Schedule: Computer Vision sessions

Computer vision has led the artificial intelligence renaissance, and pushing it further forward is PyTorch, a flexible framework for training models. Mo Patel offers an overview of computer vision fundamentals and walks you through PyTorch code explanations for notable objection classification and object detection models. Read more.

Benoit Dherin explains how machine learning is applied to image classification, discusses evolving methods and challenges, and walks you through creating increasingly sophisticated image classification models using TensorFlow. Read more.

Dmytro Dzhulgakov explores PyTorch 1.0, from its start as a popular deep learning framework for flexible research to its evolution into an end-to-end platform for building and deploying AI models at production scale. Read more.

Many industries, including banking, financial sectors, and insurance, continuously face the problem of detecting fraudulent activities. Giorgia Fortuna explores state-of-the-art innovations in fraud detection and explains how unsupervised ML fits into the picture, focusing on signature checks and face recognition. Read more.

Machine learning and object recognition have matured to the point that exciting applications are now possible. Anmol Jagetia demonstrates how to create a Pokédex that uses a camera phone to recognize the Pokémon it's looking at in real time. You'll see how to gather data, prepare your dataset, tune models, and deploy it to a mobile device, using the same tech that is used in self-driving cars. Read more.

Neural networks are particularly vulnerable to adversarial inputs. Carefully designed perturbations can lead a well-trained model to misbehave, raising new concerns about safety-critical and security-critical applications. Pin-Yu Chen offers an overview of CLEVER, a comprehensive robustness measure that can be used to assess the robustness of any neural network classifiers. Read more.

Daeil Kim delineates the advantages of synthetic data and explains how to avoid traps that lead to dead zones and false positives. He also reviews work on simulations for synthetic data in application verticals in which it is traditionally difficult to manually acquire significant datasets. Read more.

eLife’s mission is to accelerate discovery and encourage responsible behaviors in science. Daniel Ecer and Paul Shannon detail eLife’s journey in using NLP, computer vision, and similarity algorithms to find more diverse peer reviewers, apply semantics to archive content, automate the submission process, and find insights into the sentiment of scholarly content. Read more.

How can AI impact national security? Collaborating with the UK Home Office Counterterrorism Unit, ASI Data Science built a tool that removes extremist propaganda from the web. Drawing on this experience, Marc Warner discusses the role of AI in the fight against terror and explains how shared access to this technology may be part of the answer. Read more.

Even in the age of big data, labeled data is a scarce resource in many machine learning use cases. Florian Wilhelm evaluates generative adversarial networks (GANs) when used to extract information from vehicle registrations under a varying amount of labeled data, compares the performance with supervised learning techniques, and demonstrates a significant improvement when using unlabeled data. Read more.

Transfer learning allows data scientists to leverage insights from large labeled datasets. The general idea of transfer learning is to use knowledge learned from tasks for which a lot of labeled data is available in settings where only little labelled data is available. Lars Hulstaert explains what transfer learning is and demonstrates how it can boost your NLP or CV pipelines. Read more.

In recent years, we’ve seen a shift from traditional vision algorithms to deep neural network algorithms. While many companies expect to move to deep learning for some or all of their algorithms, they may have a significant investment in classical vision. Paul Brasnett explains how to express and adapt a classical vision algorithm to become a trainable DNN. Read more.

Dramatic progress has been made in computer vision. Deep neural networks (DNNs) trained on millions of images can recognize thousands of different objects, and they can be customized to new use cases. Vanja Paunic and Patrick Buehler outline simple methods and tools that enable users to easily and quickly adapt Microsoft's state-of-the-art DNNs for use in their own computer vision solutions. Read more.

Detection of moving vessels with satellite sensors is a challenging problem. Satellite imagery is expensive, covers a very small area, and can be acquired only at predefined acquisition opportunities. Natalie Fridman dives into this challenging problem and shares ISI's AI-based solution along with successful examples of detecting maritime vessels with ISI's satellites. Read more.