Schedule: Computer Vision sessions

Siwei Lyu reviews the evolution of techniques behind the generation of fake media and discusses several projects in digital media forensics for the detection of fake media, with a special focus on recent work on detecting AI-generated fake videos (DeepFakes). Read more.

At the core of today's problems with image classification and deep learning lies one fundamental truth: most AI systems operate by choosing the path of least resistance, not the path of highest long-term quality. Matt Zeiler discusses Clarifai's approach to closing the loop on AI and the techniques it employs to counter the AI quality regression phenomenon. Read more.

Deep neural networks (DNNs) have enabled AI breakthroughs, but serving DNNs at scale has been challenging: Fast and cheap? Won’t be accurate. Fast and accurate? Won’t be cheap. Join Ted Way, Maharshi Patel, and Aishani Bhalla to learn how to use Python and TensorFlow to train and deploy computer vision models on Intel FPGAs with Azure Machine Learning and Project Brainwave. Read more.

There's significant interest in applying deep learning-based solutions to problems in medicine and healthcare. Eric Oermann and Katie Link identify actionable medical problems, recast them as tractable deep learning problems, and discuss techniques to solve them. Read more.

Anoop Katti explores the shortcomings of the existing techniques for understanding 2D documents and offers an overview of the Character Grid (Chargrid), a new processing pipeline pioneered by data scientists at SAP. Read more.

Machine learning models are often susceptible to adversarial deception of their input at test time, which leads to poorer performance. Alina Matyukhina investigates the feasibility of deception in source code attribution techniques in real-world environments and explores attack scenarios on users' identities in open source projects—along with possible protection methods. Read more.

Artificial intelligence has shown great potential to revolutionize clinical medicine and healthcare delivery. However, incorporating these algorithms into clinical workflows involves a big challenge: convincing clinicians and regulators to trust a “black box” solution. Behrooz Hashemian explains how he's helping make deep neural networks interpretable to provide evidence for clinical decisions. Read more.

While deep learning has led to many advancements in computer vision, most research has focused on ground-based imagery. Ryan Mukherjee and Neil Fendley offer an overview of functional Map of the World (fMoW), an ImageNet for satellite imagery built to address this issue, and explain how you can attack or defend these deep learning models. Read more.

Clinical radiology currently faces several clinical issues: improving imaging efficiency, reducing risks, and developing higher imaging quality. Enhao Gong and Greg Zaharchuk explain how Subtle Medical's deep learning/AI solution addresses these problems by enabling faster MRI and faster PET and low-dose scans, providing real clinical and financial benefit to hospitals. Read more.

Humayun Irshad offers an overview of an active learning framework that uses a crowdsourcing approach to solve parking sign recognition—a real-world problem in transportation and autonomous driving for which a large amount of unlabeled data is available. The solution generates an accurate model, quickly and cost-effectively, despite the unevenness of the data. Read more.