Research Scientist (6-month contract), SCIS

The candidate will be responsible for leading and conducting research on (a) developing techniques to incorporate memory of past recent inference executions to optimize embodied AI model inference in persistent environments, and (b) subsequently, incorporating the generalized reasoning ability of LLMs/VLMs to support optimized execution of reasoning and planification in embodied AI models. Successful candidate will be part of an active research team led by Prof Archan Misra from School of Computing and Information Systems, Singapore Management University.

The candidate will have the following key duties:

• Develop and implement new DNN models and structures to include memory embeddings of past inference execution to support accelerated and optimized execution of large, embodied AI models.
• Generate appropriate datasets (including enhanced models to generate virtual world data and associated query sequences designed to support variable amounts of cached reasoning) to support training and generalization of such optimization approaches over state-of-the-art large vision-language (VLM) models.
• Create and validate novel spatiotemporal optimizations of such hybrid DNN inferencing pipelines to reduce latency of visual grounding AI models without adversely affecting accuracy.
• Provide leadership to enable the submission of 1-2 high-quality research publications and public release of benchmark datasets.
• Other duties as assigned.

Qualifications

• Minimum Ph.D. in Computer Science/Electrical Engineering or relevant disciplines from a reputable institute, with particular focus on optimized execution of complex computational models.
• 3+ years real-world experience AI algorithms for human-object interaction and video-based physical world understanding and reasoning.
• Proven track record and professional experience in multi-modal data analysis and AI model development for multimedia and vision applications.
• 3 years of project-level experience with video transformer-based AI models and their use in action recognition/prediction and/or video grounding tasks.
• Ability to work cooperatively as part of a small, agile academic research team is essential.
• High-quality publications in high-quality academic conferences and journals, especially in the area of Computer Vision.