RESEARCH SPOTLIGHT
Hi, I am
Fan Feng
HOST INSTITUTE
Aston University
PROJECT TITLE
Probabilistic Photonic Machine Learning
KEY WORDS
Probabilistic photonic machine learning, photonic generative models, photonic neuromorphic computing
RESEARCH AIM
My research aims to develop reliable AI systems that compute using light so that complex tasks, such as generating images and making predictions under uncertainty, can be performed with less energy.
RESEARCH OVERVIEW
My research explores how light can perform the calculations behind artificial intelligence. I develop methods to help light-based computers handle uncertainty and remain reliable when their components are imperfect or change over time. I also investigate how these systems can generate new images and process information using approaches inspired by the brain. By designing AI algorithms together with the optical devices that run them, I aim to build reliable, energy-efficient computing systems for future AI applications.
RESEARCH CHALLENGE
The main challenge is making light-based AI reliable while realising its potential to reduce energy use. Optical devices are affected by noise, manufacturing imperfections, and changes over time, which can lead to inaccurate predictions or poor-quality generated images. My research develops learning methods that account for these physical limitations and uncertainty, so light-based systems can perform useful AI tasks consistently and efficiently.
RESEARCH INNOVATION
My approach combines an understanding of how optical devices behave with AI methods that explicitly account for uncertainty. This helps the system estimate changes inside the hardware and adapt its calculations accordingly. I bring this perspective to image generation and brain-inspired computing, designing the learning algorithms and optical hardware together. The innovation lies in connecting the physics of the devices directly to the reliability and quality of the AI’s outputs.
RESEARCH IMPACT
What excites me most is seeing a light-based system turn random inputs into recognisable images. It connects the physics of optical devices with AI’s ability to learn and create. I also enjoy discovering how understanding the imperfections of real hardware can help us design more reliable systems, with the potential to use less energy.
RESEARCH SUMMARY
My research focuses on probabilistic photonic machine learning, photonic generative models, and photonic neuromorphic computing. I develop physics-informed Bayesian methods to infer hidden device states and quantify uncertainty in analogue photonic neural networks, with the aim of maintaining reliable computation under fabrication variability, noise, and temporal drift. For generative modelling, I investigate diffusion and flow-matching approaches that map stochastic inputs to structured images through photonic neural architectures, using hardware-aware training and model distillation to accommodate quantisation and physical device constraints. In photonic neuromorphic computing, I study how optical linear transformations and nonlinear responses can support neural information processing. These directions share a focus on co-designing learning algorithms and photonic hardware to balance computational accuracy, generative fidelity, and implementation cost, towards reliable and energy-efficient optical AI systems.
RESEARCH VISUAL
RESEARCH JOURNEY
RESEARCH SECONDMENT
1. CNRS (French National Centre for Scientific Research)
- Country: France
- Duration: 3 months
- Main objectives: Develop probabilistic photonic machine learning methods and investigate photonic neuromorphic computing architectures.
- Expected skills: Probabilistic modelling, photonic neural network design, and learning algorithms that account for optical device behaviour.
2. Hewlett Packard Enterprise (HPE)
- Country: To be confirmed
- Duration: To be confirmed
- Main objectives: Develop and evaluate photonic generative models, focusing on implementing image generation algorithms within photonic hardware constraints.
- Expected skills: Generative modelling, joint design of algorithms and photonic hardware, and evaluation of generation quality and computational efficiency.
LEARNING GOALS
I am most looking forward to learning photonic integrated circuit (PIC) design and hardware–software co-design. I want to understand how to translate AI models into practical photonic circuits and design algorithms around their physical capabilities and constraints. These skills will help me connect my research in probabilistic learning and generative models with practical hardware implementation.
RESEARCH BENEFITS
The secondments will help me connect my AI models with practical photonic hardware. At CNRS, I aim to strengthen the probabilistic and neuromorphic foundations of my research, while at HPE I will focus on photonic generative models. Learning PIC design and hardware–software co-design will help me identify realistic device constraints, refine my algorithms, and develop models that are better suited to physical implementation.
CAREER DEVELOPMENT
These secondments will help me build a career in photonic AI by strengthening my skills in PIC design and hardware–software co-design. Experience at CNRS and HPE will give me insight into both academic research and industrial development, helping me translate research ideas into practical technologies. They will also help me build professional connections and prepare for interdisciplinary research and development roles.
MORE THAN PHD
Why did you decide to join POSTDIGITAL+ ?
I decided to join POSTDIGITAL+ because of its many opportunities for collaboration and exchange. Its broad research scope and strong interdisciplinary focus allow me to connect photonics, machine learning, and hardware design. I value the chance to learn from researchers with different backgrounds, share ideas across institutions, and develop my research through both academic and industrial perspectives.
What do you think is the biggest advantage of being part of the International Doctoral Network?
The biggest advantage is learning from people with different expertise across countries, universities, and industry. For my research, this means connecting AI methods with photonic device design and practical engineering. These collaborations can bring fresh perspectives to difficult problems, give me access to complementary skills and facilities, and build lasting relationships that support my future career.
What skill have you developed the most so far?
The skill I have developed most is adapting machine learning algorithms to the physical constraints of photonic hardware. I have learned to account for noise, limited precision, and device behaviour when designing and training models. This has helped me connect abstract algorithms with what a physical system can realistically achieve.
What are your career ambitions after completing your doctorate?
After completing my doctorate, I plan to pursue a postdoctoral research position in photonic AI. I want to deepen my expertise in probabilistic learning, photonic generative models, and hardware–software co-design, while developing greater research independence and building international collaborations.
What inspired you to pursue this research Area?
I was inspired by the possibility of using light to make AI more energy-efficient. The idea that physical devices could learn, handle uncertainty, and generate images fascinated me. This research brings together my interests in physics and machine learning, while offering the opportunity to turn mathematical ideas into practical computing technologies.
What surprised you most since starting your PhD?
What surprised me most is how rewarding it feels to make even a small step forward on a difficult problem at the frontier of research. I also hadn’t expected so much independence in choosing my methods and setting the pace of my work. Taking ownership of these decisions has been challenging, but it has made the progress feel much more meaningful.