Searcharxiv⌕ Search

arXiv subjects

Zhanqiu Zhang

Publications and source records attributed to Zhanqiu Zhang.

18 recordsLinked to original sources

MMDM: Measuring Massive Multitask Decision Making

System-1 Decision Models directly make structured probabilistic decisions over query-specified options given a state in a single forward pass, without autoregressively generating reasoning trajectories or textual responses. Recent evaluations cover diverse tasks but often rely on existing benchmarks, where prior exposure to test material can obscure generalization. Moreover, real-world decision making often involves long contexts, large option spaces, and complex logical dependencies, challenges that existing benchmarks do not adequately capture together. Thus, we introduce MMDM (Massive Multitask Decision Making), a benchmark of 5,448 high-quality questions across 16 task categories, combining newly constructed tasks with source-grounded recompositions and rewrites for single-select, multi-select, and soft-label decision making in fresh contexts. It covers realistic scenarios such as agent state tracking, tool and skill selection, preference judgments, and rubric scoring, with option spaces ranging from 2 to 1,024 and inputs extending to 64,817 tokens. With MMDM, we conduct a comprehensive quantitative evaluation of 28 decision models, showing that many perform well on common knowledge yet struggle with interdependent options, constrained decisions, and multi-select tasks. For example, Perplexity Decider-V1.2 reaches 78.75% Hard Single-select Accuracy on world knowledge, but only 53.95% on option logic and 47.95% on constraint planning, while the best Hard Multi-select Accuracy across all models is 36.81%. These results highlight joint reasoning, constraint-aware selection, and complete multi-select recovery as key priorities for decision model development, with MMDM providing a testbed for measuring this progress.

cs.LG↗

HiSTEVE 1.0: Harnessing Human-Interface VLA Agents for Open-World Games

Building virtual embodied agents that operate in open-ended games like human players is a long-standing goal requiring both long-horizon autonomy and reliable native-interface control. As a procedurally generated open world, Minecraft provides a natural testbed for studying the integration of these capabilities. Existing agents typically provide either native visuomotor control without long-horizon autonomy or high-level planning executed through environment APIs or predefined skills. To bridge this gap, we propose HiSTEVE, a hierarchical VLA agent with a capability-aware agent harness. At the orchestration level, we equip the harness with structured memory for tracking progress and a VLM planner for selecting subtasks within the VLA's capabilities. To reliably determine subtask completion, we further introduce a policy-coupled task completion state detector that conditions on VLA policy context to signal candidate completion. At the control level, a single VLA executes all subtasks with vision-only perception and native keyboard-and-mouse actions. To mitigate interference between heterogeneous GUI and non-GUI interactions, we introduce regime-specific RMSNorm branches and separate mouse-movement token embeddings. We also train the VLA through world knowledge injection, imitation learning, and reinforcement learning with environment feedback to improve reliability and recovery. HiSTEVE achieves state-of-the-art atomic-task performance, scoring 85.0% on MineStudio Bench and a 37.4% average success rate on OpenHA, while also completing long-horizon tasks such as crafting an iron pickaxe from scratch like human players.

cs.CV↗

Progressive Agent Skill Generation via Reinforcement Learning

Recent large language model agents often use external skills as modular procedural units that condition inference and improve complex task solving. Thus, automatically generating high-quality skills from documents or experience has become an important problem. Existing skill generation methods largely rely on heuristics or pipeline-style consolidation, which must be specially designed for different evidence sources. In contrast, learning-based approaches offer a more unified way to model skill generation across heterogeneous sources. However, learning-based skill generation remains challenging because skills lack a natural supervision signal based on relevance or correctness; their value can largely be determined only by whether they improve the behavior of the agent on downstream tasks. To address this challenge, we propose Skill-$α$, a reinforcement learning method that learns a unified policy for progressive skill generation. Specifically, we construct each skill by repeatedly applying the learned policy to successive source evidence and introduce a novel rollback reward that evaluates each edit by comparing downstream execution under the original and edited skills on an anchored query. Extensive experiments show that Skill-$α$ generates more effective skills than methods based on heuristics or pipelines in both document-to-skill and experience-to-skill settings. Under the main GPT-4o worker, Skill-$α$ improves average downstream success rates over the strongest skill-generation baseline by 3.1 points on CL-Bench and 6.7 points on tau2-bench. Further ablations and analysis validate the importance of rollback reward and progressive generation.

