swiftly and cap global warming to 1.5 C above pre-industrial level. The Intergovernmental Panel on Climate Change report Climate Change 2022: Mitigation of climate change, which was released in April, concludes that "without immediate and deep emissions reductions across all sectors, limiting global warming to 1.5 C is beyond reach"1. According to the report, in order to reach the goal, global greenhouse gas emissions should peak before 2025 and decrease by 43% by 2030.

Stochastic planning can be reduced to probabilistic inference in large discrete graphical models, but hardness of inference requires approximation schemes to be used. In this paper we argue that such applications can be disentangled along two dimensions. The first is the direction of information flow in the idealized exact optimization objective, i.e., forward vs backward inference. The second is the type of approximation used to compute this objective, e.g., Belief Propagation (BP) vs mean field variational inference (MFVI). This new categorization allows us to unify a large amount of isolated efforts in prior work explaining their connections and differences as well as potential improvements. An extensive experimental evaluation over large stochastic planning problems shows the advantage of forward BP over several algorithms based on MFVI. An analysis of practical limitations of MFVI motivates a novel algorithm, collapsed state variational inference (CSVI), which provides a tighter approximation and achieves comparable planning performance with forward BP.

Equilibrium propagation (EP) is an alternative to backpropagation (BP) that allows the training of deep neural networks with local learning rules. It thus provides a compelling framework for training neuromorphic systems and understanding learning in neurobiology. However, EP requires infinitesimal teaching signals, thereby limiting its applicability in noisy physical systems. Moreover, the algorithm requires separate temporal phases and has not been applied to large-scale problems. Here we address these issues by extending EP to holomorphic networks. We show analytically that this extension naturally leads to exact gradients even for finite-amplitude teaching signals. Importantly, the gradient can be computed as the first Fourier coefficient from finite neuronal activity oscillations in continuous time without requiring separate phases. Further, we demonstrate in numerical simulations that our approach permits robust estimation of gradients in the presence of noise and that deeper models benefit from the finite teaching signals. Finally, we establish the first benchmark for EP on the ImageNet 32x32 dataset and show that it matches the performance of an equivalent network trained with BP. Our work provides analytical insights that enable scaling EP to large-scale problems and establishes a formal framework for how oscillations could support learning in biological and neuromorphic systems.

Robots and intelligent systems that sense or interact with the world are increasingly being used to automate a wide array of tasks. The ability of these systems to complete these tasks depends on a large range of technologies such as the mechanical and electrical parts that make up the physical body of the robot and its sensors, perception algorithms to perceive the environment, and planning and control algorithms to produce meaningful actions. Therefore, it is often necessary to consider the interactions between these components when designing an embodied system. This thesis explores work on the task-driven co-optimization of robotics systems in an end-to-end manner, simultaneously optimizing the physical components of the system with inference or control algorithms directly for task performance. We start by considering the problem of optimizing a beacon-based localization system directly for localization accuracy. Designing such a system involves placing beacons throughout the environment and inferring location from sensor readings. In our work, we develop a deep learning approach to optimize both beacon placement and location inference directly for localization accuracy. We then turn our attention to the related problem of task-driven optimization of robots and their controllers. In our work, we start by proposing a data-efficient algorithm based on multi-task reinforcement learning. Our approach efficiently optimizes both physical design and control parameters directly for task performance by leveraging a design-conditioned controller capable of generalizing over the space of physical designs. We then follow this up with an extension to allow for the optimization over discrete morphological parameters such as the number and configuration of limbs. Finally, we conclude by exploring the fabrication and deployment of optimized soft robots.

In this work, we propose a framework for energy-efficient trajectory design of an unmanned aerial vehicle (UAV)-based portable access point (PAP) deployed to serve a set of ground nodes (GNs). In addition to the PAP and GNs, the system consists of a set of intelligent reflecting surfaces (IRSs) mounted on man-made structures to increase the number of bits transmitted per Joule of energy consumed measured as the global energy efficiency (GEE). The GEE trajectory for the PAP is designed by considering the UAV propulsion energy consumption and the Peukert effect of the PAP battery, which represents an accurate battery discharge profile as a non-linear function of the UAV power consumption profile. The GEE trajectory design problem is solved in two phases: in the first, a path for the PAP and feasible positions for the IRS modules are found using a multi-tier circle packing method, and the required IRS phase shift values are calculated using an alternate optimization method that considers the interdependence between the amplitude and phase responses of an IRS element; in the second phase, the PAP flying velocity and user scheduling are calculated using a novel multilap trajectory design algorithm. Numerical evaluations show that: neglecting the Peukert effect overestimates the available flight time of the PAP; after a certain threshold, increasing the battery size reduces the available flight time of the PAP; the presence of IRS modules improves the GEE of the system compared to other baseline scenarios; the multi-lap trajectory saves more energy compared to a single-lap trajectory developed using a combination of sequential convex programming and Dinkelbach algorithm.