A Unified FPGA Virtualization Framework for General-Purpose Deep Neural Networks in the Cloud

INFerence-as-a-Service (INFaaS) has become a primary workload in the cloud. However, existing FPGA-based Deep Neural Network (DNN) accelerators are mainly optimized for the fastest speed of a single task, while the multi-tenancy of INFaaS has not been explored yet. As the demand for INFaaS keeps growing, simply increasing the number of FPGA-based DNN accelerators is not cost-effective, while merely sharing these single-task optimized DNN accelerators in a time-division multiplexing way could lead to poor isolation and high-performance loss for INFaaS. On the other hand, current cloud-based DNN accelerators have excessive compilation overhead, especially when scaling out to multi-FPGA systems for multi-tenant sharing, leading to unacceptable compilation costs for both offline deployment and online reconfiguration. Therefore, it is far from providing efficient and flexible FPGA virtualization for public and private cloud scenarios. Aiming to solve these problems, we propose a unified virtualization framework for general-purpose deep neural networks in the cloud, enabling multi-tenant sharing for both the Convolution Neural Network (CNN), and the Recurrent Neural Network (RNN) accelerators on a single FPGA. The isolation is enabled by introducing a two-level instruction dispatch module and a multi-core based hardware resources pool. Such designs provide isolated and runtime-programmable hardware resources, which further leads to performance isolation for multi-tenant sharing. On the other hand, to overcome the heavy re-compilation overheads, a tiling-based instruction frame package design and a two-stage static-dynamic compilation, are proposed. Only the lightweight runtime information is re-compiled with ∼1 ms overhead, thus guaranteeing the private cloud’s performance. Finally, the extensive experimental results show that the proposed virtualized solutions achieve up to 3.12× and 6.18× higher throughput in the private cloud compared with the static CNN and RNN baseline designs, respectively.

Evaluation of multi-task learning in deep learning-based positioning classification of mandibular third molars

Pell and Gregory, and Winter’s classifications are frequently implemented to classify the mandibular third molars and are crucial for safe tooth extraction. This study aimed to evaluate the classification accuracy of convolutional neural network (CNN) deep learning models using cropped panoramic radiographs based on these classifications. We compared the diagnostic accuracy of single-task and multi-task learning after labeling 1330 images of mandibular third molars from digital radiographs taken at the Department of Oral and Maxillofacial Surgery at a general hospital (2014–2021). The mandibular third molar classifications were analyzed using a VGG 16 model of a CNN. We statistically evaluated performance metrics [accuracy, precision, recall, F1 score, and area under the curve (AUC)] for each prediction. We found that single-task learning was superior to multi-task learning (all p < 0.05) for all metrics, with large effect sizes and low p-values. Recall and F1 scores for position classification showed medium effect sizes in single and multi-task learning. To our knowledge, this is the first deep learning study to examine single-task and multi-task learning for the classification of mandibular third molars. Our results demonstrated the efficacy of implementing Pell and Gregory, and Winter’s classifications for specific respective tasks.

An Adaptive Imitation Learning Framework for Robotic Complex Contact-Rich Insertion Tasks

Complex contact-rich insertion is a ubiquitous robotic manipulation skill and usually involves nonlinear and low-clearance insertion trajectories as well as varying force requirements. A hybrid trajectory and force learning framework can be utilized to generate high-quality trajectories by imitation learning and find suitable force control policies efficiently by reinforcement learning. However, with the mentioned approach, many human demonstrations are necessary to learn several tasks even when those tasks require topologically similar trajectories. Therefore, to reduce human repetitive teaching efforts for new tasks, we present an adaptive imitation framework for robot manipulation. The main contribution of this work is the development of a framework that introduces dynamic movement primitives into a hybrid trajectory and force learning framework to learn a specific class of complex contact-rich insertion tasks based on the trajectory profile of a single task instance belonging to the task class. Through experimental evaluations, we validate that the proposed framework is sample efficient, safer, and generalizes better at learning complex contact-rich insertion tasks on both simulation environments and on real hardware.

The Specificity of Cognitive-Motor Dual-Task Interference on Balance in Young and Older Adults

