Three questions on Chapter 2: Machine Learning System

[qyy2003]

1. Where do small GPU servers and consumer-grade desktops/laptops fit within the ML deployment spectrum? How do these common equipment fit into this landscape? Do these devices serve as intermediate platforms between cloud and edge computing, or should they be considered local computing nodes at the edge?

2. What are the key research challenges and opportunities beyond engineering implementation? While the multi-tier hierarchy is well-structured for deployment, what open research questions remain in optimizing ML across these layers? What are the fundamental trade-offs, and are there emerging paradigms that could redefine this architecture?

3. Are there practical toy examples or lab exercises for hands-on experimentation with this hierarchy? Given access to the Cloud, Edge, Mobile, and Tiny ML devices listed in Table 2.1, how can we design hands-on experiments to better understand their interactions and constraints?