Online Internet platforms require sophisticated marketing strategies to optimize user retention and platform revenue — a classical resource allocation problem. Traditional solutions adopt a two-stage pipeline: machine learning (ML) for predicting individual treatment effects to marketing actions, followed by operations research (OR) optimization for decision-making. This paradigm presents two fundamental technical challenges. First, the prediction-decision misalignment: Conventional ML methods focus solely on prediction accuracy without considering downstream optimization objectives, leading to improved predictive metrics that fail to translate to better decisions. Second, the bias-variance dilemma: Observational data suffers from multiple biases (e.g., selection bias, position bias), while experimental data (e.g., randomized controlled trials), though unbiased, is typically scarce and costly --- resulting in high-variance estimates. We propose **Bi**-level **D**ecision-**F**ocused **C**ausal **L**earning (**Bi-DFCL**) that systematically addresses these challenges. First, we develop an unbiased estimator of OR decision quality using experimental data, which guides ML model training through surrogate loss functions that bridge discrete optimization gradients. Second, we establish a bi-level optimization framework that jointly leverages observational and experimental data, solved via implicit differentiation. This novel formulation enables our unbiased OR estimator to correct learning directions from biased observational data, achieving optimal bias-variance tradeoff. Extensive evaluations on public benchmarks, industrial marketing datasets, and large-scale online A/B tests demonstrate the effectiveness of Bi-DFCL, showing statistically significant improvements over state-of-the-art. Currently, Bi-DFCL has been deployed across several marketing scenarios at Meituan, one of the largest online food delivery platforms in the world.

Marketing optimization plays an important role to enhance user engagement in online Internet platforms. Existing studies usually formulate this problem as a budget allocation problem and solve it by utilizing two fully decoupled stages, i.e., machine learning (ML) and operation research (OR). However, the learning objective in ML does not take account of the downstream optimization task in OR, which causes that the prediction accuracy in ML may be not positively related to the decision quality.

Online controlled experiments play a crucial role in enabling data-driven decisions across a wide range of companies. Variance reduction is an effective technique to improve the sensitivity of experiments, achieving higher statistical power while using fewer samples and shorter experimental periods. However, typical variance reduction methods (e.g., regression-adjusted estimators) are built upon the intuitional assumption of Gaussian distributions and cannot properly characterize the real business metrics with heavy-tailed distributions. Furthermore, outliers diminish the correlation between pre-experiment covariates and outcome metrics, greatly limiting the effectiveness of variance reduction.

Marketing is an important mechanism to increase user engagement and improve platform revenue, and heterogeneous causal learning can help develop more effective strategies. Most decision-making problems in marketing can be formulated as resource allocation problems and have been studied for decades. Existing works usually divide the solution procedure into two fully decoupled stages, i.e., machine learning (ML) and operation research (OR) --- the first stage predicts the model parameters and they are fed to the optimization in the second stage. However, the error of the predicted parameters in ML cannot be respected and a series of complex mathematical operations in OR lead to the increased accumulative errors. Essentially, the improved precision on the prediction parameters may not have a positive correlation on the final solution due to the side-effect from the decoupled design. In this paper, we propose a novel approach for solving resource allocation problems to mitigate the side-effects. Our key intuition is that we introduce the decision factor to establish a bridge between ML and OR such that the solution can be directly obtained in OR by only performing the sorting or comparison operations on the decision factor. Furthermore, we design a customized loss function that can conduct direct heterogeneous causal learning on the decision factor, an unbiased estimation of which can be guaranteed when the loss convergences. As a case study, we apply our approach to two crucial problems in marketing: the binary treatment assignment problem and the budget allocation problem with multiple treatments. Both large-scale simulations and online A/B Tests demonstrate that our approach achieves significant improvement compared with state-of-the-art.

We consider a fundamental update problem of avoiding forwarding loops based on the node-ordering protocol in Software Defined Networks (SDNs). Due to the distributed data plane, forwarding loops may occur during the updates and influence the network performance. The node-ordering protocol can avoid such forwarding loops by controlling the update orders of the switches and does not consume extra flow table space overhead. However, an Ω(n) lower bound on the number of rounds required by any algorithm using this protocol with loop-free constraint has been proved, where n is the number of switches in the network. To accelerate the updates, a weaker notion of loop-freedom - relaxed loop-freedom - has been introduced. Despite that, the theoretical bound of the node-ordering protocol with relaxed loop-free constraint remains unknown yet. In this article, we solve a long-standing open problem: how to derive ω(1) -round lower bound or to show that O(1)-round schedules always exist for the relaxed loop-free update problem. Specifically, we prove that any algorithm needs Ω(log n) rounds to guarantee relaxed loop freedom in the worst case. In addition, we develop a fast relaxed loop-free update algorithm named Savitar that touches the tight lower bound. For any update instance, Savitar can use at most 2⌊log2n⌋ - 1 rounds to schedule relaxed loop-free updates. Extensive experiments on Mininet using a Floodlight controller show that Savitar can significantly decrease the update time, achieve near optimal performance and save over 30% of the rounds compared with the state of the art.

Due to the unpredictable update orders in the data plane, a transient forwarding loop may be introduced in Software Defined Networks (SDNs) during the updates, which can result in packet drops and influence the application performance. The node-ordering protocol is a major update mechanism without incurring extra flow table space overhead. Existing algorithms using this protocol with loop-free constraint require Ω(n)-round lower bound, where n is the number of switches in the network. To accelerate the updates, a weaker notion of loop-freedom - relaxed loop-freedom - has been introduced. However, many problems about the theoretical bound of the node-ordering protocol with relaxed loop-free constraint remain unsolved yet. In this paper, we provide a rigorous proof to derive a Ω (log n)-round lower bound for relaxed loop-free update problem, which closes a long-term open problem whether O(1)-round schedules always exist or not. In addition, we develop a fast relaxed loop-free update algorithm named Savitar that touches the tight lower bound. Specifically, we prove that Savitar can use at most 2⌊log2n⌋ - 1 rounds to guarantee relaxed loop-freedom for any update instance. Extensive experiments on Mininet using a Floodlight controller show that Savitar can significantly decrease the update time, achieve the near optimal solution and save the number of rounds over 30% compared with the state of the art.

Consistent routing update based on Software-Defined Networks (SDN) is a complicated problem due to the asynchronous and distributed data plane. Existing ordered update approaches mostly focus on the consistent routing update problem for unicast other than multicast, which should guarantee two consistencies, drop-freeness and duplicate-freeness. In this paper, we propose Shifter, a novel dynamic ordered update scheme for consistent multicast routing update based on SDN to guarantee both consistencies. Shifter advocates configuring inport match field in the forwarding rules to avoid duplicate. In order to guarantee drop-freeness, Shifter employs a dependency graph to dynamically schedule update operations, and uses a greedy solution to solve a subproblem named Replace Operation Tree Migration Problem (ROTMP). We conduct simulations to evaluate Shifter and find that Shifter can give a near optimal solution of ROTMP with very few rounds and little runtime for multicast routing update scenarios. To the best of our knowledge, Shifter is the first ordered update scheme to guarantee the two consistencies simultaneously.