Qingpeng Cai

Wenqiao Zhang, Lei Zhu, James Hallinan et al.

In this paper, we propose a novel semi-supervised learning (SSL) framework named BoostMIS that combines adaptive pseudo labeling and informative active annotation to unleash the potential of medical image SSL models: (1) BoostMIS can adaptively leverage the cluster assumption and consistency regularization of the unlabeled data according to the current learning status. This strategy can adaptively generate one-hot "hard" labels converted from task model predictions for better task model training. (2) For the unselected unlabeled images with low confidence, we introduce an Active learning (AL) algorithm to find the informative samples as the annotation candidates by exploiting virtual adversarial perturbation and model's density-aware entropy. These informative candidates are subsequently fed into the next training cycle for better SSL label propagation. Notably, the adaptive pseudo-labeling and informative active annotation form a learning closed-loop that are mutually collaborative to boost medical image SSL. To verify the effectiveness of the proposed method, we collected a metastatic epidural spinal cord compression (MESCC) dataset that aims to optimize MESCC diagnosis and classification for improved specialist referral and treatment. We conducted an extensive experimental study of BoostMIS on MESCC and another public dataset COVIDx. The experimental results verify our framework's effectiveness and generalisability for different medical image datasets with a significant improvement over various state-of-the-art methods.

Wanqi Xue, Qingpeng Cai, Zhenghai Xue et al.

Current advances in recommender systems have been remarkably successful in optimizing immediate engagement. However, long-term user engagement, a more desirable performance metric, remains difficult to improve. Meanwhile, recent reinforcement learning (RL) algorithms have shown their effectiveness in a variety of long-term goal optimization tasks. For this reason, RL is widely considered as a promising framework for optimizing long-term user engagement in recommendation. Though promising, the application of RL heavily relies on well-designed rewards, but designing rewards related to long-term user engagement is quite difficult. To mitigate the problem, we propose a novel paradigm, recommender systems with human preferences (or Preference-based Recommender systems), which allows RL recommender systems to learn from preferences about users historical behaviors rather than explicitly defined rewards. Such preferences are easily accessible through techniques such as crowdsourcing, as they do not require any expert knowledge. With PrefRec, we can fully exploit the advantages of RL in optimizing long-term goals, while avoiding complex reward engineering. PrefRec uses the preferences to automatically train a reward function in an end-to-end manner. The reward function is then used to generate learning signals to train the recommendation policy. Furthermore, we design an effective optimization method for PrefRec, which uses an additional value function, expectile regression and reward model pre-training to improve the performance. We conduct experiments on a variety of long-term user engagement optimization tasks. The results show that PrefRec significantly outperforms previous state-of-the-art methods in all the tasks.

Yue Feng, Shuchang Liu, Zhenghai Xue et al.

Conversational recommender systems (CRSs) aim to recommend high-quality items to users through a dialogue interface. It usually contains multiple sub-tasks, such as user preference elicitation, recommendation, explanation, and item information search. To develop effective CRSs, there are some challenges: 1) how to properly manage sub-tasks; 2) how to effectively solve different sub-tasks; and 3) how to correctly generate responses that interact with users. Recently, Large Language Models (LLMs) have exhibited an unprecedented ability to reason and generate, presenting a new opportunity to develop more powerful CRSs. In this work, we propose a new LLM-based CRS, referred to as LLMCRS, to address the above challenges. For sub-task management, we leverage the reasoning ability of LLM to effectively manage sub-task. For sub-task solving, we collaborate LLM with expert models of different sub-tasks to achieve the enhanced performance. For response generation, we utilize the generation ability of LLM as a language interface to better interact with users. Specifically, LLMCRS divides the workflow into four stages: sub-task detection, model matching, sub-task execution, and response generation. LLMCRS also designs schema-based instruction, demonstration-based instruction, dynamic sub-task and model matching, and summary-based generation to instruct LLM to generate desired results in the workflow. Finally, to adapt LLM to conversational recommendations, we also propose to fine-tune LLM with reinforcement learning from CRSs performance feedback, referred to as RLPF. Experimental results on benchmark datasets show that LLMCRS with RLPF outperforms the existing methods.

Zhenghai Xue, Qingpeng Cai, Shuchang Liu et al.

In many real-world scenarios, Reinforcement Learning (RL) algorithms are trained on data with dynamics shift, i.e., with different underlying environment dynamics. A majority of current methods address such issue by training context encoders to identify environment parameters. Data with dynamics shift are separated according to their environment parameters to train the corresponding policy. However, these methods can be sample inefficient as data are used \textit{ad hoc}, and policies trained for one dynamics cannot benefit from data collected in all other environments with different dynamics. In this paper, we find that in many environments with similar structures and different dynamics, optimal policies have similar stationary state distributions. We exploit such property and learn the stationary state distribution from data with dynamics shift for efficient data reuse. Such distribution is used to regularize the policy trained in a new environment, leading to the SRPO (\textbf{S}tate \textbf{R}egularized \textbf{P}olicy \textbf{O}ptimization) algorithm. To conduct theoretical analyses, the intuition of similar environment structures is characterized by the notion of homomorphous MDPs. We then demonstrate a lower-bound performance guarantee on policies regularized by the stationary state distribution. In practice, SRPO can be an add-on module to context-based algorithms in both online and offline RL settings. Experimental results show that SRPO can make several context-based algorithms far more data efficient and significantly improve their overall performance.

Qingpeng Cai, Shuchang Liu, Xueliang Wang et al.

Recently, short video platforms have achieved rapid user growth by recommending interesting content to users. The objective of the recommendation is to optimize user retention, thereby driving the growth of DAU (Daily Active Users). Retention is a long-term feedback after multiple interactions of users and the system, and it is hard to decompose retention reward to each item or a list of items. Thus traditional point-wise and list-wise models are not able to optimize retention. In this paper, we choose reinforcement learning methods to optimize the retention as they are designed to maximize the long-term performance. We formulate the problem as an infinite-horizon request-based Markov Decision Process, and our objective is to minimize the accumulated time interval of multiple sessions, which is equal to improving the app open frequency and user retention. However, current reinforcement learning algorithms can not be directly applied in this setting due to uncertainty, bias, and long delay time incurred by the properties of user retention. We propose a novel method, dubbed RLUR, to address the aforementioned challenges. Both offline and live experiments show that RLUR can significantly improve user retention. RLUR has been fully launched in Kuaishou app for a long time, and achieves consistent performance improvement on user retention and DAU.

