Deep Learning Architectures for Image Classification and Object Detection

Object detection is a task in computer vision with many practical applications that can now be achieved with super-human levels of performance on selected benchmarks using deep neural networks. In this talk we define the object detection task and present J. Redmon's YOLO (You Only Look Once) V3 deep neural network architecture. As preliminaries to object detection and YOLOv3, we first describe image classification on the Pascal VOC and ImageNet benchmark datasets, and introduce a series of deep learning neural network architectures that include the multilayer perceptron (MLP), convolutional neural networks (CNNs), and other networks with dystopian names such as AlexNet, GoogLeNet/Inception, VGG16, ResNet, and Region-CNN (R-CNN). We conclude with note of recent developments, including capsule networks (CapNets) by G. Hinton and deep networks with visual feedback. Slides and notebooks with code will be available after the talk.

Speaker Bio:
Nelson Correa is a data scientist and machine learning consultant based in West Palm Beach. He has a Ph.D. in Electrical Engineering and over 25 years of experience in natural language processing at three startup companies and IBM Research. He has over 30 technical publications and three U.S. patents. His current interest is in developing connections between deep learning and symbolic models for natural language processing and perception.

Event Details:
Doors at 6:30 for pizza, drinks, and lively discussion. Main talk will begin around 7:15.