Searcharxiv⌕ Search

arXiv subjects

Xinbing Wang

Publications and source records attributed to Xinbing Wang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Structured In-context Environment Scaling for Large Language Model Reasoning

Large language models (LLMs) have achieved significant advancements in reasoning capabilities through reinforcement learning (RL) via environmental exploration. As the intrinsic properties of the environment determine the abilities that LLMs can learn, the environment plays a important role in the RL finetuning process. An ideal LLM reasoning environment should possess three core characteristics: scalability, generalizable reasoning, and verifiability. However, existing mathematical and coding environments are difficult to scale due to heavy reliance on expert annotation, while the skills learned in game-based environments are too specialized to generalize. To bridge this gap, we introduce the \textbf{S}tructured \textbf{I}n-context \textbf{E}nvironment (SIE) framework. SIE achieves scalability by automatically constructing reasoning environments from large-scale structured data, where the rich compositional patterns naturally support generalizable reasoning. Moreover, the explicit schemas and reasoning chains in structured data provide a foundation for rule-based verifiability. Experimental results show that SIE framework not only achieves substantial improvements in in-domain structured reasoning, but also enables the learned compositional reasoning skills to generalize effectively to out-of-domain mathematical and logical reasoning tasks. We further explored learning in information-limited partial SIEs and found that LLMs can infer the missing information through exploring the environment, leading to robust reasoning improvements and generalization performance.

cs.CL↗
arXiv2026

VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models

Visual token pruning methods effectively mitigate the quadratic computational growth caused by processing high-resolution images and video frames in vision-language models (VLMs). However, existing approaches rely on predefined pruning configurations without determining whether they achieve computation-performance optimality. In this work, we introduce , a novel framework that formulates visual token pruning as a Pareto configuration optimization problem to automatically identify optimal configurations. Our approach employs continuous relaxation and straight-through estimators to enable gradient-based search, solved via the Augmented Lagrangian method. Extensive experiments across 8 visual benchmarks demonstrate that effectively approximates the empirical Pareto frontier obtained through grid search and generalizes well across various pruning methods and VLM architectures. Furthermore, through learnable kernel functions, we investigate layer-wise pruning patterns and reveal that multi-step progressive pruning captures VLMs' hierarchical compression structure, achieving superior accuracy-efficiency trade-offs compared to single-layer approaches.

cs.CV↗
arXiv2026

Geminet: Learning the Duality-based Iterative Process for Lightweight Traffic Engineering in Changing Topologies

Recently, researchers have explored ML-based Traffic Engineering (TE), leveraging neural networks to solve TE problems traditionally addressed by optimization. However, existing ML-based TE schemes remain impractical: they either fail to handle topology changes or suffer from poor scalability due to excessive computational and memory overhead. To overcome these limitations, we propose Geminet, a lightweight and scalable ML-based TE framework that can handle changing topologies. Geminet is built upon two key insights: (i) a methodology that decouples neural networks from topology by learning an iterative gradient-descent-based adjustment process, as the update rule of gradient descent is topology-agnostic, relying only on a few gradient-related quantities; (ii) shifting optimization from path-level routing weights to edge-level dual variables, reducing memory consumption by leveraging the fact that edges are far fewer than paths. Evaluations on WAN and data center datasets show that Geminet significantly improves scalability. Its neural network size is only 0.04% to 7% of existing schemes, while handling topology variations as effectively as HARP, a state-of-the-art ML-based TE approach, without performance degradation. When trained on large-scale topologies, Geminet consumes under 10 GiB of memory, more than eight times less than the 80-plus GiB required by HARP, while achieving 5.45 times faster convergence speed, demonstrating its potential for large-scale deployment.

