FUTURES EXPO SERIES
CRUISE Lab
Key capabilities
Trustworthy , multimodal machine-learning and explainable agents in complex environments
Research Laboratory
CRUISE Lab addresses the need for transparency and adaptability in AI systems . The behaviour of machinelearning ( ML ) models needs to be properly understood . However , high-quality training data is often not available . In such cases , training a deep learning model requires a new data-efficient learning paradigm , without the need for high volumes of labelled data .
This team specialises in developing trustworthy ML techniques and models . The team has investigated a novel framework that uses GNN-based visualisation to explain the learning process of an agent in a complex environment .
Additionally , the team has developed a robust framework for continuous improvement of pretrained models , such as for question answering with multimodal data and for videolanguage tasks , in a continual learning setting . ViLCo-Bench , the first benchmark for video-language continual learning , which leverages memory efficient architecture , enables faster model training and more robust model evaluations .
The team has developed prototypes of both frameworks . Its work is at the proof-of-concept stage , and it is seeking collaboration and investment to refine these solutions and integrate trustworthiness in AI and machine learning .
> Multimodal machine learning
> Trustworthy machine learning ( explainable , robust , fair )
> Situational awareness and spatiotemporal intelligence
Differentiators
> Deep understanding of multimodal sensor fusion , robustness , and / or biases inherent to machine learning models and their implications
> Expertise in a broad suite of technologies , including deep learning , interpretable machine learning , reinforcement learning , robustness , and benchmarking
> Experience in building state-of-the-art AI algorithms and research solutions that have been adopted by many industry partners
Key customers
> Government agencies > Defence industry
Key partnerships
> US Department of Defence > US Airforce Research
Laboratory
cruiseresearchgroup . github . io / expo
54 •