Machine learning for real-time detection of Fast Radio Bursts with the uGMRT
The National Centre for Radio Astrophysics, Tata Institute of Fundamental Research (NCRA-TIFR), Pune, India, is seeking a Post-Doctoral Fellow to work with Dr Jayanta Roy on machine-learning algorithms for the detection of Fast Radio Bursts (FRBs) with the uGMRT.
Dr. Roy and the team are now building a software digital processing system to enable real-time commensal search for FRBs and pulsars with a PetaFlop system. The instrument will host a large number of A100 GPUs, several tens of TB of memory and Petabyte storage, executing real-time HPC and AI applications to ensure simultaneous detection and arc-second localisation capability over the full 300-1460 MHz radio spectrum. The design needs to separate genuine cosmological signals from interference to optimise the number of triggers for baseband recording and imaging localisation. Being one of the large-scale surveys with thousands of interferometric beams required to cover the field of view, this enables high-impact science in time-domain astronomy with the GMRT as well as provides a technology pathway for next-generation instruments. The team at NCRA is working with other national and international collaborators including industry partner NVIDIA to develop the machine learning/deep learning pipeline. The post-doctoral fellow will work with the AI team including experts from NVIDIA to build a memory bandwidth optimised pipeline aided by the mixed-precision arithmetic available on Ampere architecture. TensorFlow design with TensorRT optimization while minimizing accuracy loss will be attempted for quantising the network weights to achieve a faster and smaller-sized network. In order to achieve such automated classification, robust training sets generated from a human investigation of existing GMRT results will be needed. In addition, the post-doctoral fellow will work on data interpretation and visualisation aiming to discover new events. The fully real-time implementation of such an AI pipeline will enable the piggyback search capability at the GMRT resulting in high-impact discoveries aided with localisation of these bursts to use them as cosmological probes while studying the intergalactic medium. Prior experience in the areas of machine learning techniques and good programming skills, as well as some exposure to time-domain astronomy, would be beneficial for this project.
Research activities at NCRA-TIFR are centred on low-frequency radio astronomy, with faculty members pursuing research in diverse areas of astrophysics. NCRA-TIFR operates two large observational facilities, the Giant Metrewave Radio Telescope and the Ooty Radio Telescope.
Applications containing a curriculum vitae, a research plan, and the names and e-mail addresses of three or more referees should be sent to the Dean, NCRA-TIFR Faculty, at the e-mail address deanncra@ncra.tifr.res.in, by November 30, 2022. Applicants should also request their referees to send in recommendation letters by the same deadline. All enquiries should also be made to this e-mail address. All applicants will automatically be considered for all available Post-Doctoral Fellowships at NCRA-TIFR.
The monthly salary of Post-Doctoral Fellows at NCRA-TIFR depends on experience after the doctoral degree. Candidates are required to have at least a provisional Ph.D. degree when taking up the position. For candidates who have received their doctoral degrees, the monthly salary ranges from INR 47,000 to INR 54,000, depending on experience. Post-doctoral fellows are also entitled to a contingency grant of INR 40,000 per year. All positions are tenable for a maximum of 3 years, with continuations subject to a yearly review.
Post-doctoral Fellows at NCRA-TIFR are provided a travel grant of INR 3,00,000 over the course of the fellowship to enable them to travel for international conferences or for research collaborations. They are also entitled to a house rent allowance, and medical coverage under NCRA-TIFR's medical scheme. Moreover, some additional resource funding is expected to be available from industry partner NVIDIA to facilitate knowledge exchange.
Document Actions