cs.NI↗
arXiv2026

Inductive Reasoning for Temporal Knowledge Graphs with Emerging Entities

Reasoning on Temporal Knowledge Graphs (TKGs) is essential for predicting future events and time-aware facts. While existing methods are effective at capturing relational dynamics, their performance is limited by a closed-world assumption, which fails to account for emerging entities not present in the training. Notably, these entities continuously join the network without historical interactions. Empirical study reveals that emerging entities are widespread in TKGs, comprising roughly 25\% of all entities. The absence of historical interactions of these entities leads to significant performance degradation in reasoning tasks. Whereas, we observe that entities with semantic similarities often exhibit comparable interaction histories, suggesting the presence of transferable temporal patterns. Inspired by this insight, we propose TransFIR (Transferable Inductive Reasoning), a novel framework that leverages historical interaction sequences from semantically similar known entities to support inductive reasoning. Specifically, we propose a codebook-based classifier that categorizes emerging entities into latent semantic clusters, allowing them to adopt reasoning patterns from similar entities. Experimental results demonstrate that TransFIR outperforms all baselines in reasoning on emerging entities, achieving an average improvement of 28.6% in Mean Reciprocal Rank (MRR) across multiple datasets. The implementations are available at https://github.com/zhaodazhuang2333/TransFIR.

cs.AI↗
arXiv2026

Leaf-centric Logical Topology Design for OCS-based GPU Clusters

Recent years have witnessed the growing deployment of optical circuit switches (OCS) in commercial GPU clusters (e.g., Google A3 GPU cluster) optimized for machine learning (ML) workloads. Such clusters adopt a three-tier leaf-spine-OCS topology, servers attach to leaf-layer electronic packet switches (EPSes); these leaf switches aggregate into spine-layer EPSes to form a Pod; and multiple Pods are interconnected via core-layer OCSes. Unlike EPSes, OCSes only support circuit-based paths between directly connected spine switches, potentially inducing a phenomenon termed routing polarization, which refers to the scenario where the bandwidth requirements between specific pairs of Pods are unevenly fulfilled through links among different spine switches. The resulting imbalance induces traffic contention and bottlenecks on specific leaf-to-spine links, ultimately reducing ML training throughput. To mitigate this issue, we introduce a leaf-centric paradigm to ensure traffic originating from the same leaf switch is evenly distributed across multiple spine switches with balanced loads. Through rigorous theoretical analysis, we establish a sufficient condition for avoiding routing polarization and propose a corresponding logical topology design algorithm with polynomial-time complexity. Large-scale simulations validate up to 19.27% throughput improvement and a 99.16% reduction in logical topology computation overhead compared to Mixed Integer Programming (MIP)-based methods.

cs.NI↗
arXiv2026

Pretraining with Token-Level Adaptive Latent Chain-of-Thought

Scaling large language models by increasing parameters and training data is increasingly constrained by limited high-quality corpora and rising communication costs. This work explores an alternative axis: increasing per-token computation without expanding parameters, by internalizing latent Chain-of-Thought (CoT) into pretraining. We propose Pretraining with Token-Level Adaptive Latent CoT (adaptive latent CoT), where the model generates a variable-length latent CoT trajectory before emitting each token -- allocating longer trajectories to difficult tokens and shorter (or even zero) trajectories to easy ones. Importantly, this behavior emerges naturally from one-stage pretraining on general text and reduces computation in both training and inference via token-wise adaptive halting. Experiments with Llama architectures show that adaptive latent CoT consistently improves language modeling perplexity and broad downstream accuracy, even with fewer training FLOPs than prior recurrent baselines.

cs.CL↗
arXiv2026

PonderLM-2: Pretraining LLM with Latent Thoughts in Continuous Space

The remarkable success of Chain-of-Thought (CoT), which enhances performance by scaling generation steps at test-time, inspires us to ask: can we leverage a similar scaling of computational steps during pretraining to improve the generation of each individual token? To address this, we propose a novel pre-training methodology: Pretraining Language Models with Latent Thoughts (PonderLM-2). Our approach pretrains a language model (LM) to first generate an intermediate latent thought-the last hidden state of the current position-which is then used as input to predict the actual subsequent token. This additional computational step enables the LM to refine its prediction within unconstrained continuous space. Our experiments demonstrate that, at an identical inference cost, a LM that generates one additional latent thought per token outperforms a standard model with double the parameters. For instance, our PonderLM-2-Pythia-1.4B, pretrained on 300B tokens from the Pile, significantly surpasses the vanilla Pythia-2.8B trained on the same data on both language modeling and a range of general downstream tasks. Furthermore, increasing the number of latent thoughts generated before each actual token-forming a chain analogous to CoT-consistently improves the model's performance. The code is available at https://github.com/LUMIA-Group/PonderLM-2.

cs.CL↗
arXiv2026

RADAR: Reasoning as Discrimination with Aligned Representations for LLM-based Knowledge Graph Reasoning

