ID R&D, a Mitek company (NASDAQ: MITK), is a leading research-driven company in the biometrics space, focused on being first to market with category-defining products that solve some of the most pressing technical challenges in identity security. Our products apply AI and computer vision to detect fraud in ways that make identity solutions more secure and convenient. We are on a mission to deliver a new generation of “zero-effort” identity verification and authentication. Our team is led by some of the most distinguished biometrics professionals in the industry with decades of business and scientific experience enabling the research and development of groundbreaking, world-class voice, face, and behavioral biometric solutions. Our vibrant culture integrates the flexibility and innovation of a startup with the strength and security of our parent company. Learn more at www.idrnd.ai.
At ID R&D, we believe that teams are stronger and more innovative when they have the range of opinions and perspectives that come from diversity. Very simply, we know that diversity is a foundational strength for our organization. A diverse workforce makes for stronger teams as a wider breadth of thinking and experience broadens the power of our collective mind. Building stronger teams is foundational to serving our customers comprehensively, innovating our products and solutions, and enabling more impactful decision-making across our business.
We are Virtual 1st! Whether you choose to work remotely from your home office or in-person from one of ID R&D's offices, our practices, processes and tools are designed to enable your success. At ID R&D, the Future of Work is about flexibility and preference wherever and whenever we are working.
PROJECT ABSTRACT: Research of Black-Box Adversarial Attacks on Computer Vision Systems.
When someone attempts to fool a face recognition system by using digital image transformations (color distortions, compression, geometric transforms, filters, etc), it's known as a "spoofing" attack. Facial liveness detection aims to defend against these attacks by verifying that the presented image is not a transformed image or video but is an alive, original photo of the face.
Each transformation incorporates numerous parameters that can't feasibly be iterated over within a practical time frame. To fortify the final ensemble and enhance its resilience, it's crucial to scrutinize all transforms that could potentially pose a threat.
In this project, we propose the construction of a Reinforcement Learning system designed to emulate real-world scenarios for evading machine learning systems.
What you will do
- Propose a structure for a reinforcement learning system that can simulate real-world scenarios of attempts to evade facial liveness systems. Using the reinforcement learning system, simulate different image transformations, taking into consideration the multiple parameters each transformation can have. Evaluate the performance of the system in detecting these transformations.
- Optimize and Iterate: continuously refine the reinforcement learning system, iterating over different transformations and making adjustments to improve detection effectiveness.
- Prepare a presentation to communicate your findings, showing how the reinforcement learning system enhances the ability of facial recognition systems to defend against spoofing attacks.
Desired studies, skills and experience
- Some knowledge and experience in computer vision, machine learning, or image processing.
- Good Python3 programming skills.
- Fluency in written and spoken English (you will use it on a daily basis for internal communication).
- Experience in one or more deep learning frameworks: TensorFlow or Pytorch.
- Familiarity with Linux/Docker/Git.
- A willingness to learn and strong communication skills.
- Live within a commutable distance of our office near Barcelona.
What we offer you
- An internship with a flexible start and duration (3 to 6 months), depending on the university agreement and student availability.
- Participate in relevant initiatives, learning and using the latest IT stacks in AI/ML and Computer Vision.
- Be part of a cutting-edge machine learning company (ID R&D) that belongs to one of the global players in the ID verification industry, committed to pushing the boundaries of AI and biometrics. We encourage all our team members to engage in scientific publications based on their work and actively participate in challenges. Our teams were rated in the top 10 for Kaggle Deepfake detection, showcasing our expertise and dedication to excellence.
- Competitive monthly compensation and time flexibility.
€900 - €1,200 a month