Javier Girado

• Applying Deep Learning Neural Network (NN) to drone imagery: Car brand/model counting using 3D point cloud samples generated by drone imagery for training/testing and process it (run inference) using a 2D/3D Convolution Auto-encoder neural network. Working on a Variational Auto-encoder • Supervised Master thesis: DeConvolutional (DCNN) Generative Adversarial Network (GAN) on FPGA • nVidia GPU performance/cost evaluation across multiple platforms for Deep Learning application to reduce GPU cost (NFS Grant CHASE-CI). Performed tests on Amazon Web Service (AWS), Extreme Science and Engineering Discovery Environment (XSEDE), San Diego Super-Computer Center (SDSC) and our Linux-based cluster machines. Used AlexNet (ConvNN) to train with ImageNet (1.4 million images) and MNIST (60,000 handwritten digits). Framework: CUDA + cuDNN + Caffe and TensorFlow • Depth/Disparity map from Stereo using Deep Learning (ongoing). Use nVidia Jetson TX1 (Tegra) + ZED stereo camera to collect training samples and use computer vision techniques for ground-truth. Research if deep learning can Improve the disparity map using 3D reconstruction and scene understanding. Recover shape by matching across within-class shape and appearance variations • KidPrint project: Apply deep learning to improve the fingerprint detection and recognition for babies (0~3 months). Current algorithms work very well for adults but not for babies where the skin is too soft

- GPU architectural exploration which results in 3 proposals to use texture pipe as an image co-processor leveraging existing GPU architecture - ISA improvement for Image Processing - Optimization and/or profiling with focus on image processing - New Intrinsic functions in order to expose proprietary ISA - Deep knowledge of GPU Assembly language - OpenGL v2.x, DirectX v11.x, OpenCL v1.x - Program new features (or new blocks) for our GPU HLM simulator and OpenGL/OpenCL/DX11 test drivers - Review patents and research papers per request - Train new-hires and lead group collaboration - Develop a Microsoft (MS) DirectX v11.x test application framework - 3 patents (pending) - 2 Qualstar awards

- Integrated real-time 3D head position tracker systems (Ph.D. thesis) with multiple cameras (using face detection and recognition neural networks) and VarrierTM autostereoscopic displays for commercial applications - Researched and developed advanced, super-high resolution auto-stereoscopic virtual reality (VR) devices, incorporating and displaying 3D video sources

- Implementation of several algorithms (Computer Vision, Wavelet transform and Face Detection/Recognition with Neural Network) using Graphics Processor Units (GPU)

- Object Oriented Programming - Smalltalk - VisualAge Smalltalk