Yao Hu
Publications
SAGE: Sequence-level Adaptive Gradient Evolution for Generative Recommendation
Reinforcement learning-based preference optimization is increasingly used to align list-wise generative recommenders with complex, multi-objective user feedback, yet existing optimizers such as Gradient-Bounded Policy Optimization (GBPO) exhibit structural limitations in recommendation settings. We identify a Symmetric Conservatism failure mode in which symmetric update bounds suppress learning from rare positive signals (e.g., cold-start items), static negative-sample constraints fail to prevent diversity collapse under rejection-dominated feedback, and group-normalized multi-objective rewards lead to low-resolution training signals. To address these issues, we propose SAGE (Sequence-level Adaptive Gradient Evolution), a unified optimizer designed for list-wise generative recommendation. SAGE introduces sequence-level signal alignment via a geometric-mean importance ratio and a decoupled multi-objective advantage estimator to reduce token-level variance and mitigate reward collapse, together with asymmetric adaptive bounding that applies positive Boost updates to successful slates and an entropy-aware penalty to discourage low-diversity failures. Experiments on Amazon Product Reviews and the large-scale RecIF-Bench demonstrate consistent improvements in top-K accuracy, cold-start recall, and diversity across both Semantic-ID and native-text action spaces, while preserving numerical stability during training. These results suggest that asymmetric, sequence-aware policy optimization provides a principled and effective framework for addressing optimization failures in generative recommendation.
Guiding the Recommender: Information-Aware Auto-Bidding for Content Promotion
Modern content platforms offer paid promotion to mitigate cold start by allocating exposure via auctions. Our empirical analysis reveals a counterintuitive flaw in this paradigm: while promotion rescues low-to-medium quality content, it can harm high-quality content by forcing exposure to suboptimal audiences, polluting engagement signals and downgrading future recommendation. We recast content promotion as a dual-objective optimization that balances short-term value acquisition with long-term model improvement. To make this tractable at bid time in content promotion, we introduce a decomposable surrogate objective, gradient coverage, and establish its formal connection to Fisher Information and optimal experimental design. We design a two-stage auto-bidding algorithm based on Lagrange duality that dynamically paces budget through a shadow price and optimizes impression-level bids using per-impression marginal utilities. To address missing labels at bid time, we propose a confidence-gated gradient heuristic, paired with a zeroth-order variant for black-box models that reliably estimates learning signals in real time. We provide theoretical guarantees, proving monotone submodularity of the composite objective, sublinear regret in online auction, and budget feasibility. Extensive offline experiments on synthetic and real-world datasets validate the framework: it outperforms baselines, achieves superior final AUC/LogLoss, adheres closely to budget targets, and remains effective when gradients are approximated zeroth-order. These results show that strategic, information-aware promotion can improve long-term model performance and organic outcomes beyond naive impression-maximization strategies.
Robust Tool Use via Fission-GRPO: Learning to Recover from Execution Errors
Large language models (LLMs) can call tools effectively, yet they remain brittle in multi-turn execution: following a tool call error, smaller models often degenerate into repetitive invalid re-invocations, failing to interpret error feedback and self-correct. This brittleness hinders reliable real-world deployment, where the execution errors are inherently inevitable during tool interaction procedures. We identify a key limitation of current approaches: standard reinforcement learning (RL) treats errors as sparse negative rewards, providing no guidance on how to recover, while pre-collected synthetic error-correction datasets suffer from distribution mismatch with the model's on-policy error modes. To bridge this gap, we propose Fission-GRPO, a framework that converts execution errors into corrective supervision within the RL training loop. Our core mechanism fissions each failed trajectory into a new training instance by augmenting it with diagnostic feedback from a finetuned Error Simulator, then resampling recovery rollouts on-policy. This enables the model to learn from the precise errors it makes during exploration, rather than from static, pre-collected error cases. On the BFCL v4 Multi-Turn, Fission-GRPO improves the error recovery rate of Qwen3-8B by 5.7% absolute, crucially, yielding a 4% overall accuracy gain (42.75% to 46.75%) over GRPO and outperforming specialized tool-use agents.