IJCAI 2026 论文列表

Post-training for Large Language Models (LLMs) can be mainly categorized into offline Supervised Fine-Tuning (SFT) for knowledge acquisition and online Reinforcement Fine-Tuning (RFT) for adaptive refinement. Current state-of-the-art approaches typically employ a sequential cold-start pipeline (SFT-Then-RFT). However, we argue that this disjoint transition imposes an “alignment tax", leading to catastrophic forgetting and reward hacking during the unregularized exploration phase. In this work, we advocate for Joint Online-Offline Fine-Tuning as a superior paradigm that breaks the convention of restricting offline data to SFT and online data to RFT. By integrating full offline response generation with online rollouts—particularly within the realm of Reinforcement Learning with Verifiable Rewards (RLVR)—this approach mitigates the limitations of isolated training phases. We provide the first comprehensive survey focusing specifically on the synchronization of data provenance. We introduce a novel taxonomy for related works, analyze their theoretical advantages in balancing stability with plasticity, and outline a roadmap for next-generation post-training frameworks.

Graphs are a fundamental data structure underlying many problems in the natural and social sciences. Over the past decade, Graph Neural Networks (GNNs) have dominated graph machine learning, supported by solid theoretical foundations. Yet scientists often understand graph structure through vision: chemists read molecular diagrams and social scientists inspect network visualizations. Despite decades of work on graph visualization, most graph learning pipelines still treat graphs purely as symbolic structures, rarely leveraging the visual form of graphs. We argue that this gap deserves renewed attention in the era of powerful vision and vision language models. This survey provides a first systematic overview of the emerging area we term vision meets graphs, which treats visual depictions of graphs as first-class inputs for reasoning and learning. We organize existing work into three threads. Vision for Graph Reasoning studies how models can use visual depictions of graphs to understand structure and carry out multi-step reasoning. Vision for Graph Learning explores how visual features can complement or augment graph encoders beyond known limitations of message passing. Scientific Graphs examines domains where standardized depiction conventions support both reasoning and learning. Our goal is to clarify what current methods can and cannot do, and to outline a path toward foundation models that perceive and reason about graphs as scientists do.

Multimodal Emotion Recognition (MER) focuses on identifying and interpreting emotions from modality-compound inputs. Closely mirroring human cognitive processes in real-world environments, MER has drawn substantial attention from both academia and industry. Recently, a paradigm shift has been unveiled in MER, from leveraging small-scale, task-specific models to Large Language Models (LLMs). We refer to the latter as the MER-with-LLMs paradigm, which offers unprecedented generality, spurring numerous empirical attempts, even alongside speculation about their potential to achieve general emotional intelligence. However, with these new opportunities come new challenges, including the scarcity of emotionally annotated data, the affective gap both within and across modalities, and the opacity of affective interpretation. To systematically review existing research and guide future exploration, this paper categorizes prior works according to their focus on addressing these challenges into three directions: Affective Data Augmentation, Multimodal Affective Representation, and Multimodal Affective Reasoning. By thoroughly tracing the development, emerging trends, and remaining issues within each direction, this paper aims to provide a clear academic map of the MER-with-LLMs paradigm and foster its structured advancement.

With the evolution of artificial intelligence (AI) paradigms towards agentic AI, the widespread integration of large language models (LLMs) enhances system capabilities while also introducing situational risks and challenges of value misalignment, making value alignment in agentic AI systems a critical issue. This paper constructs a multi-level value framework encompassing L0 (universal values), L1 (cultural and industry values), and L2 (context-specific values). Guided by this framework, we conduct an in-depth analysis along the technical stack: at the LLM level, we examine value injection mechanisms through pretraining and post-training; at the single-agent level, we focus on representation and injecting values to agents, Profiles and memory, and planning and action; at the multi-agent level, we summarize collaborative alignment methods such as communication strategy optimization and multi-objective reinforcement learning. Following a systematic review of existing datasets and methods for multi-level alignment evaluation, we outline future research directions, including inter-agent value coordination mechanisms, high-quality scenario data sharing, game-theoretic design for value alignment in agent interaction and communication protocol alignment—aiming to establish a more systematic and dynamic evaluation framework and to promote robust and trustworthy value consensus in agentic AI systems within social collaboration.

