Goal-conditioned reinforcement learning (GCRL), related to a set of complex RL problems, trains an agent achieve different goals under particular scenarios. Compared the standard solutions that learn policy solely depending on states or observations, GCRL additionally requires make decisions according goals. In this survey, we provide comprehensive overview challenges and algorithms for GCRL. Firstly, answer what basic problems are studied in field. Then, explain how represented present...

Tagging systems play an essential role in various information retrieval applications such as search engines and recommender systems. Recently, Large Language Models (LLMs) have been applied tagging due to their extensive world knowledge, semantic understanding, reasoning capabilities. Despite achieving remarkable performance, existing methods still limitations, including difficulties retrieving relevant candidate tags comprehensively, challenges adapting emerging domain-specific the lack of...

Hero drafting is essential in multiplayer online battle arena (MOBA) game playing as it builds the team of each side and directly affects match outcome. State-of-the-art methods fail to consider: 1) efficiency when hero pool expanded; 2) multiround nature a MOBA 5v5 series, i.e., two teams play best-of- <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$N$</tex-math></inline-formula> same only allowed be...

Goal-conditioned reinforcement learning (GCRL), related to a set of complex RL problems, trains an agent achieve different goals under particular scenarios. Compared the standard solutions that learn policy solely depending on states or observations, GCRL additionally requires make decisions according goals. In this survey, we provide comprehensive overview challenges and algorithms for GCRL. Firstly, answer what basic problems are studied in field. Then, explain how represented present...

Lightweight and high-efficiency electromagnetic wave (EMW) absorption sulfur doped graphene (S-GS) was fabricated by reducing doping oxide (GO) using chemical method. The obtained S-GS exhibits an extremely low reflection loss (RL), wide effective frequency bandwidth at a thin coating thickness (d), light weight cost. minimum RL reaches −52.3 dB 17.5 GHz with matching of 1.22 mm. More excitedly, the EM properties could be double-adjusted. By changing amount dopant S-GS, touches −52.2 9.4...

With the development of online education system, personalized recommendation has played an essential role. In this paper, we focus on developing path systems that aim to generating and recommending entire learning given user in each session. Noticing existing approaches fail consider correlations concepts path, propose a novel framework named Set-to-Sequence Ranking-based Concept-aware Learning Path Recommendation (SRC), which formulates task under set-to-sequence paradigm. Specifically,...

In Goal-oriented Reinforcement learning, relabeling the raw goals in past experience to provide agents with hindsight ability is a major solution reward sparsity problem. this paper, enhance diversity of relabeled goals, we develop FGI (Foresight Goal Inference), new strategy that relabels by looking into future learned dynamics model. Besides, improve sample efficiency, propose use model generate simulated trajectories for policy training. By integrating these two improvements, introduce...

Exploration is crucial for training the optimal reinforcement learning (RL) policy, where key to discriminate whether a state visiting novel. Most previous work focuses on designing heuristic rules or distance metrics check novel without considering such discrimination process that can be learned. In this paper, we propose method called generative adversarial exploration (GAEX) encourage in RL via introducing an intrinsic reward output from network, generator provides fake samples of states...

Click-through rate (CTR) prediction plays an important role in personalized recommendations. Recently, sample-level retrieval-based models (e.g., RIM) have achieved remarkable performance by retrieving and aggregating relevant samples. However, their inefficiency at the inference stage makes them impractical for industrial applications. To overcome this issue, paper proposes a universal plug-and-play Retrieval-Oriented Knowledge (ROK) framework. Specifically, knowledge base, consisting of...

Integrated ranking is a critical component in industrial recommendation platforms. It combines candidate lists from different upstream channels or sources and ranks them into an integrated list, which will be exposed to users. During this process, take responsibility for channel providers, the system needs consider exposure fairness among channels, directly affects opportunities of being displayed Besides, personalization also requires user's diverse preference on besides items. Existing...

Recommender models play a vital role in various industrial scenarios, while often faced with the catastrophic forgetting problem caused by fast shifting data distribution, e.g., evolving user interests, click signals fluctuation during sales promotions, etc. To alleviate this problem, common approach is to reuse knowledge from historical data. However, preserving vast and fast-accumulating hard, which causes dramatic storage overhead. Memorizing old through parametric base then proposed,...

