Peter Tang

Peter Tang

Auto-generated profile, based on publicly available information

Peter Tang

Allegheny Health Network, Pittsburgh, Pennsylvania

Research Focus & Metrics

Academic Career Length

14Years (2013-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 April 2019 in The Journal of Hand Surgery
Verified
Journal Article
Closed access
JDJacob T. Didesch
Jacob T. Didesch
PTPeter Tang
Peter Tang

1 February 2018 in Hand Clinics
Verified
Journal Article
Closed access
DPDavid Pope
David Pope
PTPeter Tang
Peter Tang

14 August 2021
Verified
Conference Paper
Open access
VGVipul Gupta
Vipul Gupta
DCDhruv Choudhary
Dhruv Choudhary
PTPeter Tang
Peter Tang
XWXiaohan Wei
Xiaohan Wei
Xing WangXing Wang
Xing Wang
MMMichael W. Mahoney
Michael W. Mahoney
In this paper, we consider hybrid parallelism---a paradigm that employs both Data Parallelism (DP) and Model Parallelism (MP)---to scale distributed training of large recommendation models. We propose a compression framework called Dynamic Communication Thresholding (DCT) for communication-efficient hybrid training. DCT filters the entities to be communicated across the network through a simple hard-thresholding function, allowing only the most relevant information to pass through. For communication efficient DP, DCT compresses the parameter gradients sent to the parameter server during model synchronization. The threshold is updated only once every few thousand iterations to reduce the computational overhead of compression. For communication efficient MP, DCT incorporates a novel technique to compress the activations and gradients sent across the network during the forward and backward propagation, respectively. This is done by identifying and updating only the most relevant neurons of the neural network for each training sample in the data. We evaluate DCT on publicly available natural language processing and recommender models and datasets, as well as recommendation systems used in production at Facebook. DCT reduces communication by at least 100x and 20x during DP and MP, respectively. The algorithm has been deployed in production, and it improves end-to-end training time for a state-of-the-art industrial recommender model by 37%, without any loss in performance.

15 May 2019 in Journal of Bone and Joint Surgery
Verified
Journal Article
Closed access
PTPeter Tang
Peter Tang
DWDaniel R. Whiteman
Daniel R. Whiteman
CVClifford Voigt
Clifford Voigt
MMMark C. Miller
Mark C. Miller
HKHongkyun Kim
Hongkyun Kim
Background: Nerve injuries with a gap/defect represent a clinical challenge without a clear solution. Reconstruction with cable autografts is a common treatment technique, and repair with decellular nerve allograft is a newer option. The purpose of this study was to compare the functional outcomes of reconstruction with cable autografts with those of matched-diameter decellular nerve allografts to evaluate the relative importance of diameter as well as the autograft-versus-allograft nature of the reconstruction. Methods: A unilateral 10-mm sciatic nerve defect was created in 81 genetically identical male Lewis rats and then repaired with a reverse autograft, 4 or 5 sural nerve cable autografts, or a matched-diameter decellular nerve allograft. In each group, at each time point (12, 16, and 20 weeks), all 9 animals underwent functional testing and 5 of the 9 underwent histologic analysis. Functional testing included bilateral measurements of the isometric tetanic force of the tibialis anterior (primary outcome), the weight of the tibialis anterior, and the gastrocnemius compound muscle action potential (CMAP) latency. Histologic evaluation included an axon count as well as measurement of the axon density, fiber diameter, myelin thickness, and G-ratio. Results: The repair groups did not differ significantly in terms of isometric tetanic force, muscle weight, or CMAP latency, but these measurements did differ significantly according to the time after surgery (p < 0.05). The isometric tetanic force percent recovery (width of the 95% confidence interval) for the reverse autograft, cable autograft, and decellular nerve allograft was 57.7% (15.6%), 57.0% (23.4%), and 56.0% (19.7%), respectively, at 12 weeks; 69.1% (14.7%), 65.6% (18.5%), and 65.9% (29.1%) at 16 weeks; and 72.5% (18.2%), 73.7% (25.6%), and 71.8% (22.4%) at 20 weeks. Isometric tetanic force and muscle weight recovery were greater and CMAP latency was shorter at 20 and 16 weeks after surgery than they were at 12 weeks. The treatment type did not affect any of the histologic outcomes. Conclusions: In this animal study, we found that matched-diameter decellular nerve allograft was not significantly different from reverse autograft or cable graft reconstruction in terms of function and histologic outcomes. These findings support decellular nerve allograft as a viable treatment option for nerve reconstruction. Clinical Relevance: This study showed that decellular nerve allograft was no different from cable or reverse autograft in terms of outcome measures in a rat sciatic nerve defect model. If these results are applicable clinically, it would obviate the need for autograft nerve harvest and its ensuing donor site morbidity.

1 July 2021 in JBJS Case Connector
Verified
Journal Article
Closed access
RGR. David Graham
R. David Graham
JAJonathan Acosta
Jonathan Acosta
PTPeter Tang
Peter Tang
Case: We present a case of a patient who sustained a rhomboid major origin tear with subsequent seroma formation. The rhomboid major origin was reattached using an Achilles tendon allograft to supplement the repair, and the seroma wall was excised. We were able to alleviate the patient's pain and improve his shoulder function. We present our surgical technique. Conclusion: Scapular winging is most often attributed to neurogenic causes, but the goal of this case report is to help orthopaedic surgeons realize that scapular winging is not only a nerve issue but can result from an acute muscle tear.

1 April 2021 in JBJS Case Connector
Verified
Journal Article
Closed access
PTPeter Tang
Peter Tang
MSMark Schreck
Mark Schreck
Case: This is a case of an 18-year-old woman with superior and lateral scapular winging due to traumatic injuries to the insertions of the superior and middle trapezius as well as the origins of the latissimus dorsi and lower trapezius. Conclusion: Scapular winging can be caused by a muscle tear in absence of a nerve injury. This patient underwent latissimus dorsi origin repair and superior and middle trapezius repairs. An Achilles allograft was used to reconstruct her lower trapezius to provide static restraint to her scapula, while a rhomboid major transfer was performed to provide dynamic restraint.