The input toeach projection head is the output of the backbone network - a feature vector of size512 and the output dimension is the number of style dimensions - in the case of the pretrained CLEVR styleGAN2 used in our experiments,thisvalueis204. For the generator, we use a modified version of styleGAN2 that is trained to output images of resolution 128 128. For each input image of sizeH W C, we start by generating a random mask of sizeH W where each pixel value in contained in the interval[0,1]. These thresholds were obtained by visual inspection. In this experiment, we set out to measure the variability of the predicted quantile intervals as a functionofproblemdifficulty.

Test-time scaling (TTS) -- the dynamic allocation of compute during inference -- is a promising direction for improving reasoning in large language models (LLMs). However, a systematic comparison of well-known TTS strategies under identical conditions is missing, and the influence of model type and problem difficulty on performance remains unclear. To address these gaps, we conduct the first large-scale study of TTS, spanning over thirty billion tokens generated using eight open-source LLMs (7B to 235B parameters), across four reasoning datasets. We observe three consistent trends: (1) no single TTS strategy universally dominates; (2) reasoning models exhibit distinct trace-quality patterns across problem difficulty and trace length, forming short-horizon and long-horizon categories; and (3) for a given model type, the optimal TTS performance scales monotonically with compute budget. Based on these insights, we provide a practical recipe for selecting the best TTS strategy, considering problem difficulty, model type, and compute budget, providing a practical guide to effective inference-time scaling.

Reasoning language models perform well on complex tasks but are costly to deploy due to their size and long reasoning traces. We propose a routing approach that assigns each problem to the smallest model likely to solve it, reducing compute without sacrificing accuracy. Using intermediate representations from s1.1-32B, we train lightweight predictors of problem difficulty or model correctness to guide routing across a pool of reasoning models. On diverse math benchmarks, routing improves efficiency over random assignment and matches s1.1-32B's performance while using significantly less compute. Our results demonstrate that difficulty-aware routing is effective for cost-efficient deployment of reasoning models.

The rapid advancement of Large Language Models (LLMs) poses a significant challenge to existing mathematical reasoning benchmarks. However, these benchmarks tend to become easier over time as LLMs can learn from the published benchmarks. This limitation hinder the precise evaluation of the true capabilities of SOTA models. To address this challenge, this paper introduces EvolMathEval, an automated mathematical benchmark generation and evolution framework based on evolutionary testing. Experimental results demonstrate that EvolMathEval can not only generate a large volume of high-difficulty problems through continuous self-iteration, but it can also significantly enhance the complexity of public datasets like GSM8K through evolution, reducing model accuracy by an average of 48\%. Deeper investigation reveals that when solving these evolved problems, LLMs tend to bypass complex multi-step logical reasoning by relying on simplistic and fuzzy conditions, consequently leading to incorrect solutions. We define this phenomenon as the ``Pseudo Aha Moment", which we find accounts for 77\% to 100\% of errors on targeted problems. Code and resources are available at: https://anonymous.4open.science/r/EvolMathEval

Knowledge Tracing (KT) plays a central role in assessing students' skill mastery and predicting their future performance. While deep learning-based KT models achieve superior predictive accuracy compared to traditional methods, their complexity and opacity hinder their ability to provide psychologically meaningful explanations. This disconnect between model parameters and cognitive theory poses challenges for understanding and enhancing the learning process, limiting their trustworthiness in educational applications. To address these challenges, we enhance interpretable KT models by exploring human-understandable features derived from students' interaction data. By incorporating additional features, particularly those reflecting students' learning abilities, our enhanced approach improves predictive accuracy while maintaining alignment with cognitive theory. Our contributions aim to balance predictive power with interpretability, advancing the utility of adaptive learning systems.

Group Relative Policy Optimization (GRPO), which is widely adopted by R1-like reasoning models, has advanced mathematical reasoning. Nevertheless, GRPO faces challenges in reward sparsity, verbosity, and inadequate focus on problem difficulty. We propose GRPO-LEAD, enhancing GRPO with: (1) length-regularized rewards to encourage conciseness while maintaining accuracy; (2) explicit penalties for incorrect solutions to improve model precision; and (3) difficulty-aware advantage reweighting for robust generalization on challenging problems. Comprehensive evaluations demonstrate that GRPO-LEAD significantly improves reasoning accuracy, conciseness, and efficiency. Our approach achieves state-of-the-art performance for 14B-scale models, underscoring the synergy of our methods with appropriate model scale and high-quality data. Our source code, generated dataset, and models are available at https://github.com/aeroplanepaper/GRPO-LEAD.

