MEDIA ANALYTICS
PROJECTS
PEOPLE
PUBLICATIONS
PATENTS
Besides privacy-aware learning, we also develop methods for privacy-aware sensing. In particular, we develop novel computational cameras that allow computer vision analysis even in sensitive environments like hospitals or smart homes. Our key innovation is a camera that removes private information. Our adversarial training approach allows simultaneously high accuracy and high privacy through learned phase masks, which are inserted in the focal plane of the camera. Our hardware prototypes allow tasks like depth estimation or action recognition, while ensuring private face information is hidden. We also develop prototype cameras that encrypt images to prevent deciphering by humans except with a secret key, while allowing computer vision analysis such as face recognition to be performed on the encrypted image.
Privacy impacts every stakeholder in the AI solution ecosystem, including consumers, operators, solution providers and regulators. This is especially true for applications such as healthcare, safety and finance which require collecting and analyzing highly sensitive data. We develop AI solutions to assure customers that private information is not leaked at any stage of the data lifecycle. Our differential privacy method guarantees that an adversary cannot decipher training data from model outputs. Our differential privacy method provides a provable guarantee of privacy, while using significantly less data than competitors. We also develop federated training methods that securely combine private data from multiple users or enterprises, at orders of magnitude lower communication costs than competitors, while providing a guarantee of low leakage through differential privacy.
Team Members: Francesco Pittaluga
Publication Tags: VISPR, privacy, camera, deep learning, differential privacy, visual privacy, computer vision, machine learning
With over a billion sold each year, cameras are not only becoming ubiquitous, but are driving progress in a wide range of domains such as mixed reality, robotics, and more. However, severe concerns regarding the privacy implications of camera-based solutions currently limit the range of environments
While deep face recognition has benefited significantly from large-scale labeled data, current research is focused on leveraging unlabeled data to further boost performance, reducing the cost of human annotation. Prior work has mostly been in controlled settings, where the labeled and unlabeled data
Blind video deblurring restores sharp frames from a blurry sequence without any prior. It is a challenging task because the blur due to camera shake, object movement and defocusing is heterogeneous in both temporal and spatial dimensions. Traditional methods train on datasets synthesized with a single
Data privacy has emerged as an important issue as data-driven deep learning has been an essential component of modern machine learning systems. For instance, there could be a potential privacy risk of machine learning systems via the model inversion attack, whose goal is to reconstruct the input data