Learning-augmented Online Minimization of Age of Information and Transmission Costs

TL;DR

The paper addresses minimizing the sum of transmission costs and AoI in a discrete-time, time-varying wireless setting by designing online algorithms with worst-case guarantees and augmenting them with ML predictions.A robust online algorithm (PDOA) is derived via a TCP-ACK reformulation and primal-dual analysis, achieving a non-asymptotic competitive ratio of $3$; a learning-augmented algorithm (LAPDOA) integrates ML predictions to attain consistency and robustness simultaneously.Extensive simulations on synthetic and real wireless traces demonstrate(PDOA's) practical performance improvements over prior online policies and show LAPDOA providing favorable tradeoffs between prediction-driven efficiency and worst-case guarantees.The work highlights the benefit of combining online optimization with ML predictions for AoI-aware scheduling, while also outlining limitations and directions for adaptive trust in predictions.

Abstract

We consider a discrete-time system where a resource-constrained source (e.g., a small sensor) transmits its time-sensitive data to a destination over a time-varying wireless channel. Each transmission incurs a fixed transmission cost (e.g., energy cost), and no transmission results in a staleness cost represented by the Age-of-Information. The source must balance the tradeoff between transmission and staleness costs. To address this challenge, we develop a robust online algorithm to minimize the sum of transmission and staleness costs, ensuring a worst-case performance guarantee. While online algorithms are robust, they are usually overly conservative and may have a poor average performance in typical scenarios. In contrast, by leveraging historical data and prediction models, machine learning (ML) algorithms perform well in average cases. However, they typically lack worst-case performance guarantees. To achieve the best of both worlds, we design a learning-augmented online algorithm that exhibits two desired properties: (i) consistency: closely approximating the optimal offline algorithm when the ML prediction is accurate and trusted; (ii) robustness: ensuring worst-case performance guarantee even ML predictions are inaccurate. Finally, we perform extensive simulations to show that our online algorithm performs well empirically and that our learning-augmented algorithm achieves both consistency and robustness.

Figures (11)

• Figure 1: An illustration of our system model. The device sends data to AP through an unreliable wireless channel. Assuming the current time slot is $t$. At first, the device probes to know the current channel state $s(t)$. Then, the device decides whether to transmit or not. There are three cases: (i) the channel is OFF, and the device does not transmit; (ii) the channel is ON, but the device chooses not to transmit; and (iii) the channel is ON, and the device transmits.
• Figure 2: The updates of primal variables $z_i(t)$ and dual variables $y_i(t)$ in the $k$-th ACK interval $[t_k+1, t_{k+1}]$ of PDOA, where channels are OFF during $[t_k+5, t_k+7]$. The x-axis represents time and the y-axis represents the packet id. PDOA makes two ACKs at slot $t_k$ and slot $t_{k+1}$, where the ACK cost $c = 18$. The primal variables $z_i(t)$ and dual variables $y_i(t)$ are updated from slot $t_k+1$ to slot $t_{k+1}$; and in slot $t$, packets are updated from packet $t_k+1$ to packet $t$. The red bold italic $1$ denotes when the ACK marker equals or is larger than $1$. In Fig. \ref{['fig:online_primal']}, the grey areas denote the updates due to Line \ref{['line:OFF_primal']}; In Fig. \ref{['fig:online_dual']}, the grey areas denote the updates due to Lines \ref{['line:final_update_begin']}-\ref{['line:final_update_end']}.
• Figure 3: The updates of variables in the $k$-th ACK interval $[t_k+1, t_{k+1}]$ of LAPDOA, where channels are OFF during $[t_k+5, t_k+7]$. LAPDOA makes two ACKs at $t_k$ and $t_{k+1}$, and the ML prediction $\mathcal{P}$ makes its $i$-th ACK at slot $t_k+4$. The red bold italic value denotes when the ACK marker $M \ge 1$. Let ${m_1} = \lambda /c$, ${m_2} = 1/\lambda c$, ${y_1} = \lambda$, and ${y_2} = 1$. The light grey area denotes the small updates, the white area (without background) denotes the big updates, and the dark grey area denotes the zero updates.
• Figure 4: Performance comparison of online algorithms under different datasets.
• Figure 5: Performance comparison of LAPDOA under different trust parameter $\lambda$ using synthetic dataset (higher prediction quality levels lead to better prediction accuracy).

Theorems & Definitions (6)

• proof
• proof : Proof of Theorem \ref{['them:online_3']}
• proof
• proof
• proof : Proof of Lemma \ref{['lemma:ml_robustness']}
• proof