cs.LG↗

Just MLPs: Efficient Visual State Reconstruction for Multimodal Language Models

Long visual token sequences often account for a substantial fraction of the computational overhead in multimodal large language models~(MLLMs). Existing approaches reduce this cost by pruning redundant visual tokens, but permanently discard visual evidence that may become useful in subsequent layers. We instead ask whether all visual tokens can be preserved while reducing the cost of repeatedly evolving the representations through the Transformer. To answer this question, we perform low-rank interventions on visual-to-text information flow. We find that, after visual-to-text attention is blocked, restoring only a few directions recovers most of the lost accuracy, suggesting the relevant visual influence is concentrated in a low-dimensional subspace. We further observe strong predictability in layer-specific visual states: lightweight MLPs approximate them with high cosine similarity and low reconstruction error. Motivated by these findings, we propose $δ$-Vision, which replaces repeated Transformer evolution of visual tokens with lightweight low-rank adapters that construct layer-wise visual memories while preserving all visual tokens for text retrieval. Across image and video benchmarks, $δ$-Vision achieves higher accuracy than visual token pruning baselines at comparable or lower computation, while delivering competitive inference efficiency without discarding visual tokens.

cs.CV↗

GameWAM: A World Action Model for Video Games

Modern video games combine first-person perception, rapid visual changes, persistent world state, and heterogeneous native controls. Existing game agents map visual and task context directly to actions but lack explicit world dynamics modeling, whereas interactive game world models predict visual futures from supplied actions but do not serve as task policies. World-Action Models (WAMs) unify these objectives, but remain largely unexplored under the dynamics and open-ended interaction of video games. We introduce GameWAM, to our knowledge the first WAM for native closed-loop gameplay and GUI control. GameWAM jointly generates future visual observations and executable keyboard-mouse trajectories through parallel visual and action generative processes with block-causal conditioning and flow matching. To support joint world-action learning, we construct synchronized gameplay and GUI trajectories. To handle heterogeneous native control, GameWAM predicts a gameplay/GUI mode per action step and generates actions with mode-specific prediction distributions and continuous-action normalization. For long-horizon interaction, block-cycle control coordinates prediction, execution, and temporal context: it predicts beyond the committed horizon, executes short action blocks, replans from new observations, and hierarchically structures context from fine-grained within-cycle history to persistent cross-cycle history. Experiments demonstrate competitive task success with fewer executed native actions than the compared agents. We further uncover Low-Frequency Action Source Imprinting (LASI), in which low-frequency components of the sampled action source systematically steer coarse generated camera motion under fixed conditioning, revealing a source-sensitivity failure mode in generative control. Project page is available at https://yunncheng.github.io/GameWAM/.

cs.AI↗

CONDUIT: A Unified Residual-Stream Restoration Framework for KV Cache Reuse in Vision-Language Models

Vision-language models (VLMs) often answer new questions about recurring visual content, where reusing the key-value (KV) cache can avoid re-encoding expensive visual prefixes. Exact-prefix reuse, however, fails when the same visual content appears under a changed prefix. Selective recomputation can recover quality under a small visual-token budget, but only when the right stale tokens are refreshed. Raw-attention selection can waste budget on high-attention tokens with small value-norm proxy scores and on query-irrelevant images. To address these failure modes, we propose CONDUIT, a training-free refresh policy that unifies single- and multi-image reuse as residual-stream restoration. Building on norm-weighted attention, CONDUIT ranks cached visual tokens using cached-key query attention and an accessible pre-output cached-value-norm proxy, then applies empirical image-level relevance amplification before one global selection. With one image, the coefficient is one and the rule reduces to intra-image token selection. The method preserves model architecture and weights, adding only a single query-conditioned scoring pass at inference. At a 10% refresh budget, CONDUIT achieves 97.0-99.5% of the corresponding full-prefill five-dataset average across three VLM backbones and leads budgeted methods on average; on the MMLongBench-Doc latency subset, it uses 13.5% of full-prefill FLOPs and achieves a 2.99x time-to-first-token speedup.

cs.CL↗

Breaking Contextual Inertia: Reinforcement Learning with Single-Turn Anchors for Stable Multi-Turn Interaction