Qingpeng Cai, Zhenghai Xue, Chi Zhang et al.

The wide popularity of short videos on social media poses new opportunities and challenges to optimize recommender systems on the video-sharing platforms. Users sequentially interact with the system and provide complex and multi-faceted responses, including watch time and various types of interactions with multiple videos. One the one hand, the platforms aims at optimizing the users' cumulative watch time (main goal) in long term, which can be effectively optimized by Reinforcement Learning. On the other hand, the platforms also needs to satisfy the constraint of accommodating the responses of multiple user interactions (auxiliary goals) such like, follow, share etc. In this paper, we formulate the problem of short video recommendation as a Constrained Markov Decision Process (CMDP). We find that traditional constrained reinforcement learning algorithms can not work well in this setting. We propose a novel two-stage constrained actor-critic method: At stage one, we learn individual policies to optimize each auxiliary signal. At stage two, we learn a policy to (i) optimize the main signal and (ii) stay close to policies learned at the first stage, which effectively guarantees the performance of this main policy on the auxiliaries. Through extensive offline evaluations, we demonstrate effectiveness of our method over alternatives in both optimizing the main goal as well as balancing the others. We further show the advantage of our method in live experiments of short video recommendations, where it significantly outperforms other baselines in terms of both watch time and interactions. Our approach has been fully launched in the production system to optimize user experiences on the platform.

Ziru Liu, Jiejie Tian, Qingpeng Cai et al.

In recent years, Multi-task Learning (MTL) has yielded immense success in Recommender System (RS) applications. However, current MTL-based recommendation models tend to disregard the session-wise patterns of user-item interactions because they are predominantly constructed based on item-wise datasets. Moreover, balancing multiple objectives has always been a challenge in this field, which is typically avoided via linear estimations in existing works. To address these issues, in this paper, we propose a Reinforcement Learning (RL) enhanced MTL framework, namely RMTL, to combine the losses of different recommendation tasks using dynamic weights. To be specific, the RMTL structure can address the two aforementioned issues by (i) constructing an MTL environment from session-wise interactions and (ii) training multi-task actor-critic network structure, which is compatible with most existing MTL-based recommendation models, and (iii) optimizing and fine-tuning the MTL loss function using the weights generated by critic networks. Experiments on two real-world public datasets demonstrate the effectiveness of RMTL with a higher AUC against state-of-the-art MTL-based recommendation models. Additionally, we evaluate and validate RMTL's compatibility and transferability across various MTL models.

Wanqi Xue, Qingpeng Cai, Ruohan Zhan et al.

Long-term engagement is preferred over immediate engagement in sequential recommendation as it directly affects product operational metrics such as daily active users (DAUs) and dwell time. Meanwhile, reinforcement learning (RL) is widely regarded as a promising framework for optimizing long-term engagement in sequential recommendation. However, due to expensive online interactions, it is very difficult for RL algorithms to perform state-action value estimation, exploration and feature extraction when optimizing long-term engagement. In this paper, we propose ResAct which seeks a policy that is close to, but better than, the online-serving policy. In this way, we can collect sufficient data near the learned policy so that state-action values can be properly estimated, and there is no need to perform online exploration. ResAct optimizes the policy by first reconstructing the online behaviors and then improving it via a Residual Actor. To extract long-term information, ResAct utilizes two information-theoretical regularizers to confirm the expressiveness and conciseness of features. We conduct experiments on a benchmark dataset and a large-scale industrial dataset which consists of tens of millions of recommendation requests. Experimental results show that our method significantly outperforms the state-of-the-art baselines in various long-term engagement optimization tasks.

Qingpeng Cai, Ruohan Zhan, Chi Zhang et al.

The wide popularity of short videos on social media poses new opportunities and challenges to optimize recommender systems on the video-sharing platforms. Users provide complex and multi-faceted responses towards recommendations, including watch time and various types of interactions with videos. As a result, established recommendation algorithms that concern a single objective are not adequate to meet this new demand of optimizing comprehensive user experiences. In this paper, we formulate the problem of short video recommendation as a constrained Markov Decision Process (MDP), where platforms want to optimize the main goal of user watch time in long term, with the constraint of accommodating the auxiliary responses of user interactions such as sharing/downloading videos. To solve the constrained MDP, we propose a two-stage reinforcement learning approach based on actor-critic framework. At stage one, we learn individual policies to optimize each auxiliary response. At stage two, we learn a policy to (i) optimize the main response and (ii) stay close to policies learned at the first stage, which effectively guarantees the performance of this main policy on the auxiliaries. Through extensive simulations, we demonstrate effectiveness of our approach over alternatives in both optimizing the main goal as well as balancing the others. We further show the advantage of our approach in live experiments of short video recommendations, where it significantly outperforms other baselines in terms of watch time and interactions from video views. Our approach has been fully launched in the production system to optimize user experiences on the platform.

Zijian Zhang, Shuchang Liu, Jiaao Yu et al.

Multi-domain recommendation and multi-task recommendation have demonstrated their effectiveness in leveraging common information from different domains and objectives for comprehensive user modeling. Nonetheless, the practical recommendation usually faces multiple domains and tasks simultaneously, which cannot be well-addressed by current methods. To this end, we introduce M3oE, an adaptive Multi-domain Multi-task Mixture-of-Experts recommendation framework. M3oE integrates multi-domain information, maps knowledge across domains and tasks, and optimizes multiple objectives. We leverage three mixture-of-experts modules to learn common, domain-aspect, and task-aspect user preferences respectively to address the complex dependencies among multiple domains and tasks in a disentangled manner. Additionally, we design a two-level fusion mechanism for precise control over feature extraction and fusion across diverse domains and tasks. The framework's adaptability is further enhanced by applying AutoML technique, which allows dynamic structure optimization. To the best of the authors' knowledge, our M3oE is the first effort to solve multi-domain multi-task recommendation self-adaptively. Extensive experiments on two benchmark datasets against diverse baselines demonstrate M3oE's superior performance. The implementation code is available to ensure reproducibility.

