Browsing by Author "Yuan, Binhang"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    A computational framework for evaluating outcomes in infant craniosynostosis reconstruction
    (2016-02-11) Yuan, Binhang; Goldman, Ron
    Historically, surgical outcomes in craniosynostosis have been evaluated by qualitative analysis, direct and indirect anthropometry, cephalometrics, and CT craniometric analysis. Three-dimensional meshes constructed from 3dMD images acquired on patients with synostosis at multiple times across the course of surgical treatment provide ideal raw data for a novel approach to 3D geometric shape analysis of surgical results. We design a automatic computational framework for evaluating and visualizing the results of infant cranial surgeries based on 3dMD images. The goal of this framework is to assist surgeons in evaluating the efficacy of their surgical techniques. Feedback from surgeons in Texas Children's Hospital confirms that this framework is a robust computational system within which surgical outcomes in synostosis can be accurately and meaningfully evaluated. We also propose an algorithm to generate normative infant cranial models from the input of 3D meshes, which are extracted from CT scans of normal infant skulls. Comparing of the head shape of an affected subject with a normal control will more clearly illustrate in what aspect the subject's head deviates from the norm. Comparing of a post-treatment subject's head shape and an age-matched control would allow assessing of a specific treatment approach or surgical technique.
  • Thumbnail Image
    Item
    Distributed Machine Learning Scale Out with Algorithms and Systems
    (2020-12-04) Yuan, Binhang; Jermaine, Chris
    Machine learning (ML) is ubiquitous, and has powered the recent success of artificial intelligence. However, the state of affairs with respect to distributed ML is far from ideal. TensorFlow and PyTorch simply crash when an operation’s inputs and outputs cannot fit on a GPU for model parallelism, or when a model cannot fit on a single machine for data parallelism. A TensorFlow code that works reasonably well on a single machine with eight GPUs procured from a cloud provider often runs slower on two machines totaling sixteen GPUs. In this thesis, I propose solutions at both algorithm and system levels in order to scale out distributed ML. At the algorithm level, I propose a new method to distributed neural network learning, called independent subnet training (IST). In IST, per iteration, a neural network is decomposed into a set of subnetworks of the same depth as the original network, each of which is trained locally, before the various subnets are exchanged and the process is repeated. IST training has many advantages including reduction of communication volume and frequency, implicit extension to model parallelism, and memory limit decrease in each compute site. At the system level, I believe that proper computational and implementation abstractions will allow for the construction of self-configuring, declarative ML systems, especially when the goal is to execute tensor operations for ML in a distributed environment, or partitioned across multiple AI accelerators (ASICs). To this end, I first introduce a tensor relational algebra (TRA), which is expressive to encode any tensor operation that can be written in the Einstein notation, and then consider how TRA expressions can be re-written into an implementation algebra (IA) that enables effective implementation in a distributed environment, as well as how expressions in the IA can be optimized. The empirical study shows that the optimized implementation provided by IA can reach or even out-perform carefully engineered HPC or ML systems for large scale tensor manipulations and ML workflows in distributed clusters.