ORCID Profile
0000-0002-0811-6150
Current Organisations
City University of Hong Kong
,
Monash University
Does something not look right? The information on this page has been harvested from data sources that may not be up to date. We continue to work with information providers to improve coverage and quality. To report an issue, use the Feedback Form.
Publisher: IEEE
Date: 20-09-2020
Publisher: IEEE
Date: 12-2017
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 2019
Publisher: IEEE
Date: 02-05-2022
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 15-06-2021
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 09-2022
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 09-2020
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 12-2022
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 05-2019
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 03-2023
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 08-2020
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 03-2022
Publisher: ACM
Date: 17-10-2022
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 15-04-2021
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 08-2020
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Date: 09-2022
Publisher: IEEE
Date: 12-2019
Publisher: IEEE
Date: 12-2017
Publisher: Elsevier BV
Date: 2020
Publisher: Association for Computing Machinery (ACM)
Date: 06-12-2022
DOI: 10.1145/3453169
Abstract: Federated Edge Learning (FEL) allows edge nodes to train a global deep learning model collaboratively for edge computing in the Industrial Internet of Things (IIoT), which significantly promotes the development of Industrial 4.0. However, FEL faces two critical challenges: communication overhead and data privacy. FEL suffers from expensive communication overhead when training large-scale multi-node models. Furthermore, due to the vulnerability of FEL to gradient leakage and label-flipping attacks, the training process of the global model is easily compromised by adversaries. To address these challenges, we propose a communication-efficient and privacy-enhanced asynchronous FEL framework for edge computing in IIoT. First, we introduce an asynchronous model update scheme to reduce the computation time that edge nodes wait for global model aggregation. Second, we propose an asynchronous local differential privacy mechanism, which improves communication efficiency and mitigates gradient leakage attacks by adding well-designed noise to the gradients of edge nodes. Third, we design a cloud-side malicious node detection mechanism to detect malicious nodes by testing the local model quality. Such a mechanism can avoid malicious nodes participating in training to mitigate label-flipping attacks. Extensive experimental studies on two real-world datasets demonstrate that the proposed framework can not only improve communication efficiency but also mitigate malicious attacks while its accuracy is comparable to traditional FEL frameworks.
Publisher: National Institute for Health and Care Research
Date: 09-2022
DOI: 10.3310/RTLH7522
Abstract: People with language problems following stroke (aphasia) benefit from speech and language therapy. Optimising speech and language therapy for aphasia recovery is a research priority. The objectives were to explore patterns and predictors of language and communication recovery, optimum speech and language therapy intervention provision, and whether or not effectiveness varies by participant subgroup or language domain. This research comprised a systematic review, a meta-analysis and a network meta-analysis of in idual participant data. Participant data were collected in research and clinical settings. The intervention under investigation was speech and language therapy for aphasia after stroke. The main outcome measures were absolute changes in language scores from baseline on overall language ability, auditory comprehension, spoken language, reading comprehension, writing and functional communication. Electronic databases were systematically searched, including MEDLINE, EMBASE, Cumulative Index to Nursing and Allied Health Literature, Linguistic and Language Behavior Abstracts and SpeechBITE (searched from inception to 2015). The results were screened for eligibility, and published and unpublished data sets (randomised controlled trials, non-randomised controlled trials, cohort studies, case series, registries) with at least 10 in idual participant data reporting aphasia duration and severity were identified. Existing collaborators and primary researchers named in identified records were invited to contribute electronic data sets. In idual participant data in the public domain were extracted. Data on demographics, speech and language therapy interventions, outcomes and quality criteria were independently extracted by two reviewers, or available as in idual participant data data sets. Meta-analysis and network meta-analysis were used to generate hypotheses. We retrieved 5928 in idual participant data from 174 data sets across 28 countries, comprising 75 electronic (3940 in idual participant data), 47 randomised controlled trial (1778 in idual participant data) and 91 speech and language therapy intervention (2746 in idual participant data) data sets. The median participant age was 63 years (interquartile range 53–72 years). We identified 53 unavailable, but