Zhenghai Xue, Qingpeng Cai, Bin Yang et al.

The field of Reinforcement Learning (RL) has garnered increasing attention for its ability of optimizing user retention in recommender systems. A primary obstacle in this optimization process is the environment non-stationarity stemming from the continual and complex evolution of user behavior patterns over time, such as variations in interaction rates and retention propensities. These changes pose significant challenges to existing RL algorithms for recommendations, leading to issues with dynamics and reward distribution shifts. This paper introduces a novel approach called \textbf{A}daptive \textbf{U}ser \textbf{R}etention \textbf{O}ptimization (AURO) to address this challenge. To navigate the recommendation policy in non-stationary environments, AURO introduces an state abstraction module in the policy network. The module is trained with a new value-based loss function, aligning its output with the estimated performance of the current policy. As the policy performance of RL is sensitive to environment drifts, the loss function enables the state abstraction to be reflective of environment changes and notify the recommendation policy to adapt accordingly. Additionally, the non-stationarity of the environment introduces the problem of implicit cold start, where the recommendation policy continuously interacts with users displaying novel behavior patterns. AURO encourages exploration guarded by performance-based rejection sampling to maintain a stable recommendation quality in the cost-sensitive online environment. Extensive empirical analysis are conducted in a user retention simulator, the MovieLens dataset, and a live short-video recommendation platform, demonstrating AURO's superior performance against all evaluated baseline algorithms.

Jingtong Gao, Zeyu Song, Chi Lu et al.

Recommender systems are critical for delivering personalized content across digital platforms, and recent advances in Large Language Models (LLMs) offer new opportunities to enhance them with richer world knowledge and explicit reasoning capabilities. With the help of reasoning knowledge, recommendations can better infer users' underlying intents, adapt to evolving preferences, and leverage semantic relationships for improved accuracy and interpretability. However, existing reasoning-based recommendation methods often fail to fully align the LLM's reasoning process with recommendation-specific objectives due to structural disruption during integration and difficulties in translating free-form generation into accurate item predictions. In this paper, we introduce RPORec, a reinforced preference optimization framework that unifies an LLM backbone's reasoning ability with a dedicated recommendation head (Rechead) for precise item retrieval. RPORec comprises two stages: (1) Reasoning-Augmented Recommendation Modeling, where high-quality Chain-of-Thought (CoT) reasoning is generated and used as auxiliary knowledge to guide the Rechead in learning recommendation-specific representations; and (2) Advanced Reasoning Refinement and Alignment, in which the trained Rechead produces verifiable rewards to fine-tune the LLM backbone via reinforcement learning, enhancing reasoning quality, structural consistency, and task relevance. Extensive experiments on public benchmarks and large-scale online deployments show that RPORec consistently outperforms state-of-the-art LLM-based recommendation methods, demonstrating the effectiveness of reasoning-augmented recommendation modeling in real-world systems.

Shengtian Yang, Yu Li, Shuo He et al.

Reinforcement learning (RL) has equipped LLM agents with a strong ability to solve complex tasks. However, existing RL methods normally use a \emph{single} policy network, causing \emph{simplicity bias} where simple tasks occupy most parameters and dominate gradient updates, leaving insufficient capacity for complex tasks. A plausible remedy could be employing the Mixture-of-Experts (MoE) architecture in the policy network, as MoE allows different parameters (experts) to specialize in different tasks, preventing simple tasks from dominating all parameters. However, a key limitation of traditional MoE is its token-level routing, where the router assigns each token to specialized experts, which fragments phase-consistent patterns into scattered expert assignments and thus undermines expert specialization. In this paper, we propose \textbf{Phase-Aware Mixture of Experts (PA-MoE)}. It first features a lightweight \emph{phase router} that learns latent phase boundaries directly from the RL objective without pre-defining phase categories. Then, the phase router allocates temporally consistent assignments to the same expert, allowing experts to preserve phase-specific expertise. Experimental results demonstrate the effectiveness of our proposed PA-MoE.

Zijian Zhang, Shuchang Liu, Ziru Liu et al.

User simulators can rapidly generate a large volume of timely user behavior data, providing a testing platform for reinforcement learning-based recommender systems, thus accelerating their iteration and optimization. However, prevalent user simulators generally suffer from significant limitations, including the opacity of user preference modeling and the incapability of evaluating simulation accuracy. In this paper, we introduce an LLM-powered user simulator to simulate user engagement with items in an explicit manner, thereby enhancing the efficiency and effectiveness of reinforcement learning-based recommender systems training. Specifically, we identify the explicit logic of user preferences, leverage LLMs to analyze item characteristics and distill user sentiments, and design a logical model to imitate real human engagement. By integrating a statistical model, we further enhance the reliability of the simulation, proposing an ensemble model that synergizes logical and statistical insights for user interaction simulations. Capitalizing on the extensive knowledge and semantic generation capabilities of LLMs, our user simulator faithfully emulates user behaviors and preferences, yielding high-fidelity training data that enrich the training of recommendation algorithms. We establish quantifying and qualifying experiments on five datasets to validate the simulator's effectiveness and stability across various recommendation scenarios.

Jiaju Chen, Chongming Gao, Chenxiao Fan et al.

