Charis Kouzinopoulos

Intracortical brain-machine interfaces demand low-latency, energy-efficient solutions for neural decoding. Spiking Neural Networks (SNNs) deployed on neuromorphic hardware have demonstrated remarkable efficiency in neural decoding by leveraging sparse binary activations and efficient spatiotemporal processing. However, reducing the computational cost of SNNs remains a critical challenge for developing ultra-efficient intracortical neural implants. In this work, we introduce a novel adaptive pruning algorithm specifically designed for SNNs with high activation sparsity, targeting intracortical neural decoding. Our method dynamically adjusts pruning decisions and employs a rollback mechanism to selectively eliminate redundant synaptic connections without compromising decoding accuracy. Experimental evaluation on the NeuroBench Non-Human Primate (NHP) Motor Prediction benchmark shows that our pruned network achieves performance comparable to dense networks, with a maximum tenfold improvement in efficiency. Moreover, hardware simulation on the neuromorphic processor reveals that the pruned network operates at sub-μW power levels, underscoring its potential for energy-constrained neural implants. These results underscore the promise of our approach for advancing energy-efficient intracortical brain-machine interfaces with low-overhead on-device intelligence.

Fire incidents in residential and industrial areas are often the cause of human casualties and property damage. Although there are existing systems that detect fire and monitor the presence of people in indoor areas, research on their implementation in embedded platforms is limited. This article introduces an ultra-low-power embedded system for fire detection and crowd counting using efficient deep learning methods. For the prediction of fire occurrences, environmental and gas sensor along with multilayer perceptron nodes are used. For crowd counting, a custom lightweight version of YOLOv5 is introduced, using an architecture based on ShuffleNetV2, resulting in a model with low memory requirements, high accuracy predictions, and fast inference on an embedded platform. The accuracy, power consumption, and memory requirements of the proposed system are evaluated using public datasets and datasets acquired by the environmental and image sensors, and its performance is compared to that of existing approaches.

A LoRaWAN node powered using an 8 cm2 solar cell was designed and its low light harvesting performance evaluated. The embedded system is used to sense some parameters and transmit the results every 10 minutes, using the spreading factor SF7BW125 and transmitting with + 8 dBm, which allows the coverage of a small building. The node can cold start with less than 30 lux. Once started, its operations can be sustained down to 20 lux. Operation at higher spreading factors or higher RF output power is also possible if the transmission interval is increased. Such a performance enables the use of energy autonomous LPWAN nodes in poorly lit environments. The small size of the solar cell makes it possible to build small nodes.

This paper introduces an Autonomous Smart Sensing Card (ASSC), an embedded system that will be powered indoors and outdoors by harvested energy, have miniaturized dimensions and serve multi-sensorial IoT applications for smart living and working environments. It will consist of a combination of newly developed and optimized off-the-shelf or close-to-commercialization technologies such as PV harvesters, energy storage and power management units, MCUs and sensors, all packed with a form factor under 3mm in thickness. The system will introduce technical breakthroughs that will boost further miniaturization, a small footprint, ultra-low power consumption as well as short- and long-range communications.

The commonalities between the ALICE and FAIR experiments and their computing requirements led to the development of large parts of a common software framework in an experiment independent way. The FairRoot project has already shown the feasibility of such an approach for the FAIR experiments and extending it beyond FAIR to experiments at other facilities[1, 2]. The ALFA framework is a joint development between ALICE Online- Offline (O2) and FairRoot teams. ALFA is designed as a flexible, elastic system, which balances reliability and ease of development with performance using multi-processing and multithreading. A message- based approach has been adopted; such an approach will support the use of the software on different hardware platforms, including heterogeneous systems. Each process in ALFA assumes limited communication and reliance on other processes. Such a design will add horizontal scaling (multiple processes) to vertical scaling provided by multiple threads to meet computing and throughput demands. ALFA does not dictate any application protocols. Potentially, any content-based processor or any source can change the application protocol. The framework supports different serialization standards for data exchange between different hardware and software languages.

Baker and Bird and Baeza-Yates and Regnier are two of the most efficient and widely used algorithms for exact online two-dimensional pattern matching. Both use the automaton of the Aho-Corasick multiple pattern matching algorithm to locate all the occurrences of a two-dimensional pattern in a two-dimensional input string, a data structure that is considered by many as inefficient, especially when used to process long patterns or data using large alphabet sizes. This article presents variants of the Baker and Bird and the Baeza-Yates and Regnier algorithms that use the data structures of the Set Horspool, Wu-Manber, Set Backward Oracle Matching, and SOG multiple pattern matching algorithms in place of the automaton of Aho-Corasick and evaluates their performance experimentally in terms of preprocessing and searching time.