Reducing the Energy Consumption of sEMG-Based Gesture Recognition at the Edge Using Transformers and Dynamic Inference.

Hand gesture recognition applications based on surface electromiographic (sEMG) signals can benefit from on-device execution to achieve faster and more predictable response times and higher energy efficiency. However, deploying state-of-the-art deep learning (DL) models for this task on memory-constrained and battery-operated edge devices, such as wearables, requires a careful optimization process, both at design time, with an appropriate tuning of the DL models' architectures, and at execution time, where the execution of large and computationally complex models should be avoided unless strictly needed. In this work, we pursue both optimization targets, proposing a novel gesture recognition system that improves upon the state-of-the-art models both in terms of accuracy and efficiency. At the level of DL model architecture, we apply for the first time tiny transformer models (which we call bioformers) to sEMG-based gesture recognition. Through an extensive architecture exploration, we show that our most accurate bioformer achieves a higher classification accuracy on the popular Non-Invasive Adaptive hand Prosthetics Database 6 (Ninapro DB6) dataset compared to the state-of-the-art convolutional neural network (CNN) TEMPONet (+3.1%). When deployed on the RISC-V-based low-power system-on-chip (SoC) GAP8, bioformers that outperform TEMPONet in accuracy consume 7.8×-44.5× less energy per inference. At runtime, we propose a three-level dynamic inference approach that combines a shallow classifier, i.e., a random forest (RF) implementing a simple "rest detector" with two bioformers of different accuracy and complexity, which are sequentially applied to each new input, stopping the classification early for "easy" data. With this mechanism, we obtain a flexible inference system, capable of working in many different operating points in terms of accuracy and average energy consumption. On GAP8, we obtain a further 1.03×-1.35× energy reduction compared to static bioformers at iso-accuracy.

A deep learning approach for Spatio-Temporal forecasting of new cases and new hospital admissions of COVID-19 spread in Reggio Emilia, Northern Italy.

BACKGROUND: Since February 2020, the COVID-19 epidemic has rapidly spread throughout Italy. Some studies showed an association of environmental factors, such as PM10, PM2.5, NO2, temperature, relative humidity, wind speed, solar radiation and mobility with the spread of the epidemic. In this work, we aimed to predict via Deep Learning the real-time transmission of SARS-CoV-2 in the province of Reggio Emilia, Northern Italy, in a grid with a small resolution (12 km × 12 km), including satellite information. METHODS: We focused on the Province of Reggio Emilia, which was severely hit by the first wave of the epidemic. The outcomes included new SARS-CoV-2 infections and COVID-19 hospital admissions. Pollution, meteorological and mobility data were analyzed. The spatial simulation domain included the Province of Reggio Emilia in a grid of 40 cells of (12 km)2. We implemented a ConvLSTM, which is a spatio-temporal deep learning approach, to perform a 7-day moving average to forecast the 7th day after. We used as training and validation set the new daily infections and hospital admissions from August 2020 to March 2021. Finally, we assessed the models in terms of Mean Absolute Error (MAE) compared with Mean Observed Value (MOV) and Root Mean Squared Error (RMSE) on data from April to September 2021. We tested the performance of different combinations of input variables to find the best forecast model. FINDINGS: Daily new cases of infection, mobility and wind speed resulted in being strongly predictive of new COVID-19 hospital admissions (MAE = 2.72 in the Province of Reggio Emilia; MAE = 0.62 in Reggio Emilia city), whereas daily new cases, mobility, solar radiation and PM2.5 turned out to be the best predictors to forecast new infections, with appropriate time lags. INTERPRETATION: ConvLSTM achieved good performances in forecasting new SARS-CoV-2 infections and new COVID-19 hospital admissions. The spatio-temporal representation allows borrowing strength from data neighboring to forecast at the level of the square cell (12 km)2, getting accurate predictions also at the county level, which is paramount to help optimise the real-time allocation of health care resources during an epidemic emergency.

Prediction of All-Cause Mortality Following Percutaneous Coronary Intervention in Bifurcation Lesions Using Machine Learning Algorithms.

Stratifying prognosis following coronary bifurcation percutaneous coronary intervention (PCI) is an unmet clinical need that may be fulfilled through the adoption of machine learning (ML) algorithms to refine outcome predictions. We sought to develop an ML-based risk stratification model built on clinical, anatomical, and procedural features to predict all-cause mortality following contemporary bifurcation PCI. Multiple ML models to predict all-cause mortality were tested on a cohort of 2393 patients (training, n = 1795; internal validation, n = 598) undergoing bifurcation PCI with contemporary stents from the real-world RAIN registry. Twenty-five commonly available patient-/lesion-related features were selected to train ML models. The best model was validated in an external cohort of 1701 patients undergoing bifurcation PCI from the DUTCH PEERS and BIO-RESORT trial cohorts. At ROC curves, the AUC for the prediction of 2-year mortality was 0.79 (0.74-0.83) in the overall population, 0.74 (0.62-0.85) at internal validation and 0.71 (0.62-0.79) at external validation. Performance at risk ranking analysis, k-center cross-validation, and continual learning confirmed the generalizability of the models, also available as an online interface. The RAIN-ML prediction model represents the first tool combining clinical, anatomical, and procedural features to predict all-cause mortality among patients undergoing contemporary bifurcation PCI with reliable performance.

On the impact of smart sensor approximations on the accuracy of machine learning tasks.

Smart sensors present in ubiquitous Internet of Things (IoT) devices often obtain high energy efficiency by carefully tuning how the sensing, the analog to digital (A/D) conversion and the digital serial transmission are implemented. Such tuning involves approximations, i.e. alterations of the sensed signals that can positively affect energy consumption in various ways. However, for many IoT applications, approximations may have an impact on the quality of the produced output, for example on the classification accuracy of a Machine Learning (ML) model. While the impact of approximations on ML algorithms is widely studied, previous works have focused mostly on processing approximations. In this work, in contrast, we analyze how the signal alterations imposed by smart sensors impact the accuracy of ML classifiers. We focus in particular on data alterations introduced in the serial transmission from a smart sensor to a processor, although our considerations can also be extended to other sources of approximation, such as A/D conversion. Results on several types of models and on two different datasets show that ML algorithms are quite resilient to the alterations produced by smart sensors, and that the serial transmission energy can be reduced by up to 70% without a significant impact on classification accuracy. Moreover, we also show that, contrarily to expectations, the two generic approximation families identified in our work yield similar accuracy losses.