In creator economy, the online platform posts contract to incentivize creators for high-quality content. The creators can use human effort or Generative Artificial Intelligence (GenAI) to assist them to create content for rewards. In this work, we identify that the existing contract suffers from GenAI-Induced Effort Degradation, where creators may not try their best to create high-quality content, but instead turn to GenAI to save effort and extract additional reward. Our results show that GenAI-induced effort degradation arises from failing to provide the same profit for the creators generating the same quality content. With this observation, we design the contract for creator economy in the era of GenAI as a classical contract for human content and a reward for the GenAI content which depends on the estimation of the human effort cost required to achieve GenAI's content quality. To overcome the curse of dimensionality in learning the optimal contract that maximizes platform profit with a high-quality content guarantee, we design an adaptive action pruning method that gradually filters out contracts resulting in low quality. Simulation experiments on synthetic data demonstrate the effectiveness of our approach.

High gradient variance presents a significant obstacle to efficient post-training of large language models (LLMs) on memory-constrained devices. Existing practical strategies-such as reducing batch sizes or adopting gradient accumulation (GA)-suffer from an inherent trade-off: smaller batches exacerbate convergence issues due to increased gradient noise, while GA substantially prolongs training time owing to its sequential processing. In this work, we reveal that the Exponential Moving Average (EMA) in momentum-based optimizers exponentially discounts historical gradients, thereby limiting their effectiveness in stabilizing parameter updates, especially during post-training when parameter drift is minimal. Motivated by this, we propose integrating the core idea of GA directly into momentum updates via a novel Periodical Moving Average (PMA) mechanism, which structures training into fixed periods and replaces EMA with a uniform moving average within each period. We instantiate PMA within AdamW and Lion, resulting in the AdamW-PMA and Lion-PMA optimizers. Theoretical analysis establishes that AdamW-PMA matches the convergence guarantees of standard Adam. Extensive empirical evaluation on supervised fine-tuning and direct preference optimization tasks demonstrates that PMA-based methods achieve approximately $2\times$ faster training compared to GA, while yielding consistently better performance on downstream evaluations.

Online advertising platforms leverage a two-stage auction architecture to deliver personalized ads to users with low latency. The first stage efficiently selects a small subset of promising candidates out of the complete pool of ads. In the second stage, an auction is conducted within the subset to determine the winning ad for display, using click-through-rate predictions from the second-stage machine learning model. In this work, we investigate the online learning process of the first-stage subset selection policy, while ensuring game-theoretic properties in repeated two-stage ad auctions. Specifically, we model the problem as designing a combinatorial bandit mechanism with a general reward function, as well as additional requirements of truthfulness and individual rationality (IR). We establish an O(T) regret lower bound for truthful bandit mechanisms, which demonstrates the challenge of simultaneously achieving allocation efficiency and truthfulness. To circumvent this impossibility result, we introduce truthful α-approximation oracles and evaluate the bandit mechanism through α-approximation regret. Two mechanisms are proposed, both of which are ex-post truthful and ex-post IR. The first mechanism is an explore-then-commit mechanism with regret O(T2/3 ), and the second mechanism achieves an improved O(log T /ΔΦ2) regret where ΔΦ is a distribution-dependent gap, but requires additional assumptions on the oracles and information about the strategic bidders.

Two-stage bandit-based algorithms have found widespread application in modern online platforms, offering a balance between cost and accuracy. The initial stage involves coarse filtering of a small candidate set of promising items from a large corpus, while the subsequent stage refines the selection and presents a single item to the user. In this work, to the best of our knowledge, we for the first time undertake a theoretical analysis of the two-stage stochastic multi-armed bandit problem. Specifically, we model the two-stage bandit problem as a two-stage online optimization, and conduct a theoretical analysis. We demonstrate that while the optimization objective of the first stage may seem intuitive, it is, in fact, non-trivial. We devise a proxy optimization objective, emphasize the importance of a carefully designed exploration strategy, and establish the theoretical analysis for the application of Upper Confidence Bound (UCB)-based algorithms in the first stage. Furthermore, we provide a regret analysis of the proposed two-stage bandit algorithm, demonstrating a gap-dependent upper bound of [EQUATION], where [EQUATION] is the largest reward gap, and a gap-independent lower bound of [EQUATION], where n represents the horizon.