Abstract:Decision-makers often have access to a machine-learned prediction about demand, referred to as advice, which can potentially be utilized in online decision-making processes for resource allocation. However, exploiting such advice poses challenges due to its potential inaccuracy. To address this issue, we propose a framework that enhances online resource allocation decisions with potentially unreliable machine-learned (ML) advice. We assume here that this advice is represented by a general convex uncertainty set for the demand vector.
We introduce a parameterized class of Pareto optimal online resource allocation algorithms that strike a balance between consistent and robust ratios. The consistent ratio measures the algorithm's performance (compared to the optimal hindsight solution) when the ML advice is accurate, while the robust ratio captures performance under an adversarial demand process when the advice is inaccurate. Specifically, in a C-Pareto optimal setting, we maximize the robust ratio while ensuring that the consistent ratio is at least C. Our proposed C-Pareto optimal algorithm is an adaptive protection level algorithm, which extends the classical fixed protection level algorithm introduced in Littlewood (2005) and Ball and Queyranne (2009). Solving a complex non-convex continuous optimization problem characterizes the adaptive protection level algorithm. To complement our algorithms, we present a simple method for computing the maximum achievable consistent ratio, which serves as an estimate for the maximum value of the ML advice. Additionally, we present numerical studies to evaluate the performance of our algorithm in comparison to benchmark algorithms. The results demonstrate that by adjusting the parameter C, our algorithms effectively strike a balance between worst-case and average performance, outperforming the benchmark algorithms.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to use machine - learning predictions (i.e., suggestions) in online resource allocation to improve decision - making effectiveness while ensuring good performance even when the predictions are inaccurate. Specifically, the paper focuses on how to effectively use the demand predictions provided by machine learning in a convex uncertainty set to optimize resource allocation strategies.
### Background and Challenges
1. **Resource Allocation Problem**: Enterprises need to allocate sequentially arriving requests within a limited resource inventory to maximize revenue. This has applications in multiple fields such as the aviation industry, the retail industry, the hotel industry, and online advertising.
2. **Real - Time Decision - Making Challenges**: When making real - time decisions, enterprises lack accurate knowledge of future demand, which leads to the trade - off problem between low - return and high - return demands.
3. **Limitations of Existing Models**:
- **Adversarial Models**: Assuming no prediction about demand, the algorithm is too conservative.
- **Stochastic Models**: Assuming complete knowledge of the demand process, but in practice, demand prediction is often challenging, especially in new or non - stationary environments.
### Main Contributions of the Paper
1. **Introduction of Convex Uncertainty Set**: The paper proposes a new online resource allocation model. By introducing a convex uncertainty set representing machine - learning predictions, the adversarial model is enhanced. This set not only covers single - point estimates but also allows for considering errors and biases in predictions.
2. **Pareto - Optimal Algorithm**: The paper defines two performance metrics - the consistent ratio and the robust ratio - and proposes a parameterized class of Pareto - optimal algorithms that can balance between the consistent ratio and the robust ratio.
- **Consistent Ratio**: It measures the performance of the algorithm relative to the optimal hindsight solution when machine - learning suggestions are accurate.
- **Robust Ratio**: It measures the performance of the algorithm in the worst - case scenario when machine - learning suggestions are inaccurate.
3. **Adaptive Protection Level Algorithm**: The paper designs an adaptive protection level algorithm, which extends the classical fixed protection level algorithm. This algorithm can dynamically adjust the protection level according to the received low - return demands, thus maintaining good performance in different situations.
4. **Calculation of the Maximum Consistent Ratio**: The paper proposes a polynomial - time method to calculate the maximum consistent ratio \(C^*(R)\), which is the highest consistent ratio achievable given machine - learning suggestions.
### Numerical Studies
The paper verifies the performance of the proposed algorithms when machine - learning suggestions are inaccurate through numerical studies. The results show that these algorithms outperform the benchmark algorithms in both average performance and worst - case performance.
### Related Work
The paper also discusses other research related to using machine - learning suggestions in online decision - making, such as the online caching problem, the online secretary problem, and the online matching problem, and points out the uniqueness of this work, especially in dealing with convex uncertainty sets.
### Conclusion
By introducing the convex uncertainty set and the Pareto - optimal algorithm, the paper provides a method for effectively using machine - learning suggestions in online resource allocation while ensuring robustness when predictions are inaccurate. This method has important theoretical and practical values in multiple practical application scenarios.