Inference at-the-edge using embedded machine learning models is associated with challenging trade-offs between resource metrics, such as energy and memory footprint, and the performance metrics, such as computation time and accuracy. In this work, we go beyond the conventional Neural Network based approaches to explore Tsetlin Machine (TM), an emerging machine learning algorithm, that uses learning automata to create propositional logic for classification. We use algorithm-hardware co-design to propose a novel methodology for training and inference of TM. The methodology, called REDRESS, comprises independent TM training and inference techniques to reduce the memory footprint of the resulting automata to target low and ultra-low power applications. The array of Tsetlin Automata (TA) holds learned information in the binary form as bits: {0,1}, called excludes and includes, respectively. REDRESS proposes a lossless TA compression method, called the include-encoding, that stores only the information associated with includes to achieve over 99% compression. This is enabled by a novel computationally minimal training procedure, called the Tsetlin Automata Re-profiling, to improve the accuracy and increase the sparsity of TA to reduce the number of includes, hence, the memory footprint. Finally, REDRESS includes an inherently bit-parallel inference algorithm that operates on the optimally trained TA in the compressed domain, that does not require decompression during runtime, to obtain high speedups when compared with the state-of-the-art Binary Neural Network (BNN) models. In this work, we demonstrate that using REDRESS approach, TM outperforms BNN models on all design metrics for five benchmark datasets viz. MNIST, CIFAR2, KWS6, Fashion-MNIST and Kuzushiji-MNIST. When implemented on an STM32F746G-DISCO microcontroller, REDRESS obtained speedups and energy savings ranging 5-5700× compared with different BNN models.
Unlabelled: Pathogenic coding mutations are prevalent in human neuronal transcription factors (TFs) but how they disrupt development is poorly understood. Lmx1b is a master transcriptional regulator of postmitotic neurons that give rise to mature serotonin (5-HT) neurons; over two hundred pathogenic heterozygous mutations have been discovered in human yet their impact on brain development has not been investigated. Here, we developed mouse models with different DNA-binding missense mutations.
The new generation of non-hydrostatic and compressible numerical models of the ocean can explicitly simulate acoustic waves when and where space and time resolution is adapted. We show that these models can consequently propagate accurately acoustic waves and modes through a free-surface, stratified ocean evolving simultaneously both in space and time, bringing them to the state of the art of acoustic propagation modelling. To some extent, both numerical cost and memory footprint may temper their range of applications but they are an unprecedented tool to evaluate deterministically the effects of ocean variability on low-frequency acoustic propagation in a realistically-evolving ocean.
Institute of Electronics and Informatics Engineering of Aveiro (IEETA), University of Aveiro, Campus Universitário de Santiago, 3810-193 Aveiro, Portugal.
Motivation: Large-scale genomic projects grapple with the complex challenge of reducing medium- and long-term storage space and its associated energy consumption, monetary costs, and environmental footprint.
Results: We present JARVIS3, an advanced tool engineered for the efficient reference-free compression of genomic sequences. JARVIS3 introduces a pioneering approach, specifically through enhanced table memory models and probabilistic lookup-tables applied in repeat models.
Signal denoising methods based on deep learning have been extensively adopted on electroencephalogram devices. However, they are unable to deploy on edge-based portable or wearable (P/W) electronics due to the high computational complexity of the existed models. To overcome such issue, we propose an edge-based lightweight Kalman filter network (EKFNet) that does not require manual prior knowledge estimation.
We propose a novel 1-D median estimator specifically designed for the online detection of threshold-crossing signals, such as spikes in extracellular neural recordings. Compared to state-of-the-art algorithms, our method reduces estimator variance by up to eight times for a given buffer length. Likewise, for a given estimator variance, it requires a buffer length that is up to eight times smaller.