Knowledge graph reasoning (KGR) infers missing facts, with recent advances increasingly harnessing the semantic priors and reasoning abilities of Large Language Models (LLMs). However, prevailing generative paradigms are prone to memorizing surface-level co-occurrences rather than learning genuine relational semantics, limiting out-of-distribution generalization. To address this, we propose RADAR, which reformulates KGR from generative pattern matching to discriminative relational reasoning. We recast KGR as discriminative entity selection, where reinforcement learning enforces relative entity separability beyond token-likelihood imitation. Leveraging this separability, inference operates directly in representation space, ensuring consistency with the discriminative optimization and bypassing generation-induced hallucinations. Across four benchmarks, RADAR achieves 5-6% relative gains on link prediction and triple classification over strong LLM baselines, while increasing task-relevant mutual information in intermediate representations by 62.9%, indicating more robust and transferable relational reasoning.

cs.CL↗
arXiv2026

FastReChain: A Novel Bidirectional Model-Based Algorithm for Topology Engineering of OCS-Based Clusters

Optical Circuit Switching (OCS) technology is increasingly being adopted in data centers due to its advantages of low power consumption and low technology refresh costs. Unlike electrical packet switches, OCS provides programmable bandwidth for directly connected devices by configuring the mapping relationships of internal ports. Thus, how to calculate these internal port mapping relationships, i.e., Topology Engineering (ToE), is one of the key designs of OCS-based clusters. Current deployments usually design ToE algorithms by solving Integer Linear Programming (ILP) models, with the aim of minimizing modifications to links occupied by running tasks as much as possible. However, ILP-based ToE algorithms may incur excessive runtime overhead in large-scale clusters. Some existing ToE algorithms convert the ILP model into a Minimum-Cost Flow model through greedy construction, yet such greedy strategies may increase the number of affected links during the OCS reconfiguration process. To solve the aforementioned problems, we propose a novel bidirectional modeling approach, along with a corresponding FastReChain algorithm in this paper. We verify the superiority of this algorithm through simulation experiments based on real-trace data.

cs.NI↗
arXiv2026

PonderLM: Pretraining Language Models to Ponder in Continuous Space

Humans ponder before articulating complex sentence elements, enabling deeper cognitive processing through focused effort. In this work, we introduce this pondering process into language models by repeatedly invoking the forward process within a single token generation step. During pondering, instead of generating an actual token sampled from the prediction distribution, the model ponders by yielding a weighted sum of all token embeddings according to the predicted token distribution. The generated embedding is then fed back as input for another forward pass. We show that the model can learn to ponder in this way through self-supervised learning, without any human annotations. Experiments across three widely used open-source architectures-GPT-2, Pythia, and LLaMA-and extensive downstream task evaluations demonstrate the effectiveness and generality of our method. On 9 downstream benchmarks, our pondering-enhanced Pythia models significantly outperform the official Pythia models. Notably, our PonderPythia models demonstrate remarkable effectiveness: PonderPythia-2.8B surpasses Pythia-6.9B and rivals Pythia-12B, while our PonderPythia-1B matches TinyLlama-1.1B, a model trained on 10 times more data. The code is available at https://github.com/LUMIA-Group/PonderingLM.

cs.CL↗
arXiv2026

AWM: Accurate Weight-Matrix Fingerprint for Large Language Models

Protecting the intellectual property of large language models (LLMs) is crucial, given the substantial resources required for their training. Consequently, there is an urgent need for both model owners and third parties to determine whether a suspect LLM is trained from scratch or derived from an existing base model. However, the intensive post-training processes that models typically undergo-such as supervised fine-tuning, extensive continued pretraining, reinforcement learning, multi-modal extension, pruning, and upcycling-pose significant challenges to reliable identification. In this work, we propose a training-free fingerprinting method based on weight matrices. We leverage the Linear Assignment Problem (LAP) and an unbiased Centered Kernel Alignment (CKA) similarity to neutralize the effects of parameter manipulations, yielding a highly robust and high-fidelity similarity metric. On a comprehensive testbed of 60 positive and 90 negative model pairs, our method demonstrates exceptional robustness against all six aforementioned post-training categories while exhibiting a near-zero risk of false positives. By achieving perfect scores on all classification metrics, our approach establishes a strong basis for reliable model lineage verification. Moreover, the entire computation completes within 30s on an NVIDIA 3090 GPU. The code is available at https://github.com/LUMIA-Group/AWM.

