Research Associate in Deep Learning for Computational Lightfield Microscopy

Applications are invited for the above post to work with Professor Christos Bouganis and his team at Imperial College London for a project.

The successful candidate will be integral to delivering on the project called “Optical Oscilloscope: Real-time, High-throughput, Volumetric Voltage Imaging.” Our goal is to enable real-time, kilohertz, volumetric voltage imaging in 1,000 cells simultaneously within scattering mammalian brain tissue. The project is driven by a transdisciplinary consortium led by Dr Foust (Imperial, Bioengineering), and includes Professor Pier Luigi Dragotti who is an expert in machine learning, signal processing and computational imaging (Imperial, EEE), Professor Christos Bouganis (Imperial, EEE); and Dr Samuel Barnes (Imperial, Brain Sciences). 

The successful candidate will research and develop a computational platform based on GPU and FPGA devices that would accelerate the execution of deep neural networks designed by the rest of the team for volume reconstruction from lightfield video sequences produced by our lightfield microscope.

You will design, build and optimize a platform based on GPUs and FPGAs that aims to accelerate the computation of new model-based deep neural networks (DNN) for real-time neural activity extraction from lightfield microscopy data. You will research and build the infrastructure that would enable the communication of the CPU-GPU-FPGA subsystems, in order to achieve low-latency and high-throughput. You will investigate and implement strategies for the mapping of the DNN architectures for volume reconstruction developed by the rest of the team to the available computational platforms in order to minimise the latency of the computation. You will also consider the optimisation of the DNN models by investigating different quantisation methods taking into account the available hardware resources.

You will also work with postdoctoral associates in EEE and Bioengineering to develop algorithms for volume reconstruction that can be efficiently mapped into hardware.

The successful candidate:

Will have a PhD (or be very near completion or equivalent) in engineering, mathematics, physics, or a related topic. Experience with digital design techniques and familiarity in designing hardware targeting FPGAs. Experience in deploying hardware designs in FPGA platforms. Experience with programming in CUDA. The successful candidate will have a good publication record, show evidence of working well in teams, and demonstrate an ability to work to tight deadlines. Preference will be given to candidates with a strong background in digital hardware design and to those with experience with programming a GPU. The ideal candidate will be both a “tool developer” and a “tool user” with a keen interest in system design.

The opportunity to form a key part of the multidisciplinary team who will innovate methods for high-throughput, volumetric voltage imagingTraining in system architecture design and design of hardware accelerators for DNNsClose interaction with the highly motivated and diverse team of Prof. Bouganis all working on researching hardware architectures for the acceleration of Deep Neural Networks ()First access to spatially-resolved cortical microcircuit voltage data acquired on a previously inaccessible scale. Flexibility to develop your own hardware architectures/systems and research vision based on this new and exciting area.A diverse and supportive training environment promoting transdisciplinary fluencyFurther professional training and network development provided by The opportunity to continue your career at a world-leading institutionSector-leading salary and remuneration package