Description
In this highly visible role, you will invent the next generation of perceptual loss functions used across Apple’s imaging ecosystem. Your work will span algorithm development, theoretical analysis, and deployment at scale.
Responsibilities
- Create and advance perception-aligned measures of image distortion, and integrate them into training pipelines for neural compressors and other cutting edge learned image processing algorithms
- Collaborate with partners to develop algorithms trained by these next-generation perceptual loss functions, such as learned image and texture compression models that optimize for realism, fidelity, and rate efficiency
- Collaborate closely with cross-functional teams—including sensor architecture, ISP/SoC design, display HW, Vision Science, and applied ML groups—to bring research ideas into production
- Perform detailed experimentation to evaluate model fidelity to perception and to evaluate optimization performance for different image processing algorithms
- Prototype and validate new algorithms that meaningfully push rate–distortion–realism boundaries beyond state-of-the-art
- You will join a world-class organization of researchers and engineers who value creativity, rigor, and beautiful solutions to hard problems.
Minimum Qualifications
- Bachelors Degree in Computer Science, Electrical and Computer Engineering, Neuroscience, Vision Science, or equivalent and 3+ years of relevant experience
- Experience in translating complex mathematical concepts into practical algorithms aligned with perceived image realism or quality
- Experience with full-reference or no-reference image metrics, generative modeling, optimization, or realism-driven evaluation frameworks
Preferred Qualifications
- Masters or Ph.D. in Computer Science, Electrical and Computer Engineering, Neuroscience, Vision Science, or equivalent
- Experience in information theory, probabilistic modeling, and/or machine learning
- Experience in Python and modern ML frameworks such as PyTorch, TensorFlow, and JAX
- Deep expertise in image compression, texture modeling, and rate-distortion optimization, with demonstrated ability to design new metrics and algorithms that outperform classical approaches
- Publication record in machine learning, compression, or information theory venues (NeurIPS, ICLR, ICML, ISIT, or related)
- Hands-on experience building learned compression systems end-to-end, including model design, training pipelines, ablations, and integration into large-scale frameworks
- Internship or industry experience integrating research models into production-scale frameworks is a strong plus
- Basic knowledge of human visual perception is a strong plus
- Strong analytical & critical thinking, and creative problem-solving skills
- Excellent written and verbal communication skills in English
- Excellent communication, collaboration, and scientific writing skills
- Basic understanding of digital imaging, and display software and hardware
- Swift/Metal programming is a plus
