Recommender systems need to optimize various types of user feedback, e.g., clicks, likes, and shares. A typical recommender system handling multiple types of feedback has two components: a multi-task learning (MTL) module, predicting feedback such as click-through rate and like rate; and a multi-task fusion (MTF) module, integrating these predictions into a single score for item ranking. MTF is essential for ensuring user satisfaction, as it directly influences recommendation outcomes. Recently, reinforcement learning (RL) has been applied to MTF tasks to improve long-term user satisfaction. However, existing RL-based MTF methods are formula-based methods, which only adjust limited coefficients within pre-defined formulas. The pre-defined formulas restrict the RL search space and become a bottleneck for MTF. To overcome this, we propose a formula-free MTF framework. We demonstrate that any suitable fusion function can be expressed as a composition of single-variable monotonic functions, as per the Sprecher Representation Theorem. Leveraging this, we introduce a novel learnable monotonic fusion cell (MFC) to replace pre-defined formulas. We call this new MFC-based model eXtreme MTF (xMTF). Furthermore, we employ a two-stage hybrid (TSH) learning strategy to train xMTF effectively. By expanding the MTF search space, xMTF outperforms existing methods in extensive offline and online experiments.

Integration of heat and electricity supply improves the overall energy efficiency and system operational flexibility. The renewable powered heat-electricity energy system is a promising way to set up residential energy supply facilities in remote areas beyond the reach of power system infrastructures. However, the volatility of wind and solar energy brings about the risk of supply inadequacy. This article proposes a data-driven robust method to quantify two measures of such a risk in the stand-alone renewable powered heat-electricity energy system. The uncertainty of renewable generation is modeled through a family of ambiguous probability distributions around an empirical one based on the Wasserstein metric; then, the probability of heat and electricity load shedding during a short period and related penalty cost are discussed. Through a polyhedral characterization of renewable power feasible region, the load shedding probability under the Wasserstein ambiguity set comes down to a linear program. With a piecewise linear optimal value function of the penalty cost, its expectation under the worst case distribution in the Wasserstein ambiguity set also gives rise to a linear program. The proposed method requires moderate information on renewable generation and makes full use of available data, whereas sustains computational tractability. The evaluation result is robust against the inaccuracy of renewable power distributions. Case studies demonstrate the effectiveness of the proposed approach.