The performance of modern AI systems is fundamentally constrained by the quality of their underlying GPU kernels, which translate high-level algorithmic semantics into low-level hardware operations. Achieving near-optimal kernels requires expert-level understanding of hardware architectures and programming models, making kernel engineering a critical but notoriously time-consuming and non-scalable process. Recent advances in large language models and LLM-based agents have opened new possibilities for automating kernel generation and optimization. LLMs are well-suited to compress expert-level kernel knowledge that is difficult to formalize, while agentic systems further enable scalable optimization by casting kernel development as an iterative, feedback-driven loop. Rapid progress has been made in this area. However, the field remains fragmented and lacks a systematic perspective for LLM-driven kernel generation. This survey addresses this gap by providing a structured overview of existing approaches, spanning LLM-based approaches and agentic optimization workflows, and systematically organizing the datasets and benchmarks that underpin learning and evaluation in this domain. Moreover, key open challenges and future research directions are further outlined, aiming to establish a comprehensive reference for the next generation of automated kernel optimization. To keep track of this field, we maintain an open-source GitHub repository at https://github.com/flagos-ai/awesome-LLM-driven-kernel-generation.

LLM-based agents perceive, reason, act, and adapt through interaction. While scaling remains important, agentic progress also depends on extracting more learning signal from limited experience, including demonstrations, feedback, reasoning traces, tool-use records, and interaction trajectories. This survey develops an agent-centric view of data-efficient learning. We formulate an agentic learning loop, define data efficiency as capability improvement without proportional increases in supervision or real-environment trial-and-error, and synthesize methods into experience augmentation, agent structural design, and learning paradigms. We also summarize representative application domains and open challenges in experience reuse and cost-aware evaluation. Overall, the agentic era requires smarter ways to acquire, structure, reuse, and learn from limited experience.

Deep neural networks deliver strong performance but remain opaque, limiting their use in high-stakes domains that require transparency and human oversight. Concept Bottleneck Models (CBMs) address this gap by introducing a human-interpretable concept layer that mediates inputs and decisions, enabling semantic explanations and test-time intervention. This survey provides a unified review of CBMs organized along four dimensions: concept acquisition, concept-based decision making, concept intervention, and concept evaluation. We summarize the evolution of concept construction from manual annotation to lexicon-based mining, LLM/VLM-guided generation, and visually grounded discovery via prototypes and diffusion models; review emerging CBM architectures beyond strict bottlenecks; and consolidate evaluation and intervention protocols emphasizing faithfulness, sparsity, and intervenability, with particular relevance to high-stakes domains such as healthcare. We synthesize fragmented literature and outline key challenges and future directions for concept-based interpretable decision making.

Traffic forecasting is evolving, with world models emerging as a powerful framework applicable to tasks such as core state, trajectory, event, and demand forecasting. These tasks involve both visual and spatiotemporal data, yet most existing methods treat them separately, hindering a unified understanding of traffic scenes in both semantic meanings and spatiotemporal dynamics. The fusion of the two modalities is critical for building models that comprehend complex traffic scenarios. However, the fusion issue faces two fundamental misalignments: semantic, where pixels conflict with traffic concepts, and geometric, which requires spatial intelligence to map 2D inputs into 3D. This survey reframes vision-spatiotemporal fusion via the unique lens of cross-modal alignment, addressing semantic and geometric failures that limit forecasting reliability. First, we categorize existing methods into three paradigms: feature-level, semantic-level, and task-level. This reveals their progression from low-level feature manipulation to high-level architectural integration. Second, we synthesize representative techniques per paradigm, highlighting geometric challenges such as cross-view association and spatial mapping. Third, we examine current datasets and benchmarks, highlighting their deficiencies in evaluating alignment. Finally, we outline future directions, including spatiotemporal intelligence for robust perception and holistic traffic world models. The unified framework establishes a reference for robust and explainable forecasting systems.

Numerical weather prediction (NWP) models remain the cornerstone of atmospheric sciences. Yet, deep learning (DL) is challenging this paradigm by its ability to capture intricate spatio-temporal patterns and deliver ultra-fast predictions. Analogous to the foundation models (e.g., ChatGPT) in natural language processing, foundation models in the weather/climate domain have also been developed. This paper reviews DL and foundation models for weather prediction by highlighting their strengths and limitations. In particular, we carefully examine them from the perspective of their training paradigms: deterministic predictive learning, probabilistic generative learning, and pre-training & fine-tuning. For each paradigm, we summarize the underlying model architectures, training methods, and respective features. To facilitate further study, we provide a curated repository featuring categorized papers, open-source code, and benchmark datasets. Finally, we discuss and suggest potential research directions across new tasks and models in weather data storage and management, and operational deployment, further inspiring innovations in this rapidly evolving field. GitHub: https://github.com/JimengShi/DL-Foundation-Models-Weather.