Standing upright on stable and unstable surfaces requires postural control. Postural control declines as humans age, presenting greater risk of fall-related injury and other negative health outcomes. Secondary cognitive tasks can further impact balance, which highlights the importance of coordination between cognitive and motor processes. Past research indicates that this coordination relies on executive function (EF; the ability to control, maintain, and flexibly direct attention to achieve goals), which coincidentally declines as humans age. This suggests that secondary cognitive tasks requiring EF may exert a greater influence on balance compared to non-EF secondary tasks, and this interaction could be exaggerated among older adults. In the current study, we had younger and older adults complete two Surface Stability conditions (standing upright on stable vs. unstable surfaces) under varying Cognitive Load; participants completed EF (Shifting, Inhibiting, Updating) and non-EF (Processing Speed) secondary cognitive tasks on tablets, as well as a single task control scenario with no secondary cognitive task. Our primary balance measure of interest was sway area, which was measured with an array of wearable inertial measurement unit sensors. Replicating prior work, we found a main effect of Surface Stability with less sway on stable surfaces compared to unstable surfaces, and we found an interaction between Age and Surface Stability with older adults exhibiting significantly greater sway selectively on unstable surfaces compared to younger adults. New findings revealed a main effect of Cognitive Load on sway, with the single task condition having significantly less sway than two of the EF conditions (Updating and Shifting) and the non-EF condition (Processing Speed). We also found an interaction of Cognitive Load and Surface Stability on postural control, where Surface Stability impacted sway the most for the single task and two of the executive function conditions (Inhibition and Shifting). Interestingly, Age did not interact with Cognitive Load, suggesting that both age groups were equally impacted by secondary cognitive tasks, regardless the presence or type of secondary cognitive task. Taken together, these patterns suggest that cognitive demands vary in their impact on posture control across stable vs. unstable surfaces, and that EF involvement may not be the driving mechanism explaining cognitive-motor dual-task interference on balance.

ASSOCIATION BETWEEN BALANCE SELF-EFFICACY AND WALKING ABILITY IN THOSE WITH NEW LOWER LIMB AMPUTATIONS

BACKGROUND: A relationship between walking ability and self-efficacy has been demonstrated in various rehabilitation patient populations. In experienced prosthetic ambulators, walking ability is related to self-efficacy of balance, however, this relationship has not been quantified for those with newly acquired lower limb amputations (LLA). OBJECTIVE(S): To investigate the association between walking performance (objective) and self-reported walking abilities (subjective) on balance self-efficacy in those with LLA. METHODOLOGY: Cross-sectional study of 27 people (17 men; mean age=63.57±9.33) at discharge from inpatient prosthetic rehabilitation for first major unilateral LLA. Individuals completed 6m straight path walking and the L-Test under single- and dual-task conditions. The Prosthesis Evaluation Questionnaire (PEQ) was administered, and the Ambulation subscale provided subjective measures of walking ability. A single PEQ question on satisfaction with walking (16B) was also used as a proxy for subjective walking ability. The Activities-specific Balance Confidence Scale measured balance self-efficacy. Multivariable linear regression was used to evaluate the strength of association between walking ability (objective and subjective) and balance self-efficacy (dependent variable). FINDINGS: Walking velocity on the 6m straight path under single-task (p=0.011) and dual-task conditions (p=0.039), the single-task L-Test (p=0.035) and self-reported satisfaction with walking (p=0.019) were associated with self-efficacy of balance. CONCLUSION: Objective measures of walking ability that were independently associated with balance self-efficacy included straight path walking velocity under single and dual-task conditions and the single-task L-Test. Satisfaction with walking was also associated with balance self-efficacy. This highlights the interplay between physical and psychological factors during rehabilitation. More research in the area of self-efficacy and walking ability is needed to establish self-efficacy as a target during prosthetic rehabilitation for those with LLA. Layman's Abstract Self-efficacy is a person’s belief in their ability to do a certain task well. Improving self-efficacy can be done by watching others complete a task, by getting praise from experts, or by doing the task yourself. There is a link between how well some people walk and their confidence with walking, however this has not been studied in people learning to use a lower limb prosthesis. The goal of this paper was to study the link between balance self-efficacy, scores on walking tests and self-reported walking ability in those with lower limb amputations (LLA) when they leave rehabilitation. To do this, two walking tests were done (straight path and complex path) in two settings (walking only and walking with distraction). A survey about walking ability and a questionnaire on balance self-efficacy were also done. Results showed that self-efficacy of balance was related to the straight path walking test under both settings and the complex walking test during walking alone. A person’s satisfaction with walking ability was also linked. The only test not related was the complex walking test under distracting conditions. It might be that more time is needed for people with LLA to confidently do this task. This shows the link between physical and mental factors during rehabilitation. More research is needed to find other factors that might impact self-efficacy and walking ability in people with LLA when they leave rehabilitation. Article PDF Link: https://jps.library.utoronto.ca/index.php/cpoj/article/view/36695/28904 How To Cite: Frengopoulos C, Zia Z, Payne M.W.C, Viana R, Hunter S.W. Association between balance self-efficacy and walking ability in those with new lower limb amputations. Canadian Prosthetics & Orthotics Journal. 2022; Volume 5, Issue 1, No.4. https://doi.org/10.33137/cpoj.v5i1.36695 Corresponding Author: Courtney Frengopoulos,University of Western Ontario, Room 1408, Elborn College, London, Ontario, Canada, N6G 1H1.E-Mail: [email protected] ID: https://orcid.org/0000-0002-4131-2727

Head-Mounted and Hand-Held Displays Diminish the Effectiveness of Fall-Resisting Skills

