Large Reasoning Models (LRMs) significantly improve the reasoning ability of Large Language Models (LLMs) by learning to reason, exhibiting promising performance in complex task-solving. However, their deliberative reasoning process leads to inefficiencies in token usage, memory consumption, and inference time. Thus, this survey provides a review of efficient inference methods designed specifically for LRMs, focusing on mitigating token inefficiency while preserving the reasoning quality. First, we introduce a taxonomy to group the recent methods into two main categories: (a) explicit compact Chain-of-Thought (CoT), which reduces tokens while keeping the explicit reasoning structure, and (b) implicit latent CoT, which encodes reasoning steps within hidden representations instead of explicit tokens. Meanwhile, we discuss their strengths and weaknesses. Then, we conduct empirical analyses on existing methods from performance and efficiency aspects. Besides, we present open challenges in this field, including human-centric controllable reasoning, trade-off between interpretability and efficiency of reasoning, ensuring safety of efficient reasoning, and broader applications of efficient reasoning. In addition, we highlight key insights for enhancing LRMs' inference efficiency via techniques such as model merging, new architectures, and agent routers. We hope this work serves as a valuable guide, helping researchers overcome challenges in this vibrant field\footnote{https://github.com/yueliu1999/Awesome-Efficient-Inference-for-LRMs}.

Large Language Models (LLMs) are increasingly employed as automated evaluators to assess the safety of generated content, yet their reliability in this role remains uncertain. This study evaluates a diverse set of 11 LLM judge models across critical safety domains, examining three key aspects: self-consistency in repeated judging tasks, alignment with human judgments, and susceptibility to input artifacts such as apologetic or verbose phrasing. Our findings reveal that biases in LLM judges can significantly distort the final verdict on which content source is safer, undermining the validity of comparative evaluations. Notably, apologetic language artifacts alone can skew evaluator preferences by up to 98\%. Contrary to expectations, larger models do not consistently exhibit greater robustness, while smaller models sometimes show higher resistance to specific artifacts. To mitigate LLM evaluator robustness issues, we investigate jury-based evaluations aggregating decisions from multiple models. Although this approach both improves robustness and enhances alignment to human judgements, artifact sensitivity persists even with the best jury configurations. These results highlight the urgent need for diversified, artifact-resistant methodologies to ensure reliable safety assessments.

Equivalence checking, i.e., determining whether two programs produce identical outputs for all possible inputs, underpins a broad range of applications, including software refactoring, testing, and optimization. We present the task of equivalence checking as a new way to evaluate the code reasoning abilities of large language models (LLMs). We introduce EquiBench, a dataset of 2400 program pairs spanning four programming languages and six equivalence categories. These pairs are systematically generated through program analysis, compiler scheduling, and superoptimization, covering nontrivial structural transformations that demand deep semantic reasoning beyond simple syntactic variations. Our evaluation of 17 state-of-the-art LLMs shows that OpenAI o3-mini achieves the highest overall accuracy of 78.0%. In the most challenging categories, the best accuracies are 62.3% and 68.8%, only modestly above the 50% random baseline for binary classification, indicating significant room for improvement in current models' code reasoning capabilities.

In this work, we introduce a new framework for active experimentation, the Prediction-Guided Active Experiment (PGAE), which leverages predictions from an existing machine learning model to guide sampling and experimentation. Specifically, at each time step, an experimental unit is sampled according to a designated sampling distribution, and the actual outcome is observed based on an experimental probability. Otherwise, only a prediction for the outcome is available. We begin by analyzing the non-adaptive case, where full information on the joint distribution of the predictor and the actual outcome is assumed. For this scenario, we derive an optimal experimentation strategy by minimizing the semi-parametric efficiency bound for the class of regular estimators. We then introduce an estimator that meets this efficiency bound, achieving asymptotic optimality. Next, we move to the adaptive case, where the predictor is continuously updated with newly sampled data. We show that the adaptive version of the estimator remains efficient and attains the same semi-parametric bound under certain regularity assumptions. Finally, we validate PGAE's performance through simulations and a semi-synthetic experiment using data from the US Census Bureau. The results underscore the PGAE framework's effectiveness and superiority compared to other existing methods.

Despite significant advancements in salient object detection(SOD) in optical remote sensing images(ORSI), challenges persist due to the intricate edge structures of ORSIs and the complexity of their contextual relationships. Current deep learning approaches encounter difficulties in accurately identifying boundary features and lack efficiency in collaboratively modeling the foreground and background by leveraging contextual features. To address these challenges, we propose a stronger multifaceted collaborative salient object detector in ORSIs, termed LBA-MCNet, which incorporates aspects of localization, balance, and affinity. The network focuses on accurately locating targets, balancing detailed features, and modeling image-level global context information. Specifically, we design the Edge Feature Adaptive Balancing and Adjusting(EFABA) module for precise edge localization, using edge features to guide attention to boundaries and preserve spatial details. Moreover, we design the Global Distributed Affinity Learning(GDAL) module to model global context. It captures global context by generating an affinity map from the encoders final layer, ensuring effective modeling of global patterns. Additionally, deep supervision during deconvolution further enhances feature representation. Finally, we compared with 28 state of the art approaches on three publicly available datasets. The results clearly demonstrate the superiority of our method.