Recent developments in Artificial Intelligence (AI) have led to new ways for users to search through vast information. However, users may have questions that are grounded in the real world that require spatial inference, for which language models are not well suited. Conversely, traditional spatial search methods, like spatial pattern matching, answer spatial reasoning questions correctly but are noise-intolerant, slow, and brittle. Presently, there are opportunities to integrate AI and spatial pattern matching to enable robust and flexible spatial search. This paper surveys existing spatial pattern matching methods, including the few that apply machine learning to the problem, discussing their efficiency and limitations, and describing opportunities to further enable spatial search through AI.

Person re-identification via 3D skeletons is an important emerging research area that attracts increasing attention within the pattern recognition community. With distinctive advantages across various application scenarios, numerous 3D skeleton based person re-identification (SRID) methods with diverse skeleton modeling and learning paradigms have been proposed in recent years. In this paper, we provide a comprehensive review and analysis of recent SRID advances. First of all, we define the SRID task and provide an overview of its origin and major advancements. Secondly, we formulate a systematic taxonomy that organizes existing methods into three categories centered on hand-crafted, sequence-based, and graph-based modeling. Then, we elaborate on the representative models along these three types with an illustration of foundational mechanisms. Meanwhile, we provide an overview of mainstream supervised, self-supervised, and unsupervised SRID learning paradigms and corresponding common methods. A thorough evaluation of state-of-the-art SRID methods is further conducted over various types of benchmarks and protocols to compare their effectiveness, efficiency, and key properties. Finally, we present the key challenges and prospects to advance future research, and highlight interdisciplinary applications of SRID with a case study. A curated collection of valuable resources is available at https://github.com/Kali-Hac/3D-SRID-Survey.

Multivariate Time Series Forecasting (MTSF) plays a crucial role across diverse fields, ranging from economics to energy to traffic. In recent years, deep learning has demonstrated outstanding performance in MTSF tasks. In MTSF, modeling the correlations among different channels is critical, as leveraging information from other related channels can significantly improve the prediction accuracy of a specific channel. This study systematically reviews the channel modeling strategies for time series and proposes a taxonomy organized into three hierarchical levels: the strategy perspective, the mechanism perspective, and the characteristic perspective. On this basis, we provide a structured analysis of these methods and conduct an in-depth examination of the advantages and limitations of different channel strategies. Finally, we summarize and discuss some future research directions to provide useful research guidance. Moreover, we maintain an up-to-date GitHub repository which includes all the papers discussed in the survey.

Controllable generation guided by external knowledge is a key requirement in modern generative deep learning applications, enabling the synthesis of samples with explicit constraints on semantic content, structural properties, and variability. In 3D Computed Tomography (CT), such control is essential for clinical applications, including data augmentation, privacy-preserving data sharing, and the simulation of specific anatomical or pathological scenarios. While research on conditional 3D CT generation has expanded rapidly, the diversity of existing approaches makes systematic comparison difficult and obscures fundamental design choices. In this survey, we propose a conditioning-centric taxonomy that organizes the literature along three orthogonal dimensions: the type of external knowledge (K), the knowledge integration paradigm (I), and the generative architecture (A). This factorization defines an explicit design space (K x I x A) that provides a unified perspective on prior work. Using this framework, we systematize existing methods, identify dominant trends and recurring design patterns, and highlight underexplored regions of the design space that point toward promising directions for future research.

Large Language Models (LLMs) are transforming blockchain security and analytics, yet a systematic evaluation of their capabilities remains limited. This survey provides a comprehensive, AI‑centric assessment of LLM‑based methods across over 70 recent studies spanning 11 application domains, such as security auditing, transaction fraud detection, and cryptocurrency portfolio management. Our unified taxonomy standardizes task formulations and evaluation practices to enable a comparison of six LLM roles across domains. For each domain, we review input representations tailored to blockchain data; LLM architectures, learning and inference paradigms, e.g., fine‑tuning, retrieval‑augmented generation, and agentic strategies. Our review analyzes the strengths, limitations, and emerging patterns of LLM roles observed in current systems. Finally, we provide practical guidance for selecting LLMs for specific roles and outline promising research directions.