Large language model (LLM)-based generative list-wise recommendation has advanced rapidly, but decoding remains sequential and thus latency-prone. To accelerate inference without changing the target distribution, speculative decoding (SD) uses a small draft model to propose several next tokens at once and a target LLM to verify and accept the longest prefix, skipping multiple steps per round. In generative recommendation, however, each item is represented by multiple semantic-ID tokens, often with separators, and current drafts typically treat these tokens uniformly. This overlooks two practical facts: (i) a token's semantics depend on its within-item slot, and (ii) uncertainty tends to increase with speculation depth. Without modeling these effects, SD's speedups can be limited. We introduce PAD-Rec, Position-Aware Drafting for generative Recommendation, a lightweight module that augments the draft model with two complementary signals. Item position embeddings explicitly encode the within-item slot of each token, strengthening structural awareness. Step position embeddings encode the draft step, allowing the model to adapt to depth-dependent uncertainty and improve proposal quality. To harmonize these signals with base features, we add simple gates: a learnable coefficient for item slots and a context-driven gate for draft steps. The module is trainable, easy to integrate with standard draft models, and adds negligible inference overhead. Extensive experiments on four real-world datasets show up to 3.1x wall-clock speedup and about 5% average wall-clock speedup gain over strong SD baselines, while largely preserving recommendation quality.

Jingtong Gao, Ling Pan, Yejing Wang et al.

Reinforcement Learning (RL) has emerged as a pivotal method for improving the reasoning capabilities of Large Language Models (LLMs). However, prevalent RL approaches such as Proximal Policy Optimization (PPO) and Group-Regularized Policy Optimization (GRPO) face critical limitations due to their reliance on sparse outcome-based rewards and inadequate mechanisms for incentivizing exploration. These limitations result in inefficient guidance for reasoning. Specifically, sparse rewards fail to deliver sufficient feedback, particularly for challenging problems. Furthermore, such rewards induce systematic biases that prioritize exploitation of familiar trajectories over novel solution discovery. These shortcomings critically hinder performance in complex reasoning tasks, which inherently demand iterative refinement across intermediate steps. To address these challenges, we propose an Intrinsic Motivation guidEd exploratioN meThOd foR LLM Reasoning (i-MENTOR), a method designed to deliver dense rewards and amplify exploration in the RL-based paradigm. i-MENTOR introduces three innovations: trajectory-aware exploration rewards that mitigate bias in token-level strategies while maintaining computational efficiency; error-conditioned reward allocation to ensure efficient exploration on challenging samples while intrinsically stabilizing training; and advantage-preserving integration that maintains advantage distribution integrity while incorporating exploratory guidance. Experiments across 4 public datasets demonstrate i-MENTOR's effectiveness, achieving a 22.23\% improvement on AIME 2024.

Jingtong Gao, Yewen Li, Shuai Mao et al.

Auto-bidding, with its strong capability to optimize bidding decisions within dynamic and competitive online environments, has become a pivotal strategy for advertising platforms. Existing approaches typically employ rule-based strategies or Reinforcement Learning (RL) techniques. However, rule-based strategies lack the flexibility to adapt to time-varying market conditions, and RL-based methods struggle to capture essential historical dependencies and observations within Markov Decision Process (MDP) frameworks. Furthermore, these approaches often face challenges in ensuring strategy adaptability across diverse advertising objectives. Additionally, as offline training methods are increasingly adopted to facilitate the deployment and maintenance of stable online strategies, the issues of documented behavioral patterns and behavioral collapse resulting from training on fixed offline datasets become increasingly significant. To address these limitations, this paper introduces a novel offline Generative Auto-bidding framework with Value-Guided Explorations (GAVE). GAVE accommodates various advertising objectives through a score-based Return-To-Go (RTG) module. Moreover, GAVE integrates an action exploration mechanism with an RTG-based evaluation method to explore novel actions while ensuring stability-preserving updates. A learnable value function is also designed to guide the direction of action exploration and mitigate Out-of-Distribution (OOD) problems. Experimental results on two offline datasets and real-world deployments demonstrate that GAVE outperforms state-of-the-art baselines in both offline evaluations and online A/B tests. By applying the core methods of this framework, we proudly secured first place in the NeurIPS 2024 competition, 'AIGB Track: Learning Auto-Bidding Agents with Generative Models'.

Yewen Li, Shuai Mao, Jingtong Gao et al.

Auto-bidding is essential in facilitating online advertising by automatically placing bids on behalf of advertisers. Generative auto-bidding, which generates bids based on an adjustable condition using models like transformers and diffusers, has recently emerged as a new trend due to its potential to learn optimal strategies directly from data and adjust flexibly to preferences. However, generative models suffer from low-quality data leading to a mismatch between the condition, like return to go, and true action value, especially in long sequential decision-making. Besides, the majority preference in the dataset may hinder models' generalization ability on minority advertisers' preferences. While it is possible to collect high-quality data and retrain multiple models for different preferences, the high cost makes it unaffordable, hindering the advancement of auto-bidding into the era of large foundation models. To address this, we propose a flexible and practical Generative Auto-bidding scheme using post-training Search, termed GAS, to refine a base policy model's output and adapt to various preferences. We use weak-to-strong search alignment by training small critics for different preferences and an MCTS-inspired search to refine the model's output. Specifically, a novel voting mechanism with transformer-based critics trained with policy indications could enhance search alignment performance. Additionally, utilizing the search, we provide a fine-tuning method for high-frequency preference scenarios considering computational efficiency. Extensive experiments conducted on the real-world dataset and online A/B test on the Kuaishou advertising platform demonstrate the effectiveness of GAS, achieving significant improvements, e.g., 4.60% increment of target cost.

Yewen Li, Zhiyi Lyu, Peng Jiang et al.

The growing scale of ad auctions on online advertising platforms has intensified competition, making manual bidding impractical and necessitating auto-bidding to help advertisers achieve their economic goals. Current auto-bidding methods have evolved to use offline reinforcement learning or generative methods to optimize bidding strategies, but they can sometimes behave counterintuitively due to the black-box training manner and limited mode coverage of datasets, leading to challenges in understanding task status and generalization in dynamic ad environments. Large language models (LLMs) offer a promising solution by leveraging prior human knowledge and reasoning abilities to improve auto-bidding performance. However, directly applying LLMs to auto-bidding faces difficulties due to the need for precise actions in competitive auctions and the lack of specialized auto-bidding knowledge, which can lead to hallucinations and suboptimal decisions. To address these challenges, we propose a hierarchical Large autoBidding Model (LBM) to leverage the reasoning capabilities of LLMs for developing a superior auto-bidding strategy. This includes a high-level LBM-Think model for reasoning and a low-level LBM-Act model for action generation. Specifically, we propose a dual embedding mechanism to efficiently fuse two modalities, including language and numerical inputs, for language-guided training of the LBM-Act; then, we propose an offline reinforcement fine-tuning technique termed GQPO for mitigating the LLM-Think's hallucinations and enhancing decision-making performance without simulation or real-world rollout like previous multi-turn LLM-based methods. Experiments demonstrate the superiority of a generative backbone based on our LBM, especially in an efficient training manner and generalization ability.