Quantization techniques are widely used to improve inference speed and deployment of large language models. While a wide body of work examines the impact of quantized LLMs on English tasks, none have examined the effect of quantization across languages. We conduct a thorough analysis of quantized multilingual LLMs, focusing on their performance across languages and at varying scales. We use automatic benchmarks, LLM-as-a-Judge methods, and human evaluation, finding that (1) harmful effects of quantization are apparent in human evaluation, and automatic metrics severely underestimate the detriment: a 1.7% average drop in Japanese across automatic tasks corresponds to a 16.0% drop reported by human evaluators on realistic prompts; (2) languages are disparately affected by quantization, with non-Latin script languages impacted worst; and (3) challenging tasks such as mathematical reasoning degrade fastest. As the ability to serve low-compute models is critical for wide global adoption of NLP technologies, our results urge consideration of multilingual performance as a key evaluation criterion for efficient models.

We consider the problem of sequentially conducting multiple experiments where each experiment corresponds to a hypothesis testing task. At each time point, the experimenter must make an irrevocable decision of whether to reject the null hypothesis (or equivalently claim a discovery) before the next experimental result arrives. The goal is to maximize the number of discoveries while maintaining a low error rate at all time points measured by local False Discovery Rate (FDR). We formulate the problem as an online knapsack problem with exogenous random budget replenishment. We start with general arrival distributions and show that a simple policy achieves a $O(\sqrt{T})$ regret. We complement the result by showing that such regret rate is in general not improvable. We then shift our focus to discrete arrival distributions. We find that many existing re-solving heuristics in the online resource allocation literature, albeit achieve bounded loss in canonical settings, may incur a $\Omega(\sqrt{T})$ or even a $\Omega(T)$ regret. With the observation that canonical policies tend to be too optimistic and over claim discoveries, we propose a novel policy that incorporates budget safety buffers. It turns out that a little more safety can greatly enhance efficiency -- small additional logarithmic buffers suffice to reduce the regret from $\Omega(\sqrt{T})$ or even $\Omega(T)$ to $O(\ln^2 T)$. From a practical perspective, we extend the policy to the scenario with continuous arrival distributions as well as time-dependent information structures. We conduct both synthetic experiments and empirical applications on a time series data from New York City taxi passengers to validate the performance of our proposed policies. Our results emphasize how effective policies should be designed in online resource allocation problems with exogenous budget replenishment.

Sampling-based planning algorithms like Rapidly-exploring Random Tree (RRT) are versatile in solving path planning problems. RRT* offers asymptotical optimality but requires growing the tree uniformly over the free space, which leaves room for efficiency improvement. To accelerate convergence, informed approaches sample states in an ellipsoidal subset of the search space determined by current path cost during iteration. Learning-based alternatives model the topology of the search space and infer the states close to the optimal path to guide planning. We combine the strengths from both sides and propose Neural Informed RRT* with Point-based Network Guidance. We introduce Point-based Network to infer the guidance states, and integrate the network into Informed RRT* for guidance state refinement. We use Neural Connect to build connectivity of the guidance state set and further boost performance in challenging planning problems. Our method surpasses previous works in path planning benchmarks while preserving probabilistic completeness and asymptotical optimality. We demonstrate the deployment of our method on mobile robot navigation in the real world.

Emergent properties have been widely adopted as a term to describe behavior not present in smaller models but observed in larger models. Recent work suggests that the trade-off incurred by quantization is also an emergent property, with sharp drops in performance in models over 6B parameters. In this work, we ask "are quantization cliffs in performance solely a factor of scale?" Against a backdrop of increased research focus on why certain emergent properties surface at scale, this work provides a useful counter-example. We posit that it is possible to optimize for a quantization friendly training recipe that suppresses large activation magnitude outliers. Here, we find that outlier dimensions are not an inherent product of scale, but rather sensitive to the optimization conditions present during pre-training. This both opens up directions for more efficient quantization, and poses the question of whether other emergent properties are inherent or can be altered and conditioned by optimization and architecture design choices. We successfully quantize models ranging in size from 410M to 52B with minimal degradation in performance.

Ubiquitous cyber-intrusions endanger the security of our devices constantly. They may bring irreversible damages to the system and cause leakage of privacy. Thus, it is of vital importance to promptly detect these intrusions. Traditional methods such as Decision Trees and Support Vector Machine (SVM) are used to classify normal internet connections and cyber-intrusions. However, the intrusions are largely fewer than normal connections, which limits the capability of these methods. Anomaly detection methods such as Isolation Forest can handle the imbalanced data. Nevertheless, when the features of data increase, these methods lack enough ability to learn the distribution. Generative adversarial network (GAN) has been proposed to solve the above issues. With its strong generative ability, it only needs to learn the distribution of normal status, and identify the abnormal status when intrusion occurs. But existing models are not suitable to process discrete values, leading to immense degradation of detection performance. To cope with these challenges, in this paper, we propose a novel GAN-based model with specifically-designed loss function to detect cyber-intrusions. Experiment results show that our model outperforms state-of-the-art models and remarkably reduce the overhead.