Recently, increasing attention has been paid to LLM-based recommender systems, but their deployment is still under exploration in the industry. Most deployments utilize LLMs as feature enhancers, generating augmentation knowledge offline stage. However, recommendation scenarios, involving numerous users and items, even generation with consumes considerable time resources. This inefficiency stems from autoregressive nature of LLMs, a promising direction for acceleration speculative decoding,...

Click-Through Rate (CTR) prediction is a fundamental technique for online advertising recommendation and the complex competitive auction process also brings many difficulties to CTR optimization. Recent studies have shown that introducing posterior information contributes performance of prediction. However, existing work doesn't fully capitalize on benefits overlooks data bias brought by auction, leading biased suboptimal results. To address these limitations, we propose Auction Information...

Click-through rate (CTR) prediction is crucial for personalized online services. Sample-level retrieval-based models, such as RIM, have demonstrated remarkable performance. However, they face challenges including inference inefficiency and high resource consumption due to the retrieval process, which hinder their practical application in industrial settings. To address this, we propose a universal plug-and-play <u>r</u>etrieval-<u>o</u>riented <u>k</u>nowledge (ROK) framework that bypasses...

Graph neural networks (GNNs) have emerged as state-of-the-art methods to learn from graph-structured data for recommendation. However, most existing GNN-based recommendation focus on the optimization of model structures and learning strategies based pre-defined graphs, neglecting importance graph construction stage. Earlier works usually rely speciffic rules or crowdsourcing, which are either too simplistic labor-intensive. Recent start utilize large language models (LLMs) automate...

With the development of recommender systems, it becomes an increasingly common need to mix multiple item sequences from different sources. Therefore, integrated ranking stage is proposed be responsible for this task with re-ranking models. However, existing methods ignore relation between sequences, thus resulting in local optimum over interaction session. To resolve challenge, paper, we propose a new model named NFIRank (News Feed Integrated Ranking reinforcement learning) and formulate...

In Goal-oriented Reinforcement learning, relabeling the raw goals in past experience to provide agents with hindsight ability is a major solution reward sparsity problem. this paper, enhance diversity of relabeled goals, we develop FGI (Foresight Goal Inference), new strategy that relabels by looking into future learned dynamics model. Besides, improve sample efficiency, propose use model generate simulated trajectories for policy training. By integrating these two improvements, introduce...

Exploration is crucial for training the optimal reinforcement learning (RL) policy, where key to discriminate whether a state visiting novel. Most previous work focuses on designing heuristic rules or distance metrics check novel without considering such discrimination process that can be learned. In this paper, we propose method called generative adversarial exploration (GAEX) encourage in RL via introducing an intrinsic reward output from network, generator provides fake samples of states...

Hero drafting is essential in MOBA game playing as it builds the team of each side and directly affects match outcome. State-of-the-art methods fail to consider: 1) efficiency when hero pool expanded; 2) multi-round nature a 5v5 series, i.e., two teams play best-of-N same only allowed be drafted once throughout series. In this paper, we formulate process combinatorial propose novel algorithm based on neural networks Monte-Carlo tree search, named JueWuDraft. Specifically, design long-term...

With the development of online education system, personalized recommendation has played an essential role. In this paper, we focus on developing path systems that aim to generating and recommending entire learning given user in each session. Noticing existing approaches fail consider correlations concepts path, propose a novel framework named Set-to-Sequence Ranking-based Concept-aware Learning Path Recommendation (SRC), which formulates task under set-to-sequence paradigm. Specifically,...

Click-Through Rate (CTR) prediction is a fundamental technique in recommendation and advertising systems. Recent studies have shown that implementing multi-scenario recommendations contributes to strengthening information sharing improving overall performance. However, existing models only consider coarse-grained explicit scenario modeling depends on pre-defined identification from manual prior rules, which biased sub-optimal. To address these limitations, we propose Scenario-Aware...

In reinforcement learning applications like robotics, agents usually need to deal with various input/output features when specified different state/action spaces by their developers or physical restrictions. This indicates unnecessary re-training from scratch and considerable sample inefficiency, especially follow similar solution steps achieve tasks. this paper, we aim transfer high-level goal-transition knowledge alleviate the challenge. Specifically, propose PILoT, i.e., Planning...