potentially eligible, randomised controlled trials (46 of these appeared to include speech and language therapy). Relevant in idual participant data were filtered into each analysis. Statistically significant predictors of recovery included age (functional communication, in idual participant data: 532, n = 14 randomised controlled trials) and sex (overall language ability, in idual participant data: 482, n = 11 randomised controlled trials functional communication, in idual participant data: 532, n = 14 randomised controlled trials). Older age and being a longer time since aphasia onset predicted poorer recovery. A negative relationship between baseline severity score and change from baseline ( p 0.0001) may reflect the reduced improvement possible from high baseline scores. The frequency, duration, intensity and dosage of speech and language therapy were variously associated with auditory comprehension, naming and functional communication recovery. There were insufficient data to examine spontaneous recovery. The greatest overall gains in language ability [14.95 points (95% confidence interval 8.7 to 21.2 points) on the Western Aphasia Battery-Aphasia Quotient] and functional communication [0.78 points (95% confidence interval 0.48 to 1.1 points) on the Aachen Aphasia Test-Spontaneous Communication] were associated with receiving speech and language therapy 4 to 5 days weekly for auditory comprehension [5.86 points (95% confidence interval 1.6 to 10.0 points) on the Aachen Aphasia Test-Token Test], the greatest gains were associated with receiving speech and language therapy 3 to 4 days weekly. The greatest overall gains in language ability [15.9 points (95% confidence interval 8.0 to 23.6 points) on the Western Aphasia Battery-Aphasia Quotient] and functional communication [0.77 points (95% confidence interval 0.36 to 1.2 points) on the Aachen Aphasia Test-Spontaneous Communication] were associated with speech and language therapy participation from 2 to 4 (and more than 9) hours weekly, whereas the highest auditory comprehension gains [7.3 points (95% confidence interval 4.1 to 10.5 points) on the Aachen Aphasia Test-Token Test] were associated with speech and language therapy participation in excess of 9 hours weekly (with similar gains notes for 4 hours weekly). While clinically similar gains were made alongside different speech and language therapy intensities, the greatest overall gains in language ability [18.37 points (95% confidence interval 10.58 to 26.16 points) on the Western Aphasia Battery-Aphasia Quotient] and auditory comprehension [5.23 points (95% confidence interval 1.51 to 8.95 points) on the Aachen Aphasia Test-Token Test] were associated with 20–50 hours of speech and language therapy. Network meta-analyses on naming and the duration of speech and language therapy interventions across language outcomes were unstable. Relative variance was acceptable ( 30%). Subgroups may benefit from specific interventions. Data sets were graded as being at a low risk of bias but were predominantly based on highly selected research participants, assessments and interventions, thereby limiting generalisability. Frequency, intensity and dosage were associated with language gains from baseline, but varied by domain and subgroup. These exploratory findings require confirmatory study designs to test the hypotheses generated and to develop more tailored speech and language therapy interventions. This study is registered as PROSPERO CRD42018110947. This project was funded by the National Institute for Health and Care Research (NIHR) Health and Social Care Delivery Research programme and will be published in full in Health and Social Care Delivery Research Vol. 10, No. 28. See the NIHR Journals Library website for further project information. Funding was also provided by The Tavistock Trust for Aphasia.
Publisher: SAGE Publications
Date: 18-05-2022
DOI: 10.1177/17474930221097477
Abstract: Stroke rehabilitation interventions are routinely personalized to address in iduals’ needs, goals, and challenges based on evidence from aggregated randomized controlled trials (RCT) data and meta-syntheses. In idual participant data (IPD) meta-analyses may better inform the development of precision rehabilitation approaches, quantifying treatment responses while adjusting for confounders and reducing ecological bias. We explored associations between speech and language therapy (SLT) interventions frequency (days/week), intensity (h/week), and dosage (total SLT-hours) and language outcomes for different age, sex, aphasia severity, and chronicity subgroups by undertaking prespecified subgroup network meta-analyses of the RELEASE database. MEDLINE, EMBASE, and trial registrations were systematically searched (inception-Sept2015) for RCTs, including ⩾ 10 IPD on stroke-related aphasia. We extracted demographic, stroke, aphasia, SLT, and risk of bias data. Overall-language ability, auditory comprehension, and functional communication outcomes were standardized. A one-stage, random effects, network meta-analysis approach filtered IPD into a single optimal model, examining SLT regimen and language recovery from baseline to first post-intervention follow-up, adjusting for covariates identified a-priori. Data were dichotomized by age (⩽/ 65 years), aphasia severity (mild–moderate/ moderate–severe based on language outcomes’ median value), chronicity (⩽/ 3 months), and sex subgroups. We reported estimates of means and 95% confidence intervals. Where relative variance was high ( 50%), results were reported for completeness. 