This paper presents the conception, design and real miniature instrument implementation of a dual-purpose sensor for small satellites that can act as a star tracker and space debris detector. In the previous research work, the authors conceived, designed and implemented a breadboard consisting of a computer laptop, a camera interface and camera controller, an image sensor, an optics system, a temperature sensor and a temperature controller. It showed that the instrument was feasible. In this paper, a new real star tracker miniature instrument is designed, physically realized and tested. The implementation follows a New Space approach; it is made with Commercial Off-the-Shelf (COTS) components with space heritage. The instrument’s development, implementation and testing are presented.

We present BAE, a problem-tailored and noise-aware Bayesian algorithm for quantum amplitude estimation. In a fault tolerant scenario, BAE is capable of saturating the Heisenberg limit; if device noise is present, BAE can dynamically characterize it and self-adapt. We further propose aBAE, an annealed variant of BAE drawing on methods from statistical inference, to enhance robustness. Our proposals are parallelizable in both quantum and classical components, offer tools for fast noise model assessment, and can leverage preexisting information. Additionally, they accommodate experimental limitations and preferred cost trade-offs. We propose a robust benchmark for amplitude estimation algorithms and use it to test BAE against other approaches, demonstrating its competitive performance in both noisy and noiseless scenarios. In both cases, it achieves lower error than any other algorithm as a function of the cost. In the presence of decoherence, it is capable of learning when other algorithms fail.

The exponential growth in the number of parameters of Deep Neural Networkss (DNNs) in recent years has driven advancements in model capabilities while also leading to unprecedented energy consumption in the HPC world. During training, thousands of processing units are used to process massive datasets in parallel, raising challenges such as synchronization between Graphical Processing Units (GPUs) within and across nodes. The key challenge is to reduce the energy consumption of GPUs without compromising application performance. This work aims to investigate the effects of applying Dynamic Voltage and Frequency (DVFS) to GPUs in a distributed training context, analyzing the trade-offs between energy efficiency and training time. DVFS is a widely used technique for reducing the energy consumption of computing systems by adjusting the frequency and voltage of hardware components. However, most state of the art systems that use it are still agnostic to the different requirements of the training phases that compose DNN training. Moreover, few actually target distributed training and their multiple parallelism scenarios, as well as the dynamic conditions of both these workloads and environments. Thus, we developed PowGo, a tool that dynamically adjusts the GPU frequency by recognizing the different phases of an iteration and how they overlap with each other. Additionally, it considers that GPUs may be running at different speeds and tries to remove computation imbalance by changing the frequency of each GPU accordingly. Our results show that PowGo decreases training energy consumption up to 32% when compared to scenarios used in production environments, which include multi-GPU and multi-node scenarios, as well as training with volatility.