cs.CL↗
arXiv2026

Context-level Language Modeling by Learning Predictive Context Embeddings

We propose ContextLM, a framework that implicitly learns multi-token prediction by augmenting standard pretraining with an intrinsic next-context prediction objective. ContextLM builds a language model on top of context embeddings that span multiple tokens, enabling better next-token prediction by predicting the next context. Our model is fully compatible with standard autoregressive, token-by-token evaluation paradigms (e.g., perplexity). Extensive experiments with GPT-2 and Pythia backbones (up to 1.5B parameters and 300B training tokens) reveal that ContextLM shifts the Pareto frontier of scaling laws, exhibiting superior efficiency in parameters, training tokens, and FLOPs. Our results show that ContextLM could already achieve the baseline perplexity using 39\% fewer parameters and demonstrates robust generalization improvements on extensive downstream tasks under equivalent parameter counts.

cs.CL↗
arXiv2026

Flow of Spans: Generalizing Language Models to Dynamic Span-Vocabulary via GFlowNets

Standard autoregressive language models generate text token-by-token from a fixed vocabulary, inducing a tree-structured state space when viewing token sampling as an action, which limits flexibility and expressiveness. Recent work introduces dynamic vocabulary by sampling retrieved text spans but overlooks that the same sentence can be composed of spans of varying lengths, lacking explicit modeling of the directed acyclic graph (DAG) state space. This leads to restricted exploration of compositional paths and is biased toward the chosen path. Generative Flow Networks (GFlowNets) are powerful for efficient exploring and generalizing over state spaces, particularly those with a DAG structure. However, prior GFlowNets-based language models operate at the token level and remain confined to tree-structured spaces, limiting their potential. In this work, we propose Flow of SpanS (FOSS), a principled GFlowNets framework for span generation. FoSS constructs a dynamic span vocabulary by segmenting the retrieved text flexibly, ensuring a DAG-structured state space, which allows GFlowNets to explore diverse compositional paths and improve generalization. With specialized reward models, FoSS generates diverse, high-quality text. Empirically, FoSS improves MAUVE scores by up to 12.5% over Transformer on text generation and achieves 3.5% gains on knowledge-intensive tasks, consistently outperforming state-of-the-art methods. Scaling experiments further demonstrate FoSS benefits from larger models, more data, and richer retrieval corpora, retaining its advantage over strong baselines.

cs.AI↗
arXiv2026

: One LLM Token for Explicit Graph Structural Understanding

Large language models show great potential in unstructured data understanding, but still face significant challenges with graphs due to their structural hallucination. Existing approaches mainly either verbalize graphs into natural language, which leads to excessive token consumption and scattered attention, or transform graphs into trainable continuous embeddings (i.e., soft prompt), but exhibit severe misalignment with original text tokens. To solve this problem, we propose to incorporate one special token to fully represent the Structure Of Graph within a unified token space, facilitating explicit topology input and structural information sharing. Specifically, we propose a topology-aware structural tokenizer that maps each graph topology into a highly selective single token. Afterwards, we construct a set of hybrid structure Question-Answering corpora to align new structural tokens with existing text tokens. With this approach, empowers LLMs to understand, generate, and reason in a concise and accurate manner. Extensive experiments on five graph-level benchmarks demonstrate the superiority of our method, achieving a performance improvement of 9.9% to 41.4% compared to the baselines while exhibiting interpretability and consistency. Furthermore, our method provides a flexible extension to node-level tasks, enabling both global and local structural understanding. The codebase is publicly available at https://github.com/Jingyao-Wu/SOG.

cs.CL↗
arXiv2026

FLAME: Empowering Frozen LLMs for Knowledge Graph Completion

Traditional knowledge graph completion (KGC) methods rely solely on structural information and struggle with sparsity, while Large Language Models (LLMs) address these limitations through rich world knowledge and strong context modeling. Fine-tuning LLMs is effective but costly, while non-fine-tuned LLMs are efficient but suboptimal. To address this trade-off, we propose \textbf{FLAME}, a framework that extracts context-aware hidden states from intermediate layers of frozen LLMs to train data-efficient KGC classifiers. We bridge LLM-KG semantic gaps via subgraph-based entity descriptions and employ sliced mutual information (SMI) to quantify task-relevant information in representations. Experiments demonstrate that FLAME achieves 47\% improvement over non-fine-tuned LLM baselines and, to our knowledge, is the first to achieve fine-tuned performance with $188\times$ memory efficiency and $26.11\times$ speedup.