While LLMs demonstrate strong reasoning capabilities when provided with full information in a single turn, they exhibit substantial vulnerability in multi-turn interactions. Specifically, when information is revealed incrementally or requires updates, models frequently fail to integrate new constraints, leading to a collapse in performance compared to their single-turn baselines. We term the root cause as \emph{Contextual Inertia}: a phenomenon where models rigidly adhere to previous reasoning traces. Even when users explicitly provide corrections or new data in later turns, the model ignores them, preferring to maintain consistency with its previous (incorrect) reasoning path. To address this, we introduce \textbf{R}einforcement \textbf{L}earning with \textbf{S}ingle-\textbf{T}urn \textbf{A}nchors (\textbf{RLSTA}), a generalizable training approach designed to stabilize multi-turn interaction across diverse scenarios and domains. RLSTA leverages the model's superior single-turn capabilities as stable internal anchors to provide reward signals. By aligning multi-turn responses with these anchors, RLSTA empowers models to break contextual inertia and self-calibrate their reasoning based on the latest information. Experiments show that RLSTA significantly outperforms standard fine-tuning and abstention-based methods. Notably, our method exhibits strong cross-domain generalization (e.g., math to code) and proves effective even without external verifiers, highlighting its potential for general-domain applications. Code is available at https://github.com/Tencent/RLSTA.

cs.AI↗

BEYOND DIALOGUE: A Profile-Dialogue Alignment Framework Towards General Role-Playing Language Model

The rapid advancement of large language models (LLMs) has revolutionized role-playing, enabling the development of general role-playing models. However, current role-playing training has two significant issues: (I) Using a predefined role profile to prompt dialogue training for specific scenarios usually leads to inconsistencies and even conflicts between the dialogue and the profile, resulting in training biases. (II) The model learns to imitate the role based solely on the profile, neglecting profile-dialogue alignment at the sentence level. In this work, we propose a simple yet effective framework called BEYOND DIALOGUE, designed to overcome these hurdles. This framework innovatively introduces "beyond dialogue" tasks to align dialogue with profile traits based on each specific scenario, thereby eliminating biases during training. Furthermore, by adopting an innovative prompting mechanism that generates reasoning outcomes for training, the framework allows the model to achieve fine-grained alignment between profile and dialogue at the sentence level. The aforementioned methods are fully automated and low-cost. Additionally, the integration of automated dialogue and objective evaluation methods forms a comprehensive framework, paving the way for general role-playing. Experimental results demonstrate that our model excels in adhering to and reflecting various dimensions of role profiles, outperforming most proprietary general and specialized role-playing baselines. All code and datasets are available at https://github.com/yuyouyu32/BeyondDialogue.

cs.CL↗

Cultivating Game Sense for Yourself: Making VLMs Gaming Experts

Developing agents capable of fluid gameplay in first/third-person games without API access remains a critical challenge in Artificial General Intelligence (AGI). Recent efforts leverage Vision Language Models (VLMs) as direct controllers, frequently pausing the game to analyze screens and plan action through language reasoning. However, this inefficient paradigm fundamentally restricts agents to basic and non-fluent interactions: relying on isolated VLM reasoning for each action makes it impossible to handle tasks requiring high reactivity (e.g., FPS shooting) or dynamic adaptability (e.g., ACT combat). To handle this, we propose a paradigm shift in gameplay agent design: instead of directly controlling gameplay, VLM develops specialized execution modules tailored for tasks like shooting and combat. These modules handle real-time game interactions, elevating VLM to a high-level developer. Building upon this paradigm, we introduce GameSense, a gameplay agent framework where VLM develops task-specific game sense modules by observing task execution and leveraging vision tools and neural network training pipelines. These modules encapsulate action-feedback logic, ranging from direct action rules to neural network-based decisions. Experiments demonstrate that our framework is the first to achieve fluent gameplay in diverse genres, including ACT, FPS, and Flappy Bird, setting a new benchmark for game-playing agents.

cs.CL↗

Multi-Task Model Merging via Adaptive Weight Disentanglement

Model merging has recently gained attention as an economical and scalable approach to incorporate task-specific weights from various tasks into a unified multi-task model. For example, in Task Arithmetic (TA), adding the fine-tuned weights of different tasks can enhance the model's performance on those tasks, while subtracting them leads to task forgetting. Although TA is highly effective, interference among task still hampers the performance of the merged model. Existing methods for handling conflicts between task generally rely on empirical selection, resulting in suboptimal performance. In this paper, we introduce an Adaptive Weight Disentanglement method. We begin by theoretically proving that task vectors employed in model merging should be orthogonal to minimize interference among tasks. Guided by this insight, we initialize redundant vectors such that, when subtracted from the original task vectors, the resulting vectors exhibit increased orthogonality. Additionally, we impose an norm constraint on the redundant vectors to preserve the performance of the task-specific models. Experimental results demonstrate the effectiveness of our proposed technique: it successfully extracts redundant vectors, and after their subtraction, the task vectors not only retain robust performance but also achieve superior fusion outcomes. Our code is available at \href{https://github.com/FarisXiong/AWD.git}{https://github.com/FarisXiong/AWD.git}.