Recent advances have shown that scaling test-time computation enables large language models (LLMs) to solve increasingly complex problems across diverse domains. One effective paradigm for test-time scaling (TTS) involves LLM generators producing multiple solution candidates, with LLM verifiers assessing the correctness of these candidates without reference answers. In this paper, we study generative verifiers, which perform verification by generating chain-of-thought (CoT) reasoning followed by a binary verdict. We systematically analyze verification dynamics across three dimensions - problem difficulty, generator capability, and verifier generation capability - with empirical studies on 12 benchmarks across mathematical reasoning, knowledge, and natural language reasoning tasks using 14 open-source models (2B to 72B parameter range) and GPT-4o. Our experiments reveal three key findings about verification effectiveness: (1) Easy problems allow verifiers to more reliably certify correct responses; (2) Weak generators produce errors that are easier to detect than strong generators; (3) Verification ability is generally correlated with the verifier's own problem-solving capability, but this relationship varies with problem difficulty. These findings reveal opportunities to optimize basic verification strategies in TTS applications. First, given the same verifier, some weak generators can nearly match stronger ones in post-verification TTS performance (e.g., the Gemma2-9B to Gemma2-27B performance gap shrinks by 75.5%). Second, we identify cases where strong verifiers offer limited advantage over weak ones, as both fail to provide meaningful verification gains, suggesting that verifier scaling alone cannot overcome fundamental verification challenges.

Intermediate token generation (ITG), where a model produces output before the solution, has been proposed as a method to improve the performance of language models on reasoning tasks. While these reasoning traces or Chain of Thoughts (CoTs) are correlated with performance gains, the mechanisms underlying them remain unclear. A prevailing assumption in the community has been to anthropomorphize these tokens as "thinking", treating longer traces as evidence of higher problem-adaptive computation. In this work, we critically examine whether intermediate token sequence length reflects or correlates with problem difficulty. To do so, we train transformer models from scratch on derivational traces of the A* search algorithm, where the number of operations required to solve a maze problem provides a precise and verifiable measure of problem complexity. We first evaluate the models on trivial free-space problems, finding that even for the simplest tasks, they often produce excessively long reasoning traces and sometimes fail to generate a solution. We then systematically evaluate the model on out-of-distribution problems and find that the intermediate token length and ground truth A* trace length only loosely correlate. We notice that the few cases where correlation appears are those where the problems are closer to the training distribution, suggesting that the effect arises from approximate recall rather than genuine problem-adaptive computation. This suggests that the inherent computational complexity of the problem instance is not a significant factor, but rather its distributional distance from the training data. These results challenge the assumption that intermediate trace generation is adaptive to problem difficulty and caution against interpreting longer sequences in systems like R1 as automatically indicative of "thinking effort".

Reinforcement learning with verifiable rewards (RL VR) has achieved remarkable success in enhancing the reasoning capabilities of large language models (LLMs). However, existing RL VR methods often suffer from exploration inefficiency due to mismatches between the training data's difficulty and the model's capability. LLMs fail to discover viable reasoning paths when problems are overly difficult, while learning little new capability when problems are too simple. Building on this analysis, we propose SEELE, a novel supervision-aided RL VR framework that dynamically adjusts problem difficulty to stay within the high-efficiency region. Unlike previous hint-based approaches, SEELE deliberately and adaptively adjusts the hint length for each problem to achieve an optimal difficulty. To determine the optimal hint length, SEELE employs a multi-round rollout sampling strategy. In each round, it fits an item response theory model to the accuracy-hint pairs collected in preceding rounds to predict the required hint length for the next round. Experimental results show that SEELE outperforms Group Relative Policy Optimization (GRPO) and Supervised Fine-tuning (SFT) by +11.8 and +10.5 points, respectively, and surpasses the best previous supervision-aided approach by +3.6 points on average across six math reasoning benchmarks.