Conventional Federated Learning (FL) pipelines focus on the collaborative training of a global dense model across client devices. Sparsity has been increasingly adopted in FL, during or after local optimization, for a range of objectives, including reducing communication and computation costs, supporting unlearning, enhancing privacy guarantees, and improving local personalization. In this survey, we introduce a novel taxonomy of sparse FL methods that systematically organizes the existing literature according to their core objectives and methodological choices. Using this taxonomy, we analyze and categorize prior work, highlighting the underlying intuitions, technical mechanisms, benefits, and limitations of each class of approaches. Finally, we identify open challenges, expose research gaps, and extract guidance to help practitioners understand and adopt sparsity mechanisms in FL.

Reinforcement learning (RL) is increasingly applied in complex, safety-critical domains, yet the lack of rigorous behavioral guarantees for neural network-based policies remains a major barrier to deployment. Recent advances in policy expressiveness and scale have intensified this challenge, leading to a rapidly growing but conceptually fragmented body of work on RL policy verification. This survey provides a unifying perspective on RL verification methods. We introduce a taxonomy that clarifies relationships among existing approaches along three axes: verification paradigm (formal versus probabilistic), temporal scope (step-wise versus multi-step), and guarantees strength. Beyond taxonomy, we unify underlying theoretical foundations, make implicit assumptions and limitations explicit, and identify emerging directions.

The literature on AI fairness has grown rapidly, proposing a large number of bias mitigation techniques that are commonly organized into pre-, in-, and post-processing methods. This pipeline-centric view offers an operational, lifecycle-based perspective on where mitigation can be applied. In deployment settings, however, practitioners also face an additional question: whether a mitigation family is applicable given the resources and access rights available in a concrete system. In this survey, we use resources broadly to denote data access/control, training capability, and deployment-time interface/decision control. Accordingly, we introduce a resource-aware taxonomy that complements existing taxonomies by classifying AI bias mitigation methods according to the conditions that make them practically implementable. We use this taxonomy to structure and reinterpret existing literature on the topic, highlighting which mitigation families remain feasible under resource constraints.

Graph representation learning (GRL) serves as a canonical paradigm for modeling complex networks. However, real-world AI systems inherently manifest as evolving heterogeneous entities with complex interactions, posing significant challenges to static or homogeneous modeling. To address these complexities, representation learning for Dynamic Heterogeneous Graphs (DHGs) has emerged as a vital approach for learning low-dimensional representations that simultaneously preserve structural semantics and temporal dynamics. This survey presents the first systematic review of DHG representation learning methods. We first introduce a unified formal definition that encompasses both discrete-time and continuous-time DHGs from the perspective of temporal granularity. Building upon this formulation, we propose a novel algorithm-centric taxonomy that categorizes existing literature, including early embedding-based approaches, graph neural network (GNN)-based models, and relatively recent Transformer-based DHG methods, while explicitly highlighting their intrinsic modeling biases with respect to dynamic granularity. Furthermore, we summarize representative applications of DHG representation learning, along with commonly used datasets and benchmarks. Finally, we discuss promising research directions that guide future advances in this rapidly evolving field.

Large language models (LLMs) require substantial investments and are increasingly deployed in high-stakes domains, making it critical to protect LLM-related assets and to trace their provenance. Identity technologies such as fingerprinting and watermarking address these needs by enabling ownership verification and attribution, and have rapidly emerged as an active research focus. However, existing techniques lack a systematic organisation, leading to two key issues, terminological confusion and isolated research lines, that have hindered the development of this research field. To this end, we present a comprehensive review of LLM identity techniques, focusing on fingerprinting and watermarking across the LLM lifecycle, including datasets, models, and generated content. We make three primary contributions. First, we introduce implicit identity (Implicit-ID for short) as a unifying abstraction and distinguish fingerprinting from watermarking. Second, we propose a lifecycle-based taxonomy that organises techniques by asset type and verification role, aligning each with asset protection or provenance. Third, we establish an evaluation framework around three objectives---identifiability, robustness, and deployability. Together, these contributions structure the landscape of LLM identity techniques, clarify terminology, and highlight directions toward secure deployment.