cs.CL↗
arXiv2026

Controlling Underestimation Bias in Constrained Reinforcement Learning for Safe Exploration

Constrained Reinforcement Learning (CRL) aims to maximize cumulative rewards while satisfying constraints. However, existing CRL algorithms often encounter significant constraint violations during training, limiting their applicability in safety-critical scenarios. In this paper, we identify the underestimation of the cost value function as a key factor contributing to these violations. To address this issue, we propose the Memory-driven Intrinsic Cost Estimation (MICE) method, which introduces intrinsic costs to mitigate underestimation and control bias to promote safer exploration. Inspired by flashbulb memory, where humans vividly recall dangerous experiences to avoid risks, MICE constructs a memory module that stores previously explored unsafe states to identify high-cost regions. The intrinsic cost is formulated as the pseudo-count of the current state visiting these risk regions. Furthermore, we propose an extrinsic-intrinsic cost value function that incorporates intrinsic costs and adopts a bias correction strategy. Using this function, we formulate an optimization objective within the trust region, along with corresponding optimization methods. Theoretically, we provide convergence guarantees for the proposed cost value function and establish the worst-case constraint violation for the MICE update. Extensive experiments demonstrate that MICE significantly reduces constraint violations while preserving policy performance comparable to baselines.

cs.LG↗
arXiv2026

Extreme Value Policy Optimization for Safe Reinforcement Learning

Ensuring safety is a critical challenge in applying Reinforcement Learning (RL) to real-world scenarios. Constrained Reinforcement Learning (CRL) addresses this by maximizing returns under predefined constraints, typically formulated as the expected cumulative cost. However, expectation-based constraints overlook rare but high-impact extreme value events in the tail distribution, such as black swan incidents, which can lead to severe constraint violations. To address this issue, we propose the Extreme Value policy Optimization (EVO) algorithm, leveraging Extreme Value Theory (EVT) to model and exploit extreme reward and cost samples, reducing constraint violations. EVO introduces an extreme quantile optimization objective to explicitly capture extreme samples in the cost tail distribution. Additionally, we propose an extreme prioritization mechanism during replay, amplifying the learning signal from rare but high-impact extreme samples. Theoretically, we establish upper bounds on expected constraint violations during policy updates, guaranteeing strict constraint satisfaction at a zero-violation quantile level. Further, we demonstrate that EVO achieves a lower probability of constraint violations than expectation-based methods and exhibits lower variance than quantile regression methods. Extensive experiments show that EVO significantly reduces constraint violations during training while maintaining competitive policy performance compared to baselines.

cs.LG↗
arXiv2026

KaLM: Knowledge-aligned Autoregressive Language Modeling via Dual-view Knowledge Graph Contrastive Learning

Autoregressive large language models (LLMs) pre-trained by next token prediction are inherently proficient in generative tasks. However, their performance on knowledge-driven tasks such as factual knowledge querying remains unsatisfactory. Knowledge graphs (KGs), as high-quality structured knowledge bases, can provide reliable knowledge for LLMs, potentially compensating for their knowledge deficiencies. Aligning LLMs with explicit, structured knowledge from KGs has been a challenge; previous attempts either failed to effectively align knowledge representations or compromised the generative capabilities of LLMs, leading to less-than-optimal outcomes. This paper proposes \textbf{KaLM}, a \textit{Knowledge-aligned Language Modeling} approach, which fine-tunes autoregressive LLMs to align with KG knowledge via the joint objective of explicit knowledge alignment and implicit knowledge alignment. The explicit knowledge alignment objective aims to directly optimize the knowledge representation of LLMs through dual-view knowledge graph contrastive learning. The implicit knowledge alignment objective focuses on incorporating textual patterns of knowledge into LLMs through triple completion language modeling. Notably, our method achieves a significant performance boost in evaluations of knowledge-driven tasks, specifically embedding-based knowledge graph completion and generation-based knowledge graph question answering.

cs.CL↗