Optimistic Learning for Communication Networks

TL;DR

Optimistic Learning (OpL) integrates offline predictive models with online convex optimization to accelerate decision-making in dynamic communication networks while preserving online robustness. By incorporating gradient or function predictions into online algorithms (notably OFTRL and related variants), OpL achieves regret that scales with the prediction error, up to $\mathcal{O}(1)$ when forecasts are accurate, and gracefully degrades to standard OCO bounds otherwise. The tutorial develops the theory (definitions, regret bounds, adaptivity), presents multiple OpL algorithms, and demonstrates applications to caching, edge computing, network slicing, and O-RAN workload assignment, including memory-aware and discrete-placement problems. It also outlines future directions, including hybrid optimism, SEA models, non-convex settings, and joint predictor-learner design, to broaden OpL’s applicability in next-generation networks. Overall, OpL provides a principled, universal framework for leveraging predictive information in network control with strong worst-case guarantees and practical performance gains.

Abstract

AI/ML-based tools are at the forefront of resource management solutions for communication networks. Deep learning, in particular, is highly effective in facilitating fast and high-performing decision-making whenever representative training data is available to build offline accurate models. Conversely, online learning solutions do not require training and enable adaptive decisions based on runtime observations, alas are often overly conservative. This extensive tutorial proposes the use of optimistic learning (OpL) as a decision engine for resource management frameworks in modern communication systems. When properly designed, such solutions can achieve fast and high-performing decisions -- comparable to offline-trained models -- while preserving the robustness and performance guarantees of the respective online learning approaches. We introduce the fundamental concepts, algorithms and results of OpL, discuss the roots of this theory and present different approaches to defining and achieving optimism. We proceed to showcase how OpL can enhance resource management in communication networks for several key problems such as caching, edge computing, network slicing, and workload assignment in decentralized O-RAN platforms. Finally, we discuss the open challenges that must be addressed to unlock the full potential of this new resource management approach.

Figures (17)

• Figure 1: (a): OCO-based transmission control in a wireless network with fast-changing channel gains $\bm w_t$, mertikopoulos-iot. (b): Optimistic Learning-based transmission control in a wireless networks, using channel gain predictions $\bm{ \tilde{w}}_t$.
• Figure 2: Paper Organization: Sections and Main Results.
• Figure 3: The typical OCO template for the interaction between a learner and an oblivious adversary. The learning algorithm $\mathcal{A}$ observes the past decisions of the learner and past decisions (i.e., functions) of the adversary and yields the next action.
• Figure 4: Sequence of events in OCO: (i) The learner makes its decision $\bm x_t$; (ii) the adversary decides the cost function $f_t(\cdot)$ and the learner observes it; (iii) the learner updates its parameters ($\sigma_t$ or $\eta_t$).
• Figure 5: (a): The decision space is $x\in [-1,1]\subset \mathbb R$, the objective (cost) function is linear, $f_t=c_tx$, and the cost parameters $c_t$ change in each slot, alternating -1 and 1. FTL is unstable (ping-pong) as it is heavily influenced by the sign of the aggregate cost at each slot; while FTRL is converging to a stable decision. (b): The evolution of average regret for different time windows $t\!=\!1,\ldots,T$ (with $T\!=\!50$) for the two algorithms demonstrates that FTL does not learn.