cs.LG↗

Unveiling and Consulting Core Experts in Retrieval-Augmented MoE-based LLMs

Retrieval-Augmented Generation (RAG) significantly improved the ability of Large Language Models (LLMs) to solve knowledge-intensive tasks. While existing research seeks to enhance RAG performance by retrieving higher-quality documents or designing RAG-specific LLMs, the internal mechanisms within LLMs that contribute to the effectiveness of RAG systems remain underexplored. In this paper, we aim to investigate these internal mechanisms within the popular Mixture-of-Expert (MoE)-based LLMs and demonstrate how to improve RAG by examining expert activations in these LLMs. Our controlled experiments reveal that several core groups of experts are primarily responsible for RAG-related behaviors. The activation of these core experts can signify the model's inclination towards external/internal knowledge and adjust its behavior. For instance, we identify core experts that can (1) indicate the sufficiency of the model's internal knowledge, (2) assess the quality of retrieved documents, and (3) enhance the model's ability to utilize context. Based on these findings, we propose several strategies to enhance RAG's efficiency and effectiveness through expert activation. Experimental results across various datasets and MoE-based LLMs show the effectiveness of our method.

cs.AI↗

Compressing Deep Graph Neural Networks via Adversarial Knowledge Distillation

Deep graph neural networks (GNNs) have been shown to be expressive for modeling graph-structured data. Nevertheless, the over-stacked architecture of deep graph models makes it difficult to deploy and rapidly test on mobile or embedded systems. To compress over-stacked GNNs, knowledge distillation via a teacher-student architecture turns out to be an effective technique, where the key step is to measure the discrepancy between teacher and student networks with predefined distance functions. However, using the same distance for graphs of various structures may be unfit, and the optimal distance formulation is hard to determine. To tackle these problems, we propose a novel Adversarial Knowledge Distillation framework for graph models named GraphAKD, which adversarially trains a discriminator and a generator to adaptively detect and decrease the discrepancy. Specifically, noticing that the well-captured inter-node and inter-class correlations favor the success of deep GNNs, we propose to criticize the inherited knowledge from node-level and class-level views with a trainable discriminator. The discriminator distinguishes between teacher knowledge and what the student inherits, while the student GNN works as a generator and aims to fool the discriminator. To our best knowledge, GraphAKD is the first to introduce adversarial training to knowledge distillation in graph domains. Experiments on node-level and graph-level classification benchmarks demonstrate that GraphAKD improves the student performance by a large margin. The results imply that GraphAKD can precisely transfer knowledge from a complicated teacher GNN to a compact student GNN.

cs.LG↗

Learning Hierarchy-Aware Knowledge Graph Embeddings for Link Prediction

Knowledge graph embedding, which aims to represent entities and relations as low dimensional vectors (or matrices, tensors, etc.), has been shown to be a powerful technique for predicting missing links in knowledge graphs. Existing knowledge graph embedding models mainly focus on modeling relation patterns such as symmetry/antisymmetry, inversion, and composition. However, many existing approaches fail to model semantic hierarchies, which are common in real-world applications. To address this challenge, we propose a novel knowledge graph embedding model -- namely, Hierarchy-Aware Knowledge Graph Embedding (HAKE) -- which maps entities into the polar coordinate system. HAKE is inspired by the fact that concentric circles in the polar coordinate system can naturally reflect the hierarchy. Specifically, the radial coordinate aims to model entities at different levels of the hierarchy, and entities with smaller radii are expected to be at higher levels; the angular coordinate aims to distinguish entities at the same level of the hierarchy, and these entities are expected to have roughly the same radii but different angles. Experiments demonstrate that HAKE can effectively model the semantic hierarchies in knowledge graphs, and significantly outperforms existing state-of-the-art methods on benchmark datasets for the link prediction task.

cs.LG↗

Duality-Induced Regularizer for Semantic Matching Knowledge Graph Embeddings

Semantic matching models -- which assume that entities with similar semantics have similar embeddings -- have shown great power in knowledge graph embeddings (KGE). Many existing semantic matching models use inner products in embedding spaces to measure the plausibility of triples and quadruples in static and temporal knowledge graphs. However, vectors that have the same inner products with another vector can still be orthogonal to each other, which implies that entities with similar semantics may have dissimilar embeddings. This property of inner products significantly limits the performance of semantic matching models. To address this challenge, we propose a novel regularizer -- namely, DUality-induced RegulArizer (DURA) -- which effectively encourages the entities with similar semantics to have similar embeddings. The major novelty of DURA is based on the observation that, for an existing semantic matching KGE model (primal), there is often another distance based KGE model (dual) closely associated with it, which can be used as effective constraints for entity embeddings. Experiments demonstrate that DURA consistently and significantly improves the performance of state-of-the-art semantic matching models on both static and temporal knowledge graph benchmarks.

