Integrated sensing and communication (ISAC) is a key technology for the sixth-generation (6G) wireless networks, where the same spectral and hardware resources are used for both communication and environmental sensing. Many optimization problems in ISAC require accurate sensing and communication channel models, which are often difficult to obtain. Machine learning (ML) is a powerful tool for solving ISAC problems by enabling data-driven solutions that can bypass the reliance on explicit models. This talk will explore how ML techniques can improve ISAC performance beyond traditional optimization approaches. Two case studies will be discussed: sensing-assisted predictive beamforming and cooperative sensing through ML. These examples will demonstrate the potential of ML to enable end-to-end signal processing for ISAC in 6G wireless networks.

In this talk, I will begin with a brief introduction to quantum computing, highlighting the importance and opportunities for pursuing fundamental research in the quantum Internet. In particular, I will discuss how quantum networks can enable quantum information transmission, parallel processing, and distributed processing. Next, I will introduce online learning theory and explain how it can help us explore compelling challenges in building quantum networks and the quantum Internet. To this end, I will delve into the quantum path selection problem, as well as the quantum border gateway protocol (QBGP) if time allows. Finally, I will outline several exciting open research problems at the intersection of quantum networks and quantum computing.

Like water that adapts to any container, documents need Natural Language Processing (NLP) systems that adapt and move across modalities pages and scales to find verifiable evidence. In this talk, I will share a practical agenda to build NLP systems that ingest text, images, layout, tables and figures and produce traceable answers. We emphasise three pillars: integration, learning and retrieval. Integration: fuse multimodal features and layout aware encodings so text and visual content are interpreted together. Learning: train specialist teachers across modalities and distil their feature knowledge into compact deployable students for NLP tasks. Retrieval: adopt a retrieval first approach, using multipage and multimodal retrieval to find candidate passages, tables and figures, then chain those candidates into a clear evidence trail. I demonstrate how graph-based encodings and multiscale reasoning work together, and how multiteacher distillation compacts expert knowledge into deployable students. Then, with concise multimodal case studies and retrieval centric metrics, I show measurable gains in evidence grounding, generalisation and operational readiness. I conclude with practical measures to control complexity and annotation cost, and present simple experiments and evaluation criteria for different domains.

In an era dominated by massive foundation models, smaller players risk being left behind—unable to afford the scale, data, or manpower that large AI systems demand. This talk introduces the concept of Prudent AI—an approach that emphasizes right-sized, lightweight, and explainable intelligence delivered through just-in-time, Plug-and-Play AI Boxes. Focusing on applications like early anomaly detection in multivariate time series, we demonstrate how our AI Boxes use sparse data, minimal compute, and human-guided refinement to detect rare but critical events. The architecture integrates symbolic reasoning, data-driven refinement, and secure edge deployment, showing how being small can actually be a strength in resource-constrained settings. Through this, we reimagine how organizations can adopt AI that is transparent, agile, and sustainable.