Contextual Plackett-Luce: Efficient Neural Model for Ambiguous Sequence Selection

Researchers have introduced Contextual Plackett-Luce (CPL), a novel neural model designed to address the challenges of probabilistic sequence selection under ambiguity. This structured prediction problem is critical in tasks like trajectory forecasting and representative subset selection, where inputs often have multiple valid outputs but training data provides only single instances. CPL extends the classical Plackett-Luce model using an Ising-style parameterization with unary and pairwise interactions. It functions as a hybrid approach, combining the computational efficiency of parallel scoring with the expressive power of lightweight autoregressive selection. By decoupling these processes, CPL effectively captures multi-modal dependencies without the high computational costs typical of fully autoregressive models on modern hardware like GPUs. Evaluations on multi-modal path prediction and subset selection tasks demonstrate that CPL achieves superior structural consistency and robustness compared to strong parallel baselines. This advancement offers a significant improvement in handling ambiguous supervision in machine learning applications, balancing speed and accuracy in structured prediction environments.