cs.CL↗

Rethinking Graph Convolutional Networks in Knowledge Graph Completion

Graph convolutional networks (GCNs) -- which are effective in modeling graph structures -- have been increasingly popular in knowledge graph completion (KGC). GCN-based KGC models first use GCNs to generate expressive entity representations and then use knowledge graph embedding (KGE) models to capture the interactions among entities and relations. However, many GCN-based KGC models fail to outperform state-of-the-art KGE models though introducing additional computational complexity. This phenomenon motivates us to explore the real effect of GCNs in KGC. Therefore, in this paper, we build upon representative GCN-based KGC models and introduce variants to find which factor of GCNs is critical in KGC. Surprisingly, we observe from experiments that the graph structure modeling in GCNs does not have a significant impact on the performance of KGC models, which is in contrast to the common belief. Instead, the transformations for entity representations are responsible for the performance improvements. Based on the observation, we propose a simple yet effective framework named LTE-KGE, which equips existing KGE models with linearly transformed entity embeddings. Experiments demonstrate that LTE-KGE models lead to similar performance improvements with GCN-based KGC methods, while being more computationally efficient. These results suggest that existing GCNs are unnecessary for KGC, and novel GCN-based KGC models should count on more ablation studies to validate their effectiveness. The code of all the experiments is available on GitHub at https://github.com/MIRALab-USTC/GCN4KGC.

cs.AI↗

ConE: Cone Embeddings for Multi-Hop Reasoning over Knowledge Graphs

Query embedding (QE) -- which aims to embed entities and first-order logical (FOL) queries in low-dimensional spaces -- has shown great power in multi-hop reasoning over knowledge graphs. Recently, embedding entities and queries with geometric shapes becomes a promising direction, as geometric shapes can naturally represent answer sets of queries and logical relationships among them. However, existing geometry-based models have difficulty in modeling queries with negation, which significantly limits their applicability. To address this challenge, we propose a novel query embedding model, namely Cone Embeddings (ConE), which is the first geometry-based QE model that can handle all the FOL operations, including conjunction, disjunction, and negation. Specifically, ConE represents entities and queries as Cartesian products of two-dimensional cones, where the intersection and union of cones naturally model the conjunction and disjunction operations. By further noticing that the closure of complement of cones remains cones, we design geometric complement operators in the embedding space for the negation operations. Experiments demonstrate that ConE significantly outperforms existing state-of-the-art methods on benchmark datasets.

cs.AI↗

Technical Report of Team GraphMIRAcles in the WikiKG90M-LSC Track of OGB-LSC @ KDD Cup 2021

Link prediction in large-scale knowledge graphs has gained increasing attention recently. The OGB-LSC team presented OGB Large-Scale Challenge (OGB-LSC), a collection of three real-world datasets for advancing the state-of-the-art in large-scale graph machine learning. In this paper, we introduce the solution of our team GraphMIRAcles in the WikiKG90M-LSC track of OGB-LSC @ KDD Cup 2021. In the WikiKG90M-LSC track, the goal is to automatically predict missing links in WikiKG90M, a large scale knowledge graph extracted from Wikidata. To address this challenge, we propose a framework that integrates three components -- a basic model ComplEx-CMRC, a rule miner AMIE 3, and an inference model to predict missing links. Experiments demonstrate that our solution achieves an MRR of 0.9707 on the test dataset. Moreover, as the knowledge distillation in the inference model uses test tail candidates -- which are unavailable in practice -- we conduct ablation studies on knowledge distillation. Experiments demonstrate that our model without knowledge distillation achieves an MRR of 0.9533 on the full validation dataset.

cs.CL↗

Duality-Induced Regularizer for Tensor Factorization Based Knowledge Graph Completion

Tensor factorization based models have shown great power in knowledge graph completion (KGC). However, their performance usually suffers from the overfitting problem seriously. This motivates various regularizers -- such as the squared Frobenius norm and tensor nuclear norm regularizers -- while the limited applicability significantly limits their practical usage. To address this challenge, we propose a novel regularizer -- namely, DUality-induced RegulArizer (DURA) -- which is not only effective in improving the performance of existing models but widely applicable to various methods. The major novelty of DURA is based on the observation that, for an existing tensor factorization based KGC model (primal), there is often another distance based KGC model (dual) closely associated with it. Experiments show that DURA yields consistent and significant improvements on benchmarks.

cs.LG↗