Use of head-mounted displays (HMDs) and hand-held displays (HHDs) may affect the effectiveness of stability control mechanisms and impair resistance to falls. This study aimed to examine whether the ability to control stability during locomotion is diminished while using HMDs and HHDs. Fourteen healthy adults (21–46 years) were assessed under single-task (no display) and dual-task (spatial 2-n-back presented on the HMD or the HHD) conditions while performing various locomotor tasks. An optical motion capture system and two force plates were used to assess locomotor stability using an inverted pendulum model. For perturbed standing, 57% of the participants were not able to maintain stability by counter-rotation actions when using either display, compared to the single-task condition. Furthermore, around 80% of participants (dual-task) compared to 50% (single-task) showed a negative margin of stability (i.e., an unstable body configuration) during recovery for perturbed walking due to a diminished ability to increase their base of support effectively. However, no evidence was found for HMDs or HHDs affecting stability during unperturbed locomotion. In conclusion, additional cognitive resources required for dual-tasking, using either display, are suggested to result in delayed response execution for perturbed standing and walking, consequently diminishing participants’ ability to use stability control mechanisms effectively and increasing the risk of falls.

Hierarchical shared transfer learning for biomedical named entity recognition

Background Biomedical named entity recognition (BioNER) is a basic and important medical information extraction task to extract medical entities with special meaning from medical texts. In recent years, deep learning has become the main research direction of BioNER due to its excellent data-driven context coding ability. However, in BioNER task, deep learning has the problem of poor generalization and instability. Results we propose the hierarchical shared transfer learning, which combines multi-task learning and fine-tuning, and realizes the multi-level information fusion between the underlying entity features and the upper data features. We select 14 datasets containing 4 types of entities for training and evaluate the model. The experimental results showed that the F1-scores of the five gold standard datasets BC5CDR-chemical, BC5CDR-disease, BC2GM, BC4CHEMD, NCBI-disease and LINNAEUS were increased by 0.57, 0.90, 0.42, 0.77, 0.98 and − 2.16 compared to the single-task XLNet-CRF model. BC5CDR-chemical, BC5CDR-disease and BC4CHEMD achieved state-of-the-art results.The reasons why LINNAEUS’s multi-task results are lower than single-task results are discussed at the dataset level. Conclusion Compared with using multi-task learning and fine-tuning alone, the model has more accurate recognition ability of medical entities, and has higher generalization and stability.

Measured and perceived exercise intensity during the performance of single-task, cognitive-motor dual-task and exergame training: a transversal study (Preprint)

UNSTRUCTURED The physical and cognitive loads during exergaming may differ from more conventional cognitive-motor dual-task trainings. The aim of this pilot transversal study was to compare exercise intensity during exergame, cognitive-motor dual-task and single-task training sessions. We recruited healthy young adults who carried out one session of each t type of training: exergaming, cognitive-motor dual-tasking and single-tasking. We used a custom-made exergame as support. The sessions lasted 30 minutes, were spaced at least 24 hours, and took place in random order for each group of 4 participants. We used heart rates to assess exercise intensity, and the modified Borg scale to assess their perception of intensity. Sixteen healthy young participants carried out all sessions. There was no difference between the different types of training in mean heart rates (p = 0.3), peak heart rates (p = 0.5) or Borg scale scores (p = 0.4). Our custom-made exergames measured and perceived physical load did not differ between cognitive-motor dual-task and single-task training. As a result, our exergame can be considered as intense as more traditional physical training. Future studies should be conducted in seniors with or without cognitive impairments and should incorporate an assessment of cognitive performance.

2BALANCE: Test-retest reliability of a cognitive-motor dual-task protocol

PURPOSE: Aside from typical symptoms such as dizziness and vertigo, persons with vestibular disorders often have cognitive and motor problems. These symptoms have been assessed in single-task condition. However, dual-tasks assessing cognitive-motor interference might be an added value as they reflect daily life situations better. Therefore, the 2BALANCE protocol was developed. In the current study, the test-retest reliability of this protocol was assessed. METHODS: The 2BALANCE protocol was performed twice in 20 healthy young adults with an in-between test interval of two weeks. Two motor tasks and five different cognitive tasks were performed in single and dual-task condition. Intraclass correlation coefficients (ICC), the standard error of measurement, and the minimal detectable difference were calculated. RESULTS: All cognitive tasks, with the exception of the mental rotation task, had favorable reliability results (0.26≤ICC≤0.91). The dynamic motor task indicated overall substantial reliability values in all conditions (0.67≤ICC≤0.98). Similar results were found for the static motor task during dual-tasking (0.50≤ICC≤0.92), but were slightly lower in single-task condition (–0.26≤ICC≤0.75). CONCLUSIONS: The 2BALANCE protocol was overall consistent across trials. However, the mental rotation task showed lowest reliability values.