Experience

• 

  FUJISOFT America, Inc.

  • United States
  • IT Services and IT Consulting
  • 1 - 100 Employee

  • Artificial Intelligence Engineer

    • Jan 2019 - Present

    AI research and development:Dimensionality reduction:A generalized form of stochastic neighborhood embedding (locality sensitive hashing, backpropagation through perplexity; implementation using auto differentiation).A generalized form of laplacian eigenmaps.Implementation of all well known dimensionality reduction techniques.Genetic algorithms:Development of new hierarchical selection algorithms.Neural networks:Training neural networks of various types: convolutional, locally connected, recurrent, residual, neural ode, neural tangent kernel, bayesian, complex valued, graph, modular, morphological(attention, hopfield), non-linear connected (quadratic and higher order).Backpropagation through nodes(layers): adaptive activation functions, non-random noise functions, smooth Hausdorff loss (shape matching), generalized divergence, matrix decomposition, special matrices, general matrix functions.Backpropagation through nodes(layers) for images and 3d objects: spatial transformer networks, radon transforms, space filling tessellations, histogram, mipmap, ray tracing, sparse spatial graph.Other: softmode function, smooth l0 loss, median based optimizationMethods: automatic differentiation, automatic inversion, inceptionism (saliency maps).New matrix identities and applications.Receptive field analysis for fractional (transpose) convolution.3D convolutional neural networks:A simple Delaunay Triangulation algorithm for ordered points and convex hull simplification.Closed form solutions for collision with twisted non-convex finite elements (isoparametric mapping).Voxelization and interpolation of meshes and cloud points for use in convolutional neural networks.Surface integrals on triangular meshes over interpolated lattices (mesh decimation, remeshing over voxel boundaries, multivariate hilbert matrices).Time series analysis:Parallelized Dynamic Time Warping with least memory usage.GUI:GUI system (window system) using OpenGL from scratch.

  • Software Engineer

    • Sep 2016 - Jan 2019

    JavaScript:Web development within an application platform such as kintone.Python:Unsupervised anomaly detection of time series using filtering techniques and matrix profiling.Supervised learning anomaly detection with a neural network built using KERAS.Preprocessing images and postprocessing text for optical character recognition.C/C++:Anomaly detection of a large database based on rules given by the user.Ruby:Extracting components of an Excel spreadsheet by using a SQL database.Other:Built an automatic testing system which starts tests based on Excel spreadsheet input.

  • Software QA Engineer

    • Mar 2016 - Aug 2016

    Python:Wrote shell scripts to test android devices using ADB.

• 

  Skyline College

  • Mathematics Tutor

    • 2009 - 2010

    Tutored students in mathematics.