Haoran He, Yuxiao Ye, Qingpeng Cai et al.

RL with Verifiable Rewards (RLVR) has emerged as a promising paradigm for improving the reasoning abilities of large language models (LLMs). Current methods rely primarily on policy optimization frameworks like PPO and GRPO, which follow generalized policy iteration that alternates between evaluating the current policy's value and improving the policy based on evaluation. While effective, they often suffer from training instability and diversity collapse, requiring complex heuristic tricks and careful tuning. We observe that standard RLVR in math reasoning can be formalized as a specialized finite-horizon Markov Decision Process with deterministic state transitions, tree-structured dynamics, and binary terminal rewards. Though large in scale, the underlying structure is simpler than general-purpose control settings for which popular RL algorithms (e.g., PPO) were developed, suggesting that several sophisticated techniques in existing methods may be reduced or even omitted. Based on this insight, we prove a surprising result: the optimal action can be recovered from the Q-function of a fixed uniformly random policy, thereby bypassing the generalized policy iteration loop and its associated heuristics. We introduce Random Policy Valuation for Diverse Reasoning (ROVER) to translate this principle into a practical and scalable algorithm for LLM math reasoning, a minimalist yet highly effective RL method that samples actions from a softmax over these uniform-policy Q-values. ROVER preserves diversity throughout training, allowing sustained exploration of multiple valid pathways. Across multiple base models and standard math reasoning benchmarks, ROVER demonstrates superior performance in both \textbf{quality} (\textbf{+8.2} on pass@1, \textbf{+16.8} on pass@256) and \textbf{diversity} (\textbf{+17.6\%}), despite its radical simplification compared to strong, complicated existing methods.

Yewen Li, Jingtong Gao, Nan Jiang et al.

Auto-bidding is central to computational advertising, achieving notable commercial success by optimizing advertisers' bids within economic constraints. Recently, large generative models show potential to revolutionize auto-bidding by generating bids that could flexibly adapt to complex, competitive environments. Among them, diffusers stand out for their ability to address sparse-reward challenges by focusing on trajectory-level accumulated rewards, as well as their explainable capability, i.e., planning a future trajectory of states and executing bids accordingly. However, diffusers struggle with generation uncertainty, particularly regarding dynamic legitimacy between adjacent states, which can lead to poor bids and further cause significant loss of ad impression opportunities when competing with other advertisers in a highly competitive auction environment. To address it, we propose a Causal auto-Bidding method based on a Diffusion completer-aligner framework, termed CBD. Firstly, we augment the diffusion training process with an extra random variable t, where the model observes t-length historical sequences with the goal of completing the remaining sequence, thereby enhancing the generated sequences' dynamic legitimacy. Then, we employ a trajectory-level return model to refine the generated trajectories, aligning more closely with advertisers' objectives. Experimental results across diverse settings demonstrate that our approach not only achieves superior performance on large-scale auto-bidding benchmarks, such as a 29.9% improvement in conversion value in the challenging sparse-reward auction setting, but also delivers significant improvements on the Kuaishou online advertising platform, including a 2.0% increase in target cost.

Ruizhi Wang, Kai Liu, Bingjie Li et al.

In online advertising, the demand-side platform (a.k.a. DSP) enables advertisers to create different ad creatives for real-time bidding. Intuitively, advertisers tend to create more ad creatives for a single photo to increase the probability of participating in bidding, further enhancing their ad cost. From the perspective of DSP, the following are two overlooked issues. On the one hand, the number of ad creatives cannot grow indefinitely. On the other hand, the marginal effects of ad cost diminish as the number of ad creatives increases. To this end, this paper proposes a two-stage framework named Automated Creatives Quota (ACQ) to achieve the automatic creation and deactivation of ad creatives. ACQ dynamically allocates the creative quota across multiple advertisers to maximize the revenue of the ad platform. ACQ comprises two components: a prediction module to estimate the cost of a photo under different numbers of ad creatives, and an allocation module to decide the quota for photos considering their estimated costs in the prediction module. Specifically, in the prediction module, we develop a multi-task learning model based on an unbalanced binary tree to effectively mitigate the target variable imbalance problem. In the allocation module, we formulate the quota allocation problem as a multiple-choice knapsack problem (MCKP) and develop an efficient solver to solve such large-scale problems involving tens of millions of ads. We performed extensive offline and online experiments to validate the superiority of our proposed framework, which increased cost by 9.34%.

Peng Cui, Yiming Yang, Fusheng Jin et al.

In online advertising, once an ad campaign is deployed, the automated bidding system dynamically adjusts the bidding strategy to optimize Cost Per Action (CPA) based on the number of ad conversions. For ads with a long conversion delay, relying solely on the real-time tracked conversion number as a signal for bidding strategy can significantly overestimate the current CPA, leading to conservative bidding strategies. Therefore, it is crucial to predict the number of long-delayed conversions. Nonetheless, it is challenging to predict ad conversion numbers through traditional regression methods due to the wide range of ad conversion numbers. Previous regression works have addressed this challenge by transforming regression problems into bucket classification problems, achieving success in various scenarios. However, specific challenges arise when predicting the number of ad conversions: 1) The integer nature of ad conversion numbers exacerbates the discontinuity issue in one-hot hard labels; 2) The long-tail distribution of ad conversion numbers complicates tail data prediction. In this paper, we propose the Long-Delayed Ad Conversions Prediction model for bidding strategy (LDACP), which consists of two sub-modules. To alleviate the issue of discontinuity in one-hot hard labels, the Bucket Classification Module with label Smoothing method (BCMS) converts one-hot hard labels into non-normalized soft labels, then fits these soft labels by minimizing classification loss and regression loss. To address the challenge of predicting tail data, the Value Regression Module with Proxy labels (VRMP) uses the prediction bias of aggregated pCTCVR as proxy labels. Finally, a Mixture of Experts (MoE) structure integrates the predictions from BCMS and VRMP to obtain the final predicted ad conversion number.