959 IPD (25 RCTs) were analyzed. For working-age participants, greatest language gains from baseline occurred alongside moderate to high-intensity SLT (functional communication 3-to-4 h/week overall-language and comprehension 9 h/week) older participants’ greatest gains occurred alongside low-intensity SLT (⩽ 2 h/week) except for auditory comprehension ( 9 h/week). For both age-groups, SLT-frequency and dosage associated with best language gains were similar. Participants ⩽ 3 months post-onset demonstrated greatest overall-language gains for SLT at low intensity/moderate dosage (⩽ 2 SLT-h/week 20-to-50 h) for those 3 months, post-stroke greatest gains were associated with moderate-intensity/high-dosage SLT (3–4 SLT-h/week ⩾ 50 hours). For moderate–severe participants, 4 SLT-days/week conferred the greatest language gains across outcomes, with auditory comprehension gains only observed for ⩾ 4 SLT-days/week mild–moderate participants’ greatest functional communication gains were associated with similar frequency (⩾ 4 SLT-days/week) and greatest overall-language gains with higher frequency SLT (⩾ 6 days/weekly). Males’ greatest gains were associated with SLT of moderate (functional communication 3-to-4 h/weekly) or high intensity (overall-language and auditory comprehension ( 9 h/weekly) compared to females for whom the greatest gains were associated with lower-intensity SLT ( 2 SLT-h/weekly). Consistencies across subgroups were also evident greatest overall-language gains were associated with 20-to-50 SLT-h in total auditory comprehension gains were generally observed when SLT 9 h over ⩾ 4 days/week. We observed a treatment response in most subgroups’ overall-language, auditory comprehension, and functional communication language gains. For some, the maximum treatment response varied in association with different SLT-frequency, intensity, and dosage. Where differences were observed, working-aged, chronic, mild–moderate, and male subgroups experienced their greatest language gains alongside high-frequency/intensity SLT. In contrast, older, moderate–severely impaired, and female subgroups within 3 months of aphasia onset made their greatest gains for lower-intensity SLT. The acceptability, clinical, and cost effectiveness of precision aphasia rehabilitation approaches based on age, sex, aphasia severity, and chronicity should be evaluated in future clinical RCTs.
Publisher: Springer Singapore
Date: 2020
Publisher: Informa UK Limited
Date: 08-10-2021
Publisher: Ovid Technologies (Wolters Kluwer Health)
Date: 03-2022
DOI: 10.1161/STROKEAHA.121.035216
Abstract: Optimizing speech and language therapy (SLT) regimens for maximal aphasia recovery is a clinical research priority. We examined associations between SLT intensity (hours/week), dosage (total hours), frequency (days/week), duration (weeks), delivery (face to face, computer supported, in idual tailoring, and home practice), content, and language outcomes for people with aphasia. Databases including MEDLINE and Embase were searched (inception to September 2015). Published, unpublished, and emerging trials including SLT and ≥10 in idual participant data on aphasia, language outcomes, and time post-onset were selected. Patient-level data on stroke, language, SLT, and trial risk of bias were independently extracted. Outcome measurement scores were standardized. A statistical inferencing, one-stage, random effects, network meta-analysis approach filtered in idual participant data into an optimal model examining SLT regimen for overall language, auditory comprehension, naming, and functional communication pre-post intervention gains, adjusting for a priori–defined covariates (age, sex, time poststroke, and baseline aphasia severity), reporting estimates of mean change scores (95% CI). Data from 959 in idual participant data (25 trials) were included. Greatest gains in overall language and comprehension were associated with to 50 hours SLT dosage (18.37 [10.58–26.16] Western Aphasia Battery–Aphasia Quotient 5.23 [1.51–8.95] Aachen Aphasia Test–Token Test). Greatest clinical overall language, functional communication, and comprehension gains were associated with 2 to 4 and 9+ SLT hours/week. Greatest clinical gains were associated with frequent SLT for overall language, functional communication (3–5+ days/week), and comprehension (4–5 days/week). Evidence of comprehension gains was absent for SLT ≤20 hours, hours/week, and ≤3 days/week. Mixed receptive-expressive therapy, functionally tailored, with prescribed home practice was associated with the greatest overall gains. Relative variance was %. Risk of trial bias was low to moderate low for meta-biases. Greatest language recovery was associated with frequent, functionally tailored, receptive-expressive SLT, with prescribed home practice at a greater intensity and duration than reports of usual clinical services internationally. These exploratory findings suggest critical therapeutic ranges, informing hypothesis-testing trials and tailoring of clinical services. URL: www.crd.york.ac.uk/PROSPERO/ Unique identifier: CRD42018110947.
No related grants have been discovered for Yi Liu.