Qingpeng Cai, Kaiping Zheng, H. V. Jagadish et al.

Cohort studies are of significant importance in the field of healthcare analysis. However, existing methods typically involve manual, labor-intensive, and expert-driven pattern definitions or rely on simplistic clustering techniques that lack medical relevance. Automating cohort studies with interpretable patterns has great potential to facilitate healthcare analysis but remains an unmet need in prior research efforts. In this paper, we propose a cohort auto-discovery model, CohortNet, for interpretable healthcare analysis, focusing on the effective identification, representation, and exploitation of cohorts characterized by medically meaningful patterns. CohortNet initially learns fine-grained patient representations by separately processing each feature, considering both individual feature trends and feature interactions at each time step. Subsequently, it classifies each feature into distinct states and employs a heuristic cohort exploration strategy to effectively discover substantial cohorts with concrete patterns. For each identified cohort, it learns comprehensive cohort representations with credible evidence through associated patient retrieval. Ultimately, given a new patient, CohortNet can leverage relevant cohorts with distinguished importance, which can provide a more holistic understanding of the patient's conditions. Extensive experiments on three real-world datasets demonstrate that it consistently outperforms state-of-the-art approaches and offers interpretable insights from diverse perspectives in a top-down fashion.

Haoran He, Emmanuel Bengio, Qingpeng Cai et al.

The Generative Flow Network (GFlowNet) is a probabilistic framework in which an agent learns a stochastic policy and flow functions to sample objects proportionally to an unnormalized reward function. A number of recent works explored connections between GFlowNets and maximum entropy (MaxEnt) RL, which modifies the standard objective of RL agents by learning an entropy-regularized objective. However, the relationship between GFlowNets and standard RL remains largely unexplored, despite the inherent similarities in their sequential decision-making nature. While GFlowNets can discover diverse solutions through specialized flow-matching objectives, connecting them can simplify their implementation through established RL principles and improve RL's diverse solution discovery capabilities. In this paper, we bridge this gap by revealing a fundamental connection between GFlowNets and one RL's most basic components -- policy evaluation. Surprisingly, we find that the value function obtained from evaluating a uniform policy is closely associated with the flow functions in GFlowNets through the lens of flow iteration under certain structural conditions. Building upon these insights, we introduce a rectified random policy evaluation (RPE) algorithm, which achieves the same reward-matching effect as GFlowNets based on simply evaluating a fixed random policy in these cases, offering a new perspective. Empirical results across extensive benchmarks demonstrate that RPE achieves competitive results compared to previous approaches, shedding light on the previously overlooked connection between (non-MaxEnt) RL and GFlowNets.

Chunhui Li, Cheng-Hao Liu, Dianbo Liu et al.

Generative Flow Networks (GFlowNets), a new family of probabilistic samplers, have recently emerged as a promising framework for learning stochastic policies that generate high-quality and diverse objects proportionally to their rewards. However, existing GFlowNets often suffer from low data efficiency due to the direct parameterization of edge flows or reliance on backward policies that may struggle to scale up to large action spaces. In this paper, we introduce Bifurcated GFlowNets (BN), a novel approach that employs a bifurcated architecture to factorize the flows into separate representations for state flows and edge-based flow allocation. This factorization enables BN to learn more efficiently from data and better handle large-scale problems while maintaining the convergence guarantee. Through extensive experiments on standard evaluation benchmarks, we demonstrate that BN significantly improves learning efficiency and effectiveness compared to strong baselines.

Wenqiao Zhang, Haochen Shi, Jiannan Guo et al.

Text-based image captioning (TextCap) requires simultaneous comprehension of visual content and reading the text of images to generate a natural language description. Although a task can teach machines to understand the complex human environment further given that text is omnipresent in our daily surroundings, it poses additional challenges in normal captioning. A text-based image intuitively contains abundant and complex multimodal relational content, that is, image details can be described diversely from multiview rather than a single caption. Certainly, we can introduce additional paired training data to show the diversity of images' descriptions, this process is labor-intensive and time-consuming for TextCap pair annotations with extra texts. Based on the insight mentioned above, we investigate how to generate diverse captions that focus on different image parts using an unpaired training paradigm. We propose the Multimodal relAtional Graph adversarIal inferenCe (MAGIC) framework for diverse and unpaired TextCap. This framework can adaptively construct multiple multimodal relational graphs of images and model complex relationships among graphs to represent descriptive diversity. Moreover, a cascaded generative adversarial network is developed from modeled graphs to infer the unpaired caption generation in image-sentence feature alignment and linguistic coherence levels. We validate the effectiveness of MAGIC in generating diverse captions from different relational information items of an image. Experimental results show that MAGIC can generate very promising outcomes without using any image-caption training pairs.

Qingpeng Cai, Can Cui, Yiyuan Xiong et al.

Data processing and analytics are fundamental and pervasive. Algorithms play a vital role in data processing and analytics where many algorithm designs have incorporated heuristics and general rules from human knowledge and experience to improve their effectiveness. Recently, reinforcement learning, deep reinforcement learning (DRL) in particular, is increasingly explored and exploited in many areas because it can learn better strategies in complicated environments it is interacting with than statically designed algorithms. Motivated by this trend, we provide a comprehensive review of recent works focusing on utilizing DRL to improve data processing and analytics. First, we present an introduction to key concepts, theories, and methods in DRL. Next, we discuss DRL deployment on database systems, facilitating data processing and analytics in various aspects, including data organization, scheduling, tuning, and indexing. Then, we survey the application of DRL in data processing and analytics, ranging from data preparation, natural language processing to healthcare, fintech, etc. Finally, we discuss important open challenges and future research directions of using DRL in data processing and analytics.

Ling Pan, Qingpeng Cai, Longbo Huang

A widely-used actor-critic reinforcement learning algorithm for continuous control, Deep Deterministic Policy Gradients (DDPG), suffers from the overestimation problem, which can negatively affect the performance. Although the state-of-the-art Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm mitigates the overestimation issue, it can lead to a large underestimation bias. In this paper, we propose to use the Boltzmann softmax operator for value function estimation in continuous control. We first theoretically analyze the softmax operator in continuous action space. Then, we uncover an important property of the softmax operator in actor-critic algorithms, i.e., it helps to smooth the optimization landscape, which sheds new light on the benefits of the operator. We also design two new algorithms, Softmax Deep Deterministic Policy Gradients (SD2) and Softmax Deep Double Deterministic Policy Gradients (SD3), by building the softmax operator upon single and double estimators, which can effectively improve the overestimation and underestimation bias. We conduct extensive experiments on challenging continuous control tasks, and results show that SD3 outperforms state-of-the-art methods.

Jianxiong Wei, Anxiang Zeng, Yueqiu Wu et al.

The slate re-ranking problem considers the mutual influences between items to improve user satisfaction in e-commerce, compared with the point-wise ranking. Previous works either directly rank items by an end to end model, or rank items by a score function that trades-off the point-wise score and the diversity between items. However, there are two main existing challenges that are not well studied: (1) the evaluation of the slate is hard due to the complex mutual influences between items of one slate; (2) even given the optimal evaluation, searching the optimal slate is challenging as the action space is exponentially large. In this paper, we present a novel Generator and Critic slate re-ranking approach, where the Critic evaluates the slate and the Generator ranks the items by the reinforcement learning approach. We propose a Full Slate Critic (FSC) model that considers the real impressed items and avoids the impressed bias of existing models. For the Generator, to tackle the problem of large action space, we propose a new exploration reinforcement learning algorithm, called PPO-Exploration. Experimental results show that the FSC model significantly outperforms the state of the art slate evaluation methods, and the PPO-Exploration algorithm outperforms the existing reinforcement learning methods substantially. The Generator and Critic approach improves both the slate efficiency(4% gmv and 5% number of orders) and diversity in live experiments on one of the largest e-commerce websites in the world.

Ling Pan, Qingpeng Cai, Longbo Huang

Recent years have witnessed a tremendous improvement of deep reinforcement learning. However, a challenging problem is that an agent may suffer from inefficient exploration, particularly for on-policy methods. Previous exploration methods either rely on complex structure to estimate the novelty of states, or incur sensitive hyper-parameters causing instability. We propose an efficient exploration method, Multi-Path Policy Optimization (MPPO), which does not incur high computation cost and ensures stability. MPPO maintains an efficient mechanism that effectively utilizes a population of diverse policies to enable better exploration, especially in sparse environments. We also give a theoretical guarantee of the stable performance. We build our scheme upon two widely-adopted on-policy methods, the Trust-Region Policy Optimization algorithm and Proximal Policy Optimization algorithm. We conduct extensive experiments on several MuJoCo tasks and their sparsified variants to fairly evaluate the proposed method. Results show that MPPO significantly outperforms state-of-the-art exploration methods in terms of both sample efficiency and final performance.

Qingpeng Cai, Ling Pan, Pingzhong Tang

Reinforcement learning algorithms such as the deep deterministic policy gradient algorithm (DDPG) has been widely used in continuous control tasks. However, the model-free DDPG algorithm suffers from high sample complexity. In this paper we consider the deterministic value gradients to improve the sample efficiency of deep reinforcement learning algorithms. Previous works consider deterministic value gradients with the finite horizon, but it is too myopic compared with infinite horizon. We firstly give a theoretical guarantee of the existence of the value gradients in this infinite setting. Based on this theoretical guarantee, we propose a class of the deterministic value gradient algorithm (DVG) with infinite horizon, and different rollout steps of the analytical gradients by the learned model trade off between the variance of the value gradients and the model bias. Furthermore, to better combine the model-based deterministic value gradient estimators with the model-free deterministic policy gradient estimator, we propose the deterministic value-policy gradient (DVPG) algorithm. We finally conduct extensive experiments comparing DVPG with state-of-the-art methods on several standard continuous control benchmarks. Results demonstrate that DVPG substantially outperforms other baselines.

Qingpeng Cai, Will Hang, Azalia Mirhoseini et al.

Heuristic algorithms such as simulated annealing, Concorde, and METIS are effective and widely used approaches to find solutions to combinatorial optimization problems. However, they are limited by the high sample complexity required to reach a reasonable solution from a cold-start. In this paper, we introduce a novel framework to generate better initial solutions for heuristic algorithms using reinforcement learning (RL), named RLHO. We augment the ability of heuristic algorithms to greedily improve upon an existing initial solution generated by RL, and demonstrate novel results where RL is able to leverage the performance of heuristics as a learning signal to generate better initialization. We apply this framework to Proximal Policy Optimization (PPO) and Simulated Annealing (SA). We conduct a series of experiments on the well-known NP-complete bin packing problem, and show that the RLHO method outperforms our baselines. We show that on the bin packing problem, RL can learn to help heuristics perform even better, allowing us to combine the best parts of both approaches.

Ling Pan, Qingpeng Cai, Qi Meng et al.

Value function estimation is an important task in reinforcement learning, i.e., prediction. The Boltzmann softmax operator is a natural value estimator and can provide several benefits. However, it does not satisfy the non-expansion property, and its direct use may fail to converge even in value iteration. In this paper, we propose to update the value function with dynamic Boltzmann softmax (DBS) operator, which has good convergence property in the setting of planning and learning. Experimental results on GridWorld show that the DBS operator enables better estimation of the value function, which rectifies the convergence issue of the softmax operator. Finally, we propose the DBS-DQN algorithm by applying dynamic Boltzmann softmax updates in deep Q-network, which outperforms DQN substantially in 40 out of 49 Atari games.

Feiyang Pan, Qingpeng Cai, An-Xiang Zeng et al.

Model-free reinforcement learning methods such as the Proximal Policy Optimization algorithm (PPO) have successfully applied in complex decision-making problems such as Atari games. However, these methods suffer from high variances and high sample complexity. On the other hand, model-based reinforcement learning methods that learn the transition dynamics are more sample efficient, but they often suffer from the bias of the transition estimation. How to make use of both model-based and model-free learning is a central problem in reinforcement learning. In this paper, we present a new technique to address the trade-off between exploration and exploitation, which regards the difference between model-free and model-based estimations as a measure of exploration value. We apply this new technique to the PPO algorithm and arrive at a new policy optimization method, named Policy Optimization with Model-based Explorations (POME). POME uses two components to predict the actions' target values: a model-free one estimated by Monte-Carlo sampling and a model-based one which learns a transition model and predicts the value of the next state. POME adds the error of these two target estimations as the additional exploration value for each state-action pair, i.e, encourages the algorithm to explore the states with larger target errors which are hard to estimate. We compare POME with PPO on Atari 2600 games, and it shows that POME outperforms PPO on 33 games out of 49 games.

Qingpeng Cai, Ling Pan, Pingzhong Tang

We study a reinforcement learning setting, where the state transition function is a convex combination of a stochastic continuous function and a deterministic function. Such a setting generalizes the widely-studied stochastic state transition setting, namely the setting of deterministic policy gradient (DPG). We firstly give a simple example to illustrate that the deterministic policy gradient may be infinite under deterministic state transitions, and introduce a theoretical technique to prove the existence of the policy gradient in this generalized setting. Using this technique, we prove that the deterministic policy gradient indeed exists for a certain set of discount factors, and further prove two conditions that guarantee the existence for all discount factors. We then derive a closed form of the policy gradient whenever exists. Furthermore, to overcome the challenge of high sample complexity of DPG in this setting, we propose the Generalized Deterministic Policy Gradient (GDPG) algorithm. The main innovation of the algorithm is a new method of applying model-based techniques to the model-free algorithm, the deep deterministic policy gradient algorithm (DDPG). GDPG optimize the long-term rewards of the model-based augmented MDP subject to a constraint that the long-rewards of the MDP is less than the original one. We finally conduct extensive experiments comparing GDPG with state-of-the-art methods and the direct model-based extension method of DDPG on several standard continuous control benchmarks. Results demonstrate that GDPG substantially outperforms DDPG, the model-based extension of DDPG and other baselines in terms of both convergence and long-term rewards in most environments.

Ling Pan, Qingpeng Cai, Zhixuan Fang et al.

Bike sharing provides an environment-friendly way for traveling and is booming all over the world. Yet, due to the high similarity of user travel patterns, the bike imbalance problem constantly occurs, especially for dockless bike sharing systems, causing significant impact on service quality and company revenue. Thus, it has become a critical task for bike sharing systems to resolve such imbalance efficiently. In this paper, we propose a novel deep reinforcement learning framework for incentivizing users to rebalance such systems. We model the problem as a Markov decision process and take both spatial and temporal features into consideration. We develop a novel deep reinforcement learning algorithm called Hierarchical Reinforcement Pricing (HRP), which builds upon the Deep Deterministic Policy Gradient algorithm. Different from existing methods that often ignore spatial information and rely heavily on accurate prediction, HRP captures both spatial and temporal dependencies using a divide-and-conquer structure with an embedded localized module. We conduct extensive experiments to evaluate HRP, based on a dataset from Mobike, a major Chinese dockless bike sharing company. Results show that HRP performs close to the 24-timeslot look-ahead optimization, and outperforms state-of-the-art methods in both service level and bike distribution. It also transfers well when applied to unseen areas.

Feiyang Pan, Qingpeng Cai, Pingzhong Tang et al.

Decision making is a challenging task in online recommender systems. The decision maker often needs to choose a contextual item at each step from a set of candidates. Contextual bandit algorithms have been successfully deployed to such applications, for the trade-off between exploration and exploitation and the state-of-art performance on minimizing online costs. However, the applicability of existing contextual bandit methods is limited by the over-simplified assumptions of the problem, such as assuming a simple form of the reward function or assuming a static environment where the states are not affected by previous actions. In this work, we put forward Policy Gradients for Contextual Recommendations (PGCR) to solve the problem without those unrealistic assumptions. It optimizes over a restricted class of policies where the marginal probability of choosing an item (in expectation of other items) has a simple closed form, and the gradient of the expected return over the policy in this class is in a succinct form. Moreover, PGCR leverages two useful heuristic techniques called Time-Dependent Greed and Actor-Dropout. The former ensures PGCR to be empirically greedy in the limit, and the latter addresses the trade-off between exploration and exploitation by using the policy network with Dropout as a Bayesian approximation. PGCR can solve the standard contextual bandits as well as its Markov Decision Process generalization. Therefore it can be applied to a wide range of realistic settings of recommendations, such as personalized advertising. We evaluate PGCR on toy datasets as well as a real-world dataset of personalized music recommendations. Experiments show that PGCR enables fast convergence and low regret, and outperforms both classic contextual-bandits and vanilla policy gradient methods.

Qingpeng Cai, Aris Filos-Ratsikas, Pingzhong Tang et al.

We study the problem of allocating impressions to sellers in e-commerce websites, such as Amazon, eBay or Taobao, aiming to maximize the total revenue generated by the platform. We employ a general framework of reinforcement mechanism design, which uses deep reinforcement learning to design efficient algorithms, taking the strategic behaviour of the sellers into account. Specifically, we model the impression allocation problem as a Markov decision process, where the states encode the history of impressions, prices, transactions and generated revenue and the actions are the possible impression allocations in each round. To tackle the problem of continuity and high-dimensionality of states and actions, we adopt the ideas of the DDPG algorithm to design an actor-critic policy gradient algorithm which takes advantage of the problem domain in order to achieve convergence and stability. We evaluate our proposed algorithm, coined IA(GRU), by comparing it against DDPG, as well as several natural heuristics, under different rationality models for the sellers - we assume that sellers follow well-known no-regret type strategies which may vary in their degree of sophistication. We find that IA(GRU) outperforms all